AI Overviews are the AI-written answer Google puts at the top of some results pages, and for a business with an address the honest headline is that there is nothing new to install, submit or mark up. Google's guidance for site owners says it in one sentence: "The best practices for SEO remain relevant for AI features in Google Search (such as AI Overviews and AI Mode). There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." The profile, the reviews and the pages you already have are the whole lever.

For a local business, the practical question is when an AI Overview appears and what it reads when it does. Every claim below that belongs to Google is quoted from Google's own pages, checked on 6 September 2026. Where Google has not documented something, this post says so rather than guessing. This article sticks to Google's AI Overview; AI search optimization for local businesses covers all five surfaces together.

What is an AI Overview?

An AI Overview is a summary Google writes at the top of a results page, with links to the pages it drew on. Google Search Help calls it "an AI-generated snapshot with key information and links to dig deeper". Search Central adds the purpose: they "help people get to the gist of a complicated topic or question more quickly, and provide a jumping off point to explore links to learn more" (AI features and your website).

It is not an ad unit, so there is no placement to buy. And your customers cannot switch it off. Google's help page says AI Overviews are "a core Google Search feature, like knowledge panels. Features cannot be turned off", though a searcher can apply the Web filter after searching to see "only text-based links without features like AI Overviews" (AI Overviews in Google Search). Site owners can still control what Google takes from their own pages.

When does Google show an AI Overview for a local search?

Less often than the coverage suggests. Google's own wording is that AI Overviews "were designed to show up on queries where they can add additional benefits beyond what people might already get on Search", and, further down the same page, that they "are only shown when our systems determine that it is additive to classic Search, and as such, often don't trigger."

Here is what that looked like on 6 September 2026, from one Australian connection that Google placed at Carole Park in Queensland while I searched for a Sydney suburb.

• "best physio in bondi" returned a Places block: three profiles, a map, a "Top rated" filter chip, and a quoted line from a customer review under each listing. No AI Overview.
• "what should I look for when choosing a physio in bondi" returned an AI Overview with headed sections on credentials, treatment approach and local fit, a panel of three cited articles, and an "Ask anything" box underneath it. No Places block.

Two searches on one day from one location is an observation, not a rule. I ran the second one twice within a few minutes and the opening sentence came back differently each time: same shape, different wording, a different thing emphasised. One AI Overview is one sample, which is worth remembering before anyone reads a trend into a screenshot.

The pattern does line up with what Google documents. The shopping question already had a good answer in the form of three profiles, so a summary added nothing. The research question had no such answer, so it got one. The searches closest to a booking are still being answered by your Business Profile, and AI Overviews are turning up on the questions people ask before that. For the profile half, whether Google reviews affect your ranking covers the three factors Google publishes for local results.

What do AI Overviews use to answer a local question?

They run on ordinary Search. Google's guide to generative AI features describes the mechanism as retrieval-augmented generation, "a technique (also known as grounding)" that relies "on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index" (Optimizing your website for generative AI features). Alongside it sits query fan-out, "a set of concurrent, related queries generated by the model to request more information".

So the entry requirement is the ordinary one. Search Central: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements."

Your Business Profile is named, twice. The generative AI guide has a section headed "Optimize your local business and ecommerce details" saying that "generative AI responses can include product listings, product information, and information about local businesses", and that Merchant Center and Google Business Profiles "can help your products and services to be visible in both AI responses and other Google Search results". The AI features guide repeats it as a checklist item: "Checking that your Merchant Center and Business Profile information is up-to-date."

What sits behind that profile is not only what you typed into it. Google's Business Profile help says the information is compiled from publicly available sources "such as crawled web content", licensed third-party data, users who contribute facts and content, and Google's own interactions with the place (how Google sources business information).

Reviews: what is documented, and what is not. Google documents reviews as an input to local ranking. Its local ranking page puts them inside prominence and says "More reviews and positive ratings can help your business's local ranking." Google also documents that generative AI features are "rooted in our core Search ranking and quality systems".

What Google has never published is a sentence saying AI Overviews read, weigh or quote your individual reviews. Not the count, not the rating, not the text. Anyone telling you how many reviews an AI Overview needs invented the number. The one Google product that does document reading review text is the Gemini API's Grounding with Google Maps tool, a developer product rather than the AI Overview on a search page. That, and the equivalent documentation from OpenAI and Perplexity, is covered in does ChatGPT recommend businesses like yours.

Is there special markup or a file for AI Overviews?

No, and Google spends a section saying so. From the mythbusting part of the generative AI guide:

You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them.

The same section rules out special schema.org markup and says there is no requirement to break your content into tiny pieces for AI to understand it. It also warns against the tactic usually sold next to this one, which is spinning up a page for every question variation. Creating separate content for every possible search "primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy". If an agency is quoting you for an llms.txt file, an AI-flavoured rewrite of your service pages or forty new suburb pages, that quote is for work Google says it does not use.

How to appear in AI Overviews as a local business

Start at the top of this list and work down. None of it is new.

1. Check you are indexed, snippet-eligible and not excluded. Indexing and snippet eligibility are the only technical requirements, and they are also how a business gets shut out by accident: nosnippet, data-nosnippet, max-snippet and noindex all limit what Google can show from your pages. There is a second switch. Google's generative AI guide says a site "must be included in Search generative AI features in Search Console to be eligible for display in generative AI features on Google Search". Inclusion is the default, so most businesses have nothing to do here. Check it anyway if an agency has been in there.
2. Finish the Business Profile. Google's local ranking page asks for the full address, the hours including special hours, the right category and attributes like parking and Wi-Fi, and says "Businesses with complete and accurate info are more likely to show up in local search results." Adding your business to Google Maps has the setup and verification steps.
3. Keep collecting reviews, and reply to them. Google's documented claim here is about local ranking rather than AI answers, and it is worth having on those terms: more reviews and positive ratings can help, and "Positive reviews and helpful replies can help your business stand out." Ask every customer, not only the ones you expect to be pleased. Google's contribution policies bar merchants from moves that "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers". How to get more Google reviews covers the timing and the follow-up.
4. Make your details identical everywhere you are listed. Profiles are built partly from crawled web content and licensed third-party data, so an old phone number on a directory is a number Google has seen. Fixing the outliers is dull work and the highest-yield hour in this list.
5. Write the page that answers the question. In the AI Overview I looked at, the cited sources were articles from clinics and health providers about how to choose a physiotherapist. Not a directory, not an SEO agency's listicle. Google's guidance points the same way, asking for a "unique point of view" and contrasting commodity content, "something like '7 Tips for First-Time Homebuyers'", with a piece that carries real experience. You know the questions your customers ask in the first appointment. Those are the pages.

What AI Overviews do to your clicks

Only Google has the aggregate data. In August 2025, Liz Reid, VP and Head of Google Search, published the company's position: "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year" (AI in Search is driving more queries and higher quality clicks, 6 August 2025).

The same post concedes the obvious. For quick-answer questions, "like 'when is the next full moon,'" people "may be satisfied with the initial response and not click further", and "user trends are shifting traffic to different sites, resulting in decreased traffic to some sites and increased traffic to others." Read it for what it is: Google's account of its own product, in aggregate, with no method published. The only number that describes your site is in your own Search Console.

AI Overviews and AI Mode are not the same thing

Google says the two features behave differently: "AI Mode and AI Overviews may use different models and techniques, so the set of responses and links they show will vary."

The difference for a local business is who arrives. Someone who opens AI Mode and asks three questions in a row is further into a decision than someone who glanced at a summary. SEO for Google AI Mode: what a local business has to change covers that surface on its own.

How to check whether you appear in AI Overviews

Google now reports it. The generative AI performance report in Search Console "shows data about how your site performs in generative AI features on Google Search", and as of 31 August 2026 Google has "rolled out these insights to all websites worldwide". Four things to know before you read it.

• It counts impressions, "how many times links to your site were shown to a user in a generative AI feature on Google Search". There is no position or ranking figure in it.
• It covers AI Overviews and AI Mode together and does not split them.
• You can group it by page, country, date or device, which is how you find out which of your pages is doing the work.
• It reports on your website, drawing on the Web search type in the ordinary performance report. Your Business Profile is a separate surface with its own reporting, so a strong profile and a quiet report are not a contradiction.

If the report is empty, Google names two causes worth checking: your site has not had enough impressions in generative AI features yet, or it has been excluded from them.

Search Console covers Google and nothing else. ChatGPT, Gemini and Perplexity answer local questions from different sources and cite different pages. For those, run the same searches your customers run, from the area they are in, and read the citation list as a work list. Cloutly's AI search visibility asks those three assistants discovery questions such as "best {category} in {locality}" once a month and records whether each location was named, and it does not sample AI Overviews, so Search Console stays the record for Google's own surfaces. How to monitor AI search visibility is the procedure for checking the assistants by hand.

Frequently asked questions

Can I stop my business appearing in AI Overviews?

For your website, yes. Search Console has a Search generative AI control under Settings, rolled out to all sites on 31 August 2026. Including your site is the default; excluding it covers AI Overviews, AI Mode and generative AI features in Discover, takes one to two days, and costs you the impressions and traffic those features send. Google says the control "isn't used as a ranking or inclusion signal affecting other parts of Search", so it is narrower than nosnippet and noindex, which are Search-wide and take your ordinary snippet with them. None of it governs your Business Profile. That is a separate surface and this switch does not touch it.

Do I need schema markup to appear in an AI Overview?

No. Google's wording is "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." It is still worth having for rich results in ordinary Search, which is the reason to keep it.

Does an AI Overview replace the map pack?

Google has published nothing saying one displaces the other. In the two searches above, on 6 September 2026, they did not appear together: the profile-shopping query returned the Places block and no AI Overview, and the question-shaped query returned an AI Overview and no Places block. One day, one location. Run your own searches, from the suburb you serve, and write down what you see.

AI search optimization is the work of making sure the systems that answer a question with generated text can find, read and repeat accurate information about your business. For a business with an address, that is five surfaces: ChatGPT, Gemini, Perplexity, and Google's own AI Overviews and AI Mode. None of the five publishes a ranking formula for local recommendations, and anyone quoting you one has invented it. All five do publish where their answers come from, and those sources overlap almost exactly with the things a local business was already meant to keep tidy.

In September 2026, page one for the term itself is Google's documentation, a Reddit thread and a run of agencies selling the acronym. Add "local business" to the query and the top result is a webinar listing. Almost none of it cites a primary source. This page quotes the vendors, links to what it quotes, says plainly where the documentation stops, and is the parent page for everything else we write about AI search for local business.

What AI search optimization means, and whether GEO and AEO are different

Google answers this one itself. Its guide to optimizing for generative AI features takes the two acronyms head on: "'AEO' stands for 'answer engine optimization' and 'GEO' for 'generative engine optimization'... From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

For Google's surfaces there is no second discipline to learn. The same page, last updated in July 2026, says the generative features "are rooted in our core Search ranking and quality systems". The AI features guide opens with the line worth pinning above a desk: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

ChatGPT and Perplexity do not run on Google's index, so there is a little more to it for them. It is still a short list, and none of it is exotic.

Grounding. Google describes retrieval-augmented generation as "a technique (also known as grounding) used to improve the quality, accuracy, and freshness of AI responses by relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index". The model does not answer from memory. Something fetches pages first, and the answer is built out of what came back. Every assistant on this page works that way for a local question.

Query fan-out. Google defines it as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results". Its example: ask "how to fix a lawn that's full of weeds" and the fan-out queries might include "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn".

Fan-out is why the old habit of tuning one page to one phrase has stopped paying. "Best physio in Bondi" is not one search behind the scenes. It is several, about opening hours, about specific injuries, about parking, about who takes which health fund. A profile and a body of reviews that happen to answer six adjacent questions get retrieved more often than a page that answers one of them beautifully.

How AI search works for local businesses: the five surfaces

ChatGPT searches rather than remembers. OpenAI's help centre says it "can search the web to answer questions with current information and links to relevant sources" and "can also use location information to find local results" (Searching the web with ChatGPT). The rewriting is the part that surprises owners: ask "what are some good restaurants near me?" and, if ChatGPT reads San Francisco off your IP address, it "may rewrite your prompt into the search query 'top restaurants San Francisco'". What actually goes out is an ordinary local search, and the pages that come back are the raw material. Restaurants have one extra surface to think about, because the same help page says ChatGPT can show reservation times from OpenTable, Resy or Yelp. We pulled the ChatGPT half of this apart properly in does ChatGPT recommend businesses like yours.

Two lines on that page are worth reading before anyone sells you a guaranteed mention.

The second catches people out. Allowing the crawler in robots.txt is not the whole job: your host or CDN has to let OpenAI's published searchbot addresses through as well.

Gemini is the only one of the five whose documentation names reviews as an ingredient. Google's Grounding with Google Maps page describes a data retrieval step where the service "queries Google Maps for relevant information (e.g., places, reviews, photos, addresses, opening hours)", and lists place questions as the first use case.

That documentation covers the Gemini API, which is what developers build on rather than what you get in the Gemini app. Read it as the clearest published statement that review text is read, not as a description of every Gemini answer.

Perplexity is the simplest and the least about your website. Its help centre says it will "search the internet in real-time" and that "each answer includes numbered citations linking to the original sources" (How does Perplexity work?); its crawler documentation names PerplexityBot as the one to let in. For a local question the answer is assembled from whatever public pages discuss your category in your area, which is usually directories, roundups and review sites before it is you.

AI Overviews and AI Mode run on the index you already rank in. Google's eligibility rule is one sentence: a page "must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements." Two other lines matter to anyone measuring this. AI Overviews "are only shown when our systems determine that it is additive to classic Search, and as such, often don't trigger", so a query that shows one today may not tomorrow. And both features use fan-out to "display a wider and more diverse set of helpful links associated with the response than with a classic web search".

One switch can shut you out of both. Since 31 August 2026 every Search Console property has a Search generative AI control under Settings. Include is the default; exclude means "you won't receive any traffic or impressions from these features". Worth two minutes to confirm nobody has touched it.

The inputs all five have in common

Six things feed a local answer on every one of those surfaces. Only some are documented, and the table says which.

Review text and citations elsewhere are the two inputs local businesses often overlook.

Review text is the input almost nobody works on. Count and rating get the attention because they are visible on the profile. But nobody writes "great service" in an answer about rotator cuff rehab or emergency gas fitting. If forty of your reviews name the treatment, the technician, the breed or the machine, an assistant has forty sentences it can match against a specific question. If all forty say "highly recommend", it has none, and you are competing on star rating alone. You get that language by asking promptly, not by asking for it.

Citations elsewhere are the input you cannot buy. Perplexity will happily answer a local question without touching your website, using the directory pages and local roundups that rank for your category. Reading the citation list under an answer is the fastest work list in this whole discipline: if the same roundup is cited every month and you are not in it, that is a job with a name.

Both of those come out of ordinary review work. How to get more Google reviews covers the ask and the timing, and adding your business to Google Maps covers finishing the profile.

What does not work

Paying. Google's local ranking page states it flatly: "There's no way to request or pay for a better local ranking on Google." OpenAI says the same about ChatGPT's ads, which began as a US test in February 2026 and has since expanded to Canada, Australia, New Zealand and several other markets: "Ads do not influence the answers ChatGPT gives you", and an ad is "always clearly labeled as sponsored and visually separated from the organic answer" (Testing ads in ChatGPT, updated 11 August 2026). Buying the ad buys the ad.

Filtering who you ask. Sending happy customers to Google and unhappy ones to a private form is against Google's prohibited and restricted content policy, which stops merchants who "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers". It also produces a review corpus that does not describe your business, which is exactly the wrong outcome when the text is being read back to a customer. Cloutly has not had a rating step in front of the review sites since 2021. Review gating is the longer version.

Asking customers to include keywords. The same policy tells merchants they should not "request that specific content be included". The template that says "please mention emergency plumbing in Newcastle" is a policy breach on the profile you can least afford to lose, in exchange for an effect no vendor has ever documented. Ask promptly instead. A customer asked the same day writes about the job they just had, and the language arrives on its own.

Writing reviews with AI. Generated review text is fabricated review text no matter who typed the prompt. Every platform that hosts reviews bans it, and it corrupts the one asset on this list that an assistant reads as evidence.

The file-and-markup tactics. Google's mythbusting section names them one by one. On llms.txt: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." On rewriting: "You don't need to write in a specific way just for generative AI search." On buying mentions: "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem." And on schema: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add", though it stays worth having for rich results.

That list is specific to Google, and Google says so: keeping an llms.txt file for other services that use one "will neither harm nor help your site's visibility or rankings in Google Search". So the file is site hygiene for the non-Google assistants if you want it. It is not the lever, and an agency selling it to a plumber as one is selling the wrong thing.

Google adds a warning about the tools built on top of all this, and it applies to every vendor in the category including us: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems."

AI answers send fewer people to your site. Being named is the point.

An assistant that answers the question well has given the customer what they came for. Some of those people will never click anything, and no vendor publishes a click-loss figure for local queries, so treat any number you are quoted as somebody's estimate rather than a measurement.

What Google does publish is the shape of what is left: "when people click from search results pages with AI Overviews, these clicks are higher quality (meaning, users are more likely to spend more time on the site)." Fewer visitors, further along.

For a business with a phone number this is less of a change than it is for a publisher. Your outcome was never a session. It was a call, a booking, a direction request or someone walking in. Being named in the answer, with the right phone number attached, does that job whether or not anyone visits your website on the way. So measure the outcome rather than the traffic, and check the naming separately.

For Google's surfaces you now have a first-party report. Search Console's Generative AI performance report was rolled out to all websites worldwide on 31 August 2026 and covers AI Overviews and AI Mode. Read it for what it is: impressions, grouped by page, country, date or device. Nothing outside Google Search appears in it, so for ChatGPT, the Gemini app and Perplexity the only honest method is still to ask them and write down what comes back.

How to get found in AI search: the monthly check

Spend ten minutes on the same day each month and write the results down. One answer is only one sample. Look for the pattern across months and the source list that keeps appearing.

1. Fix eight prompts and stop changing them. Mix the shapes: "best [category] in [suburb]", "[category] near [landmark] open [day]", "who should I see for [a problem you solve] in [suburb]", "is [your business] any good", "[your business] or [a rival you actually lose to]". The full set is in does ChatGPT recommend businesses like yours.
2. Name the location in the prompt. OpenAI's own advice when local results show the wrong area is to "include your city, neighborhood, or postal code in your question". Do not let the assistant guess it from your office connection.
3. Run each prompt in a fresh chat, signed out where you can be, in all three assistants. Memory and chat history change the rewritten query. Testing inside your logged-in account measures your account.
4. Record seven things. Whether you were named, where in the list, which rivals were named, which sources were cited, whether the phone number and hours came back right, which assistant, and the date.

Treat every result as one sample. These systems are not deterministic and a re-ask can differ, so one miss is not a verdict and one hit is not a ranking. If you would rather not run it by hand, Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were.

Where to start

If you run one site, in this order:

1. Finish the Google Business Profile. Primary category first, then hours, then every remaining field.
2. Check the crawlers can reach you: robots.txt, your CDN's bot rules, and the Search generative AI control in Search Console. Blanket AI blocks were fashionable in 2023 and some are still sitting there.
3. Start asking every customer for a review, promptly, forever. Not a drive.
4. Fix the two or three directory listings that carry an old phone number or an old address.
5. Run the monthly check for three months before you conclude anything.

If you run a network, the order changes, because your problem is variance between sites rather than absence:

1. Make one location correct end to end and use it as the reference. Category, hours, description, photos.
2. Run the check per location, not per brand. The flagship being named tells you nothing about the suburban sites, and it is usually the suburban sites that are invisible.
3. Sort by worst first. A weak location drags only itself, so the return is highest where the profile is emptiest.
4. Get every site asking for reviews the same way, so the corpus is comparable and no location is running on a manager's memory.
5. Fix the details centrally. Twenty listings drifting in twenty directions is a listings problem before it is an AI problem.

Where to go deeper

The pieces below each take one part of this page further.

How it works

• Does ChatGPT recommend businesses like yours? How to check
• AI Overviews for local businesses: how to appear
• SEO for Google AI Mode: what a local business has to change
• Perplexity and Gemini answer local questions differently from ChatGPT

Measuring it

• How to monitor AI search visibility: eight prompts, five assistants, one log
• When AI gets your business details wrong (and how to fix it)
• AI search visibility for multi-location brands: why the suburbs get missed
• AI search visibility tools: what they measure and how to choose

The reviews link

• Do Google reviews affect ranking? What Google confirms, and what it doesn't
• AI and Google reviews: do they change what AI says?
• The AI summary above your reviews: where it comes from and what to do about it

The vocabulary

• GEO vs SEO, AEO and LLMO: what the acronyms mean for a local business
• AI visibility explained: what it measures, what it can't
• Do you need an llms.txt file?

For the numbers behind any of it, AI search statistics carries a publisher, an edition and a sample on every figure.

Frequently asked questions

Does AI search replace my Google Business Profile?

No. It reads it. Google names Business Profile information in its own generative AI guidance, Gemini's Maps grounding queries Maps directly, and a ChatGPT search on a local question returns the local results your profile sits in. The profile has become more load-bearing, not less.

Which assistant should I care about most?

Nobody publishes a share of local questions by assistant, so anyone ranking them for you is guessing. Ask all three directly, watch the AI Overview on your own category search, and notice that they fail differently. ChatGPT and Perplexity miss you when you are absent from the pages they retrieve. Google's surfaces miss you when your profile is thin.

How long before any of this changes an answer?

Nobody outside the vendors can say, and it is not one lever with a lag. A review appears on your profile in a day or two. The next answer might reflect it immediately or not at all, because the answer is rebuilt from a fresh retrieval each time. Check monthly and read the trend, not the run.

Do I need a separate version of my website for AI?

No. Google warns that spinning up content for every variation of how somebody might search, done to manipulate rankings or AI responses, breaches its scaled content abuse spam policy. Write for people and organise with ordinary headings and paragraphs. The one genuine adjustment for the non-Google assistants is making sure the facts a customer needs, which for a local business means hours, address, phone, services and prices, sit in text on a page rather than inside an image or a script.

AI search visibility for a multi-location brand is two measurements, not one. "Best physio in Bondi" is a location question, answered from what the web and Google Maps hold about one address. "Is Coastline Physio any good" is a brand question, answered from what they hold about the whole network. A brand can pass the second at every assistant and fail the first at forty of its sites.

For a network of 20 to 400 sites, use how AI search decides which local businesses get named as the starting point. At one location the check is ten minutes a month and none of the sampling below applies.

The category itself does not make the distinction. Search "ai search visibility" today and page one is trackers: Ubersuggest, Rankscale, Semrush's free checker, a couple of scoring tools. They define the measurement the same way, as the share of AI answers that name your brand, and they all track a brand. Not one of them has a row for an address. That is a reasonable unit for a software company with one website. It is the wrong unit for a business with forty front doors.

The brand answer and the location answer are different questions

Marketing leads find this out the same way every time. Somebody asks ChatGPT about the brand, gets a warm paragraph back, and concludes the network is fine. Then a franchisee three train stops away asks for the category in their own postcode and the brand is nowhere. Both results are correct. The assistant answered two different questions out of two different piles of material.

The brand question retrieves what the web says about the name: your website, your press, roundups, aggregator pages, and the reviews of whichever sites happen to be documented well enough to surface. One answer for the whole network, mostly built out of things head office controls.

The location question retrieves what the web says about one suburb. Google defines query fan-out as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query", and for a local question those concurrent queries are about hours, parking, a health fund, whether you see children. Almost none of that lives on the brand website. It lives on one location's Business Profile, in one location's reviews, and on the listings for one address.

Gemini makes the split literal. Google's Grounding with Google Maps documentation describes a retrieval step where the service "queries Google Maps for relevant information (e.g., places, reviews, photos, addresses, opening hours)", and the tool call in Google's own sample takes a latitude and a longitude.

Maps is organised by address. There is no network row in it.

Two ways a network fails, and only one of them is obvious

A weak site drags only itself, until the brand answer picks it up

In the location answer, a weak site drags only itself. Nothing about a thin profile in Parramatta stops the Bondi site being named in Bondi.

The brand answer does not work that way. It is assembled from material about the name, and nothing makes the model prefer your best-documented site over your worst. A location carrying a disconnected phone number, an old address, a "permanently closed" flag or a 3.1 rating is part of what an assistant reads when somebody asks whether the brand is any good. One site can put a sentence into the brand answer that is true of that site and false of the other forty.

Nobody measures this, because head office only ever asks the brand question and it usually comes back clean. It comes back clean until the day retrieval happens to pull the wrong site. When AI gets your business details wrong covers tracing a wrong detail back to whatever is carrying it.

The brand is named, but the wrong branch is

The second failure is quieter and costs bookings the same week. A customer in Newtown asks for the category near them, the assistant names your brand, and it names the Bondi site because that is the address the web documents best. The customer rings a clinic forty minutes away, or more often does nothing.

No brand-level check shows you this, because the brand was named. It surfaces only when the check runs in the suburb the customer is standing in, which is the whole argument for prompting per location. The cause is nearly always the same: the suburban profile is thin, so retrieval falls back to the site with enough material to retrieve.

What gets read per location, and what gets read once for the brand

Sort the inputs by whether the thing exists once or once per site. That decides who fixes it, and how many times.

Most of what decides a location answer is not centrally editable. Head office can mandate it, fund it and chase it, but the material sits on 200 profiles and in 200 review corpora. That makes it a governance problem, which is where the multi-location review management playbook arrives from the reviews side.

The cheapest wins are structural rather than editorial. A duplicate profile splits one location's reviews across two cards and halves the rating a customer sees, which is a retrieval problem as much as a reputation one. Multiple locations with the same name covers clearing a duplicate, and getting a location onto Google Maps properly covers the profile. Neither is glamorous and both outrank anything you could write about AI.

How to check AI search visibility across 200 locations without a full-time job

Two hundred sites, eight prompts and three assistants is 4,800 runs a month. Nobody does that by hand, and a network that tries will do it once and never again. Do not measure everything every month. Run two tracks instead.

Track one: a fixed panel, every month. Pick ten sites and do not change them. Take them from the top of whatever you already rank on, revenue or review volume, and add one or two you know are weak so the panel is not flattering. Run the eight prompts from how to monitor AI search visibility against each, same day each month, signed out, one fresh chat per prompt, suburb written into the sentence.

Ten sites, eight prompts and three assistants is 240 runs, which is about a day for one person once a month. If that is too much, cut the assistants before you cut the sites. One assistant across ten sites tells you more than three assistants across three.

Track two: every site once a quarter, on rotation. Two hundred sites over three months is roughly 67 a month, or about 17 a week. Give each site two prompts and one assistant: the plain discovery prompt for its category and suburb, and one constrained prompt built from a filter your customers apply, such as opening on a Sunday, taking a particular health fund, or seeing children. That is 34 runs a week, which is an hour.

Do not spend rotation runs on the named prompts. "Is {brand} any good" is a brand question and the panel answers it already.

Fix the sentences once and change only the values inside them. The moment somebody rewrites a prompt, the quarter stops comparing to the last one.

Use one sheet with a location column. Date, location, assistant, prompt, named or not, rivals named, sources cited, facts correct. The location column turns 200 separate checks into a network measurement, and it is the column people forget when they set this up in a hurry.

If running it by hand does not fit, Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were.

How to read the results

Three things come out of the sheet, and only three are worth an executive's attention.

Named or not named, per site. Treat it as a binary and ignore where in the list you appeared. The ordinal is one sample of a system that is not deterministic, so a site that was second in September and fourth in October has told you nothing. Named in September and absent in October has told you something worth checking.

Which rivals were named. The most useful column, and the one people skip. The same rival in every suburb is a brand-level competitor whose brand-side material is better than yours, and that is a head-office job. A different rival in each suburb is normal, and it means the fight is local and the work goes to the sites.

Whether the facts came back right. A wrong phone number in an answer is worse than not being named, because the customer tried. Log it separately from the naming result and fix it at whatever source carries it, rather than by complaining to the assistant.

Report two network numbers, both carrying their denominator: how many of the sites you sampled were named by at least one assistant, and how many had a fact wrong. "31 of 67 sites named this quarter, 4 with a wrong phone number" survives being questioned. Report movement as flips rather than as a percentage. "Three sites started being named, one stopped" is defensible on one sample per site. "Share of voice up 2.4 points" is not, and somebody in the room will eventually ask what the denominator was.

The count tells you the size of the problem and nothing about where it is. At 200 locations the same proportion can mean one weak region or a thin spread everywhere, so put the count on the slide and the list of sites that missed in the appendix. Only the list can be worked.

What to do with a location that is never named

A site that misses once is noise. A site that misses in three consecutive quarters is a work list, and it is usually the same list in the same order.

You can only work 200 profiles if you can reach 200 profiles. Google's business groups are the mechanism, and the trade-off is written down: one group gives you every profile in one dashboard and one spreadsheet, several groups keep each region's people out of profiles they should not be editing and cost you the single view.

1. Check it has a profile, and only one. A duplicate splits the reviews and the rating across two cards, so neither card is strong enough to be retrieved. Clear it before anything else.
2. Finish the profile. Primary category first, then hours, the description, photos, every remaining field. One hour of work that changes more than the rest of the list combined.
3. Read the constrained prompt result. If "physio in {suburb} that bulk bills" names three rivals and not you, and you do bulk bill, the problem is not the assistant. Nothing public says you do.
4. Read the citation list. If the same suburb roundup or directory is cited every time and the site is not in it, that is a job with a name and an owner.
5. Get the site asking for reviews the same way as the rest of the network. A site running on a manager's memory produces a corpus nobody can match a question against. The trigger and the timing are set centrally, the wording stays local, and how to get more Google reviews covers the ask itself.
6. Give it two quarters. Nothing on this list moves an answer in a fortnight, and a site panicked over weekly gets worse rather than better.

One thing not to do, and it comes up at every network this size: do not filter who gets asked. Routing happy customers to the review sites and unhappy ones to a private form breaches Google's contributed-content policy, which says merchants must not "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers". It also produces a corpus that does not describe the location, which is the last thing you want when a machine is reading that text back to a customer. Cloutly has not had a rating step in front of the review sites since 2021.

Reporting AI search visibility to the board

Two sources, measuring different things. Say which is which on the slide.

Google's own surfaces have a first-party report. Search Console's generative AI performance report covers AI Overviews and AI Mode, and as of 31 August 2026 Google has rolled it out to all websites worldwide. Read it for what it is. It reports impressions, defined as "how many times links to your site were shown to a user in a generative AI feature on Google Search", and nothing else. No clicks, no position. You can group by pages, countries, dates or devices.

Two things about that report bite at network scale. The usual Search Console limits apply, the 1,000-row limit included, so a 400-site network with a page per location will not see every location in one view of the page table. And nothing outside Google Search appears in it at all, so it says nothing about ChatGPT, the Gemini app or Perplexity.

Everything else is your sweep, and a sweep is a sample. Report it as the counts above, each with its denominator.

Put the caveat on the slide rather than in the speaker notes. These results are a point-in-time sample of a non-deterministic system. One sample each, and a re-ask can differ. Say that up front and the first quarter's number stays a reading instead of becoming a target.

Google adds a warning that belongs on the same slide, because it applies to every vendor in this category including us: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics."

What not to put on the slide: a share-of-voice percentage with no stated denominator, an average across the network, and any number presented as a rank.

Frequently asked questions

Should we check AI search visibility by brand or by location?

Both, and separately. The brand check answers "what does an assistant say about us", which is one answer for the whole network. The location check answers "does an assistant name this site when somebody in this suburb asks", which is a different answer at every address. Networks that run only the brand check are usually measuring their best-documented site by accident.

Can one bad location affect how AI describes the whole brand?

In the location answer, no. A thin profile in one suburb costs that suburb and nothing else. In the brand answer, yes: a site carrying a wrong phone number, an old address or a much lower rating is part of the material an assistant reads when somebody asks about the brand by name, and nothing makes the model prefer your strongest site.

Is AI search visibility the same as our local ranking?

No. Local ranking is one of the inputs an assistant reads, not the output it produces, and an assistant can name a site that is nowhere near the map pack for the same query. Track both, separately, and do not let a rankings report stand in for the sweep.

About half of US adults (49%) now use AI chatbots, Google says AI Overviews has over 2.5 billion monthly active users, and 45% of US consumers have asked an AI tool for a local business recommendation. Every figure below carries its source.

Most AI SEO statistics pages are a pile of numbers with no year on them, tracing back through three blogs to a study nobody can open. This page is the opposite. Forty-four AI search statistics. We opened every one at the publisher's own page on 6 September 2026, and every one carries its edition, its sample and a link to the source. One publisher is the exception: BrightLocal sells review software of its own, so we name it in full and do not link it. Where a publisher does not state a sample, it says so.

If you want the explanation rather than the numbers, AI search optimization for local businesses is the piece these figures sit under.

How many people use AI assistants for search

• 49% of US adults say they ever use AI chatbots like ChatGPT, Gemini or Copilot, up from 33% in 2024. Pew Research Center, "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact", published 17 June 2026, 5,119 US adults surveyed 17 to 23 February 2026.
• 24% use a chatbot daily, made up of 12% who use one several times a day, 8% about once a day and 4% almost constantly. Another 25% use one several times a week or less. Pew Research Center, 17 June 2026.
• 42% use chatbots to search for information, the most common use of all. Work tasks come second, at 38% of employed adults. Pew Research Center, 17 June 2026.
• 44% of US adults use ChatGPT, up from 34% a year earlier, 23% in 2024 and 18% in 2023. Pew Research Center, 17 June 2026. Read the trend with Pew's own note attached: from 2023 to 2025 the question was "Have you ever used ChatGPT?" and went only to people who had heard at least a little about it, and in 2026 it asked about other chatbots too.
• Gemini is used by 24% of US adults, Copilot by 17%, Meta AI by 14%, Grok by 8%, Claude by 6% and Character.ai by 3%. Pew Research Center, 17 June 2026.
• ChatGPT has more than 1 billion weekly active users, up from more than 900 million six months earlier, when OpenAI also reported more than 50 million consumer subscribers. OpenAI, "A milestone in expanding access to AI", 31 August 2026, and OpenAI, "Scaling AI for everyone", 27 February 2026.
• The Gemini app has 950 million monthly active users, with daily active users tripling in the previous year. Alphabet Q2 2026 earnings call, remarks from Sundar Pichai, 22 July 2026.
• Three-quarters of consumer ChatGPT conversations are practical guidance, seeking information and writing, and 49% of messages are classified as "Asking" rather than task completion. OpenAI and Harvard's David Deming, "How people are using ChatGPT", a National Bureau of Economic Research working paper published 15 September 2025, from a privacy-preserving analysis of 1.5 million conversations on consumer plans.

The most common thing people do with an assistant is ask it something they would once have typed into a search box.

How many people use AI for local business recommendations

The public record is thinnest here and it is the part a business with an address actually needs. One current survey covers it properly.

• 45% of consumers used an AI tool to get a local business recommendation in the past year, up from 6% a year earlier. BrightLocal, Local Consumer Review Survey 2026 (1,002 US consumers).
• 31% used ChatGPT specifically for a business recommendation in the last 12 months. BrightLocal, Local Consumer Review Survey 2026 AI trust analysis (published 10 March 2026, 1,002 US consumers).
• 23% used Google's AI Mode for a business recommendation, putting it second behind ChatGPT. BrightLocal, Local Consumer Review Survey 2026 AI trust analysis (10 March 2026, 1,002 US consumers).
• 42% trust AI platforms for local recommendations as much as they trust reviews. BrightLocal, Local Consumer Review Survey 2026 (1,002 US consumers).
• 97% of the consumers who use AI for recommendations at least sometimes check the answer against real reviews. BrightLocal, Local Consumer Review Survey 2026 AI trust analysis (455 AI users).
• 82% of consumers read AI-generated review summaries. BrightLocal, Local Consumer Review Survey 2026 (1,002 US consumers).

Then there is the practitioner view. Whitespark asked 47 local search experts to score 187 factors, separately for the local pack and for AI search visibility, and the two lists disagree in a way that is easy to miss.

• The presence of a business on expert-curated "best of" and similar lists ranks 1st of 187 factors for AI search visibility. Whitespark, 2026 Local Search Ranking Factors, published 6 November 2025, 47 local search experts.
• The authority of the third-party sites a business's reviews sit on ranks 5th, high numerical Google ratings rank 8th, and the diversity of sites carrying a business's reviews ranks 11th. Whitespark, 6 November 2025.
• A dedicated page for each service ranks 2nd, prominence on key industry-relevant domains 3rd, and a scannable content structure across the whole website 10th. Whitespark, 6 November 2025.

On the local pack list the top two factors are the primary Google Business Profile category and proximity to the searcher, and neither has anything to do with reviews. On the AI list, four of the top eleven are about where else on the web your business and its reviews appear. Whether reviews change what an assistant actually says is a narrower question with a narrower answer, and do reviews affect AI search separates what is documented from what is guesswork.

What share of Google searches show an AI Overview

• 18% of the Google searches in Pew's browsing panel produced an AI summary, and 58% of the panel saw at least one during the month. Pew Research Center, 22 July 2025, 68,879 searches run by 900 US adults who agreed to share their browsing, collected in March 2025.
• 60% of US adults say they ever read the AI summaries at the top of search results. 30% say they do not and 10% are not sure. Pew Research Center, 17 June 2026, 5,119 US adults.
• AI Overviews has over 2.5 billion monthly active users. Google, "New opportunities, control and insights for website owners", 3 June 2026.
• AI Mode passed one billion monthly users within a year of launch, with queries "more than doubling every quarter since launch". Google, "A new era for AI Search", 19 May 2026.
• 80% of the desktop keywords that triggered an AI Overview were informational, transactional keywords were under 3% of all AI Overview triggers, and 82% of desktop AI Overviews appeared on keywords with fewer than 1,000 monthly searches. Semrush AI Overviews study, published 22 July 2025, 200,000 keywords (100,000 desktop, 100,000 mobile) sampled 1 to 10 September 2024. Dated. It is the only published intent breakdown of that size, and the AI Overview has changed a great deal since the sample was taken.

Two of those numbers get conflated constantly. Pew's 18% is the share of one panel's real searches that returned a summary in March 2025. Google's 2.5 billion is a user count, not a query share. Neither tells you how often an AI Overview appears on the terms your own customers use, which you can only get by looking. AI Overviews for local businesses covers how to check your own.

What AI answers do to clicks

• Users who saw an AI summary clicked a traditional search result on 8% of visits, against 15% of visits where no summary appeared. Pew Research Center, 22 July 2025.
• They clicked a link inside the summary itself on just 1% of visits. Pew Research Center, 22 July 2025.
• They ended the browsing session outright on 26% of pages carrying an AI summary, against 16% of pages without one. Pew Research Center, 22 July 2025.
• The presence of an AI Overview correlates with a 58% lower average clickthrough rate for the top-ranking page, worse than the 34.5% the same team measured in April 2025. Ahrefs, "Update: AI Overviews reduce clicks by 58%", 4 February 2026, 300,000 keywords split evenly between those with and without an AI Overview, comparing December 2023 with December 2025.
• The effect shrinks down the page: 58% at position one, 50.8% at position two, 46.4% at position three and 19.4% at position ten. Ahrefs, 4 February 2026.

Google's own position is two claims, neither of them comparable with the studies above. On the Q2 2026 earnings call it said it is "now sending billions of clicks to websites every week through AI features in Search" (Alphabet, 22 July 2026) without publishing a denominator, and its documentation says that when people click from a results page carrying an AI Overview, "these clicks are higher quality (meaning, users are more likely to spend more time on the site)" (Google Search Central, AI features and your website, updated 10 December 2025). Nobody outside Google can check either one, which is worth saying rather than repeating them as findings.

Which sources AI answers cite

• 88% of AI summaries cited three or more sources and 1% cited a single source, with a median summary length of 67 words. Pew Research Center, 22 July 2025.
• 37.9% of the URLs cited in AI Overviews also appear in the first ten results on the same search. 31.2% sit between 11 and 100, and 31% do not appear in the top 100 at all. Count organic blue links only and the non-ranking share rises to 36.7%. Ahrefs, "Update: 38% of AI Overview citations pull from the top 10", 2 March 2026, 863,000 keyword SERPs and 4 million AI Overview URLs from Ahrefs Brand Radar.
• The most-cited domains in AI Overviews are youtube.com at 22.9%, reddit.com at 18.5%, facebook.com at 10.1%, google.com at 8.8%, instagram.com at 5.6%, quora.com at 4.7% and en.wikipedia.org at 4.0%. The denominator matters: Ahrefs calls this mention share, meaning a domain's citations as a percentage of the citations collected by the 50 most-cited sources, not of every citation. Ahrefs, "The 50 most-cited websites in Google AI Overviews (September 2026)", published 2 September 2026, over 3 million US queries.
• YouTube URLs account for 5.6% of all AI Overview citations and 18.2% of the citations that do not rank in the top 100. Ahrefs, 2 March 2026.

Nearly a third of citations come from pages that do not rank at all, and the domains at the top of the list are platforms other people post on rather than company sites. It supports the same finding Whitespark reported.

How Google's AI Mode is rolling out

Every total in this section was superseded within months, so it is a chronology and the older entries are milestones rather than current facts.

• 28 October 2024. AI Overviews "will begin rolling out in more than 100 countries and territories around the world", which Google said at the time would give it "more than 1 billion global users every month". Google, "AI Overviews in Search are coming to more places around the world". The billion is a forecast in that post, not a measurement.
• 5 March 2025. AI Mode launches as an early, opt-in Labs experiment in the US, offered first to Google One AI Premium subscribers. Google, "Expanding AI Overviews and introducing AI Mode".
• 21 August 2025. AI Mode expands "to over 180 new countries and territories in English". Google, "AI Mode in Search gets new agentic features and expands globally".
• 19 May 2026. Personal Intelligence in AI Mode expands "to more people in nearly 200 countries and territories across 98 languages", with no subscription required. Google, "A new era for AI Search".
• 22 July 2026. "Since expanding AI Mode globally last October, we've surpassed 1 billion monthly active users." Alphabet Q2 2026 earnings call.

What businesses are doing about AI search

• Google finished rolling out its AI-features controls and reporting to all websites worldwide on 31 August 2026. Site owners get a Search Console toggle deciding whether their site can appear in and ground generative AI responses, plus impression and geography data for AI answers. Google, "New opportunities, control and insights for website owners", published 3 June 2026 and updated 31 August 2026.
• Sites appearing in AI Overviews and AI Mode are counted inside the existing "Web" search type in Search Console, not in a separate AI report. Google Search Central, updated 10 December 2025.
• Google says no new files or markup are needed. Its words: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search", and on llms.txt specifically, "Google Search ignores them". Google's guide to optimizing for generative AI features, updated 10 July 2026.
• 28% of domains now publish a valid llms.txt file. Ahrefs llms.txt study, published 15 June 2026, 137,210 domains using Ahrefs Web Analytics in May 2026.
• 97% of those llms.txt files received zero requests in May 2026. Of the requests that did arrive, 96% came from bots and 19.5% from AI bots, and no AI bot went looking for an llms.txt file that did not exist. Ahrefs, 15 June 2026.
• Adding schema markup did not increase AI citations. Citations in AI Overviews fell 4.6%, a drop the authors call statistically significant; AI Mode rose 2.4% and ChatGPT 2.2%, both indistinguishable from zero. In a separate look at 6 million URLs, 53% of AI-cited pages carried JSON-LD, making them almost three times more likely to have it than uncited pages, which is a correlation the experiment did not reproduce. Ahrefs schema and AI citations study, published 11 May 2026, a matched difference-in-differences design over 1,885 treated pages against 4,000 controls, August 2025 to March 2026.
• Searches for the vocabulary grew faster than the evidence for it: "generative engine optimization" up 997% and "geo vs seo" up 982% over eighteen months, with "AI search tracking" up 184%. Ahrefs, "5 AI search trends I'm seeing in 2026", 24 July 2026, Ahrefs Keywords Explorer, US volumes.
• ChatGPT Ads reached $1 billion in annualised revenue run rate in under 200 days and is available in more than 40 countries, with Ads Manager opening the platform to small and medium-sized businesses in May 2026. OpenAI, 31 August 2026.

Demand for a technique is running well ahead of the evidence that any technique works. The published evidence still points to ordinary search fundamentals and mentions of your business elsewhere on the web. If you want to know where you stand today rather than in aggregate, how to monitor AI search visibility is a procedure you can run by hand in ten minutes, and does ChatGPT recommend businesses like yours covers what one good answer does and does not prove.

How these AI SEO statistics are checked

We opened every figure above at its publisher's own page on 6 September 2026. Nothing here comes from another statistics roundup, and nothing appears without a publisher, an edition or date, and the sample where the publisher states one.

Four rules govern what gets in.

1. A figure has to be readable at source. If the study is gated, unlinkable or has moved, it does not go on the page. If the number is a count we made by tallying a list rather than a figure the publisher printed, it does not go on either.
2. Nothing sampled before 2025, unless it is the only published figure on a point that matters. One figure qualifies, the Semrush intent breakdown taken from September 2024 keywords, and it is marked Dated. The rollout chronology is a deliberate exception: those entries are milestones and each one carries its date.
3. Publisher, edition, sample. Where a publisher does not state a sample, we say so rather than fill it in.
4. One link per figure, to the primary source, unless the publisher is a competitor of ours. BrightLocal sells review software, so we cite its figures by publisher, edition and sample and do not link them. Everything else is linked.

Two things about the numbers as a set. Vendor figures and independent figures answer different questions: Google and OpenAI report their own user counts and have every reason to report them, while Pew measures real browsing on a recruited panel and Ahrefs measures its own crawl of search results. Neither kind substitutes for the other, so they are kept apart. And almost every consumer figure here is US-only, which is why the samples are stated. There is no equivalent Australian or New Zealand panel study of AI search behaviour that we could open, and we would rather say so than borrow a US number and call it local.

The review side of the same picture lives on online review statistics, which is maintained to the same rules.

Last checked: 6 September 2026. Next check: December 2026. This page is re-checked every quarter. Figures whose source has moved or been withdrawn come off, superseded editions get replaced, and the date above changes. If a number here is wrong or out of date, tell us and it gets fixed rather than left.

We keep this page current because Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named, and one sample of a non-deterministic system is exactly as much as that is worth.

AI search visibility tools all do the same four things. They put a fixed list of questions to the AI assistants on a schedule, record whether your business was named and which rivals were, repeat the run so you can compare this month against last, and turn the count into a percentage. What separates one from another is which assistants it asks, how often, whether it can ask per location rather than per brand, whether it checks that the facts in the answer are right, and whether it will show you the arithmetic behind the number.

Cloutly publishes this page, and Cloutly sells one of the tools on it. That is the disclosure, and it belongs up here rather than in a footnote. Every fact about another vendor comes from that vendor's own public pages, read on 6 September 2026, and where a vendor does not publish something this page says so instead of borrowing a figure from a comparison blog. There is no "cheaper than" anywhere below, because these products are not priced in the same unit and the arithmetic would be dishonest. The tools are listed alphabetically. If a vendor's site says something different today, the vendor's site is right and this page is stale.

The category also gets sold as AI visibility tools, GEO tools, AEO platforms or a ChatGPT visibility tracker. The labels move faster than the products do. Read the specification instead.

What AI search visibility tools actually do

The whole category is a sampler with a spreadsheet attached. It runs your prompts, stores the answers, and counts.

That matters because of what sits underneath. An assistant asked the same question twice can name a different set of businesses, so no tool here is reading a ranking. Each one samples something that moves and presents the average as a number, and a product reporting a stable rank is presenting a sample as a fact. AI visibility explained has the arithmetic, including why two tools pointed at the same business in the same week disagree. This page is the buying decision on top of it.

Read Google's guidance before you spend anything. Its guide to optimising for generative AI features carries a warning box: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems." That is not an argument against buying one. It is the standard to hold a vendor to.

Six questions that decide which AI search visibility tool you buy

Ask these on the call. Most can be answered from a pricing page before you book anything.

1. Which assistants does it ask?

Get the list, not the count. "10+ AI platforms" is a number that includes Grok, Meta and Mistral, and if your customers ask ChatGPT and Gemini then eight of the ten are padding.

Ask which assistants come with the plan you are quoted, because coverage is the most common upsell here. Profound's entry plan tracks ChatGPT and nothing else. AIclicks lets you pick three models from a list of eleven, and the fourth is the next tier up. Nightwatch is the exception, with the same six on every plan.

Then ask how the tool reaches the assistant. SE Ranking is the only one of the six that answers in public: "SE Ranking's AI Search Toolkit retrieves results through direct UI-based monitoring. It sends queries to AI platforms and captures the responses as a real user would see them." A model reached by API is not always the model your customer sees in the app.

2. How many prompts, how often, and how many samples of each?

Three numbers, and vendors usually publish the first two.

Prompt count is what you are buying. Fifty prompts is one business with a good list of questions; five hundred is a network. Frequency is how quickly you get a trend, and every tool below runs daily on some plan. The third number is the one to say out loud: how many times is each prompt asked, per assistant, per month? Vendors who publish a monthly response allowance beside the prompt count are telling you indirectly, so divide. Nightwatch's Starter is 50 prompts against 1,500 answers; Profound's is 50 against 1,500.

Watch the entry tiers on the platform products. Ahrefs includes AI prompts in every paid plan, but Lite carries five of them and 150 checks a month, which is a taste rather than a programme.

3. Does it work per location, or only per brand?

This question eliminates most of the category for a business with addresses, and nobody asks it in the demo.

A brand-level tool measures whether the assistants talk about your company. A location-level tool measures whether they name your Parramatta site when somebody in Parramatta asks. Passing the first tells you almost nothing about the second. Coastline Physio can be described warmly by every assistant on earth and still be absent from "best physio in Bondi" at nine of its eleven clinics.

So ask what the geographic unit is. It is usually nothing at all, a country, or a place you can name. Ahrefs bills a check as one prompt on one platform in one location. Nightwatch advertises prompts tracked in any location. Profound counts regions, and both published plans include one. AIclicks counts countries per project. A country is not a suburb, and for a service business the suburb is the whole question. AI search visibility for multi-location brands covers why the flagship gets named and the suburbs do not.

4. Does it check whether the answer is right, or only whether you were in it?

Being named in an answer that gives out a phone number you gave up two years ago is worse than not being named, because the customer rings the number.

Most tools here score presence and sentiment. Accuracy is rarer: of the six below, five document no accuracy check anywhere public, and ours checks one field, the phone number. Ask anyway, because the answer tells you what the product is for. If the tool does not check, the check is manual and belongs in your log. When AI gets your business details wrong is what you do once you find one.

5. Will it show you the denominator?

Every product here reports a share-of-voice or visibility percentage, and the percentage means nothing until you know what it is a percentage of.

Ahrefs publishes its definition plainly: AI Share of Voice is "the percentage of AI responses within your topic set that mention or cite your brand versus competitors". That sentence is usable, because it tells you the denominator is your topic set and therefore yours to change. Profound describes its method but publishes no formula behind any score.

The consequence is practical. Add three prompts your rivals happen to win and your score falls with nothing changed about your business. Ask what counts toward the number, whether custom prompts feed it, and whether movement is reported in percentage points or as a count of prompts that flipped. A score you cannot reconstruct is a score you cannot act on.

6. What does the price move with?

The billing units here are the plan, the check and the location. Per-plan pricing is a flat fee with a prompt allowance, predictable until you hit the ceiling. Per-check pricing, where one check is one prompt on one platform in one location, gets expensive at scale in a way you can forecast on a napkin. Per-location pricing tracks the thing you are measuring, which suits a network and is unnecessary for one shop.

Ask what the bill looks like at location twenty before you sign anything, and ask whether AI visibility sits on the tier you are being quoted or the one above it.

Best AI visibility tracking tools: what six vendors publish

Six tools, listed alphabetically. Five are products Google returns for this search; the sixth is ours. Every cell comes from the vendor's own public pages, read on 6 September 2026, and where a vendor does not publish something, this table says so.

Read the price column with the units in mind. One of these bills per check, one bills per location, and the rest bill per plan with a prompt allowance. The digits are not comparable and lining them up would mislead you.

Ahrefs

Brand Radar is the most transparent product here about how its numbers are built. Its FAQ defines the four metrics it leads with, and it says out loud where the index is weak: it "works best for established brands or topics with meaningful search demand, since the index is built on Ahrefs' keyword database and may have limited coverage for brands with little or no search volume". A suburban physiotherapy clinic is exactly that case.

It is also the only per-check model here. Plans run from US$129 a month (Lite) to US$1,499 (Enterprise, annual commitment required), each with a prompt allowance: five prompts and 150 checks on Lite, twenty and 600 on Advanced. Brand Radar sells standalone from US$199, and prompt packages start at US$50 for 2,500 checks, so a multi-location sweep prices out exactly. No contracts and no trial: asked about discounts, Ahrefs answers "We never run discounts" and points site owners at its free tier.

AIclicks

AIclicks scales on prompts and models rather than seats. Starter is US$59 a month for 30 tracked prompts, three models and one country per project; Pro is US$189 for 150 prompts, four models and unlimited countries; Business is US$499 for 300 prompts and six models. Prompts run daily on all three, and every plan publishes a monthly response allowance beside the prompt count.

Its FAQ is straight about what drives the bill: "Pricing is based on three things: the number of prompts you track, the number of AI engines monitored, and the volume of AI answers analyzed per month." Trials are three days, no credit card. The geographic unit is the country and the pricing page describes nothing smaller, which is fine for an agency and wrong for one business across eleven suburbs.

Cloutly

Our own scope, stated the way the product states it. Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month, and records whether each location was named and which rivals were. There is an on-demand check across all three with a daily allowance. It does not sample Copilot or Google AI Mode.

The unit is the location. Each site gets its own questions and its own row, and the league table sorts worst first. Share of voice counts the standard questions only, and movement between sweeps is reported as flips: how many locations gained a mention and how many lost one. The fact check reads the phone number and nothing else. Results are a point-in-time sample of a system that is not deterministic, one sample each, so a re-ask can differ.

Pricing is published per location on our plans page, in the visitor's own currency, and reads US$39, US$55 and US$69 today, with 14 days free and no card. AI search visibility sits on the Growth plan, the US$69 one, where the inclusion line names ChatGPT and Gemini and the sweep also covers Perplexity. There is no content optimisation, no crawler analytics, no prompt research and no Grok, because this is one instrument inside a review management product rather than an AEO workbench.

Nightwatch

Nightwatch is a rank tracker that added AI prompts, and its location handling is why it is on this list. Every plan carries the same six AI platforms, and prompts are "tracked daily, in any location or language" across what the pricing page calls "107,000+ locations across 190+ countries".

Plans are Starter at €79 a month for 50 AI prompts and 1,500 AI answers, Professional at €159 for 150 and 4,500, and Agency at €399 for 500 and 15,000, all with a 14-day trial and no credit card. Sentiment and citation analysis sit on every plan, but the compare table puts white-label reports, API access and unlimited seats on Professional and above, which is worth reading against the "no per-seat fees" line at the top. Prices are in euros, and it is a rank tracker first, so you buy keyword tracking to get the prompts.

Profound

Profound is the most enterprise-shaped product here. Starter is US$99 a month and tracks ChatGPT only, with 50 prompts and 1,500 responses. Growth is US$399 for three answer engines, 100 prompts and 9,000 responses. Enterprise is custom and opens up to nine engines. Prompts run daily on all three tiers, and a "billed yearly" toggle offers two months free.

Its FAQ describes the method in one line: "Profound runs structured prompts across AI platforms daily, analyzing where and how your brand appears in responses, tracking citations, sentiment, ranking, and competitive presence." What it does not publish is the formula behind any score, so question five is live on the sales call.

The line to read carefully is Regions: one on Starter, one on Growth, custom on Enterprise. Languages are the same. For a national brand asking national questions, one region is right. For eleven clinics in eleven suburbs, you are in the Enterprise conversation before you start.

SE Ranking

SE Ranking sells AI visibility inside an SEO platform, and it is the most explicit of the six about collection method, quoted in question one. Its FAQ says the toolkit analyses ChatGPT, Perplexity, Gemini, AI Mode and AI Overviews. It is also the only vendor here that answers the question buyers actually have about their score: "There's no benchmark yet as AI search is too new for the industry."

Plans are Core at US$129 a month and Growth at US$279, both with 14-day trials and no credit card, listing 100 and 250 prompts to track daily. A separate AI Search add-on is US$89 a month for 200 prompts and competitor research; whether those prompts stack with the in-plan allowance or replace it is not stated, so confirm it on the call. The free checker gives five checks a day with no account. On location it says nothing a service business can use: no city, country or market unit at all.

Do the free checks before you buy anything

Two of these cost nothing and will tell you whether you need a subscription.

Google publishes AI performance data to the people who own the site. Its guide sends you to the Generative AI performance report in Search Console to see "how people are discovering your content through generative AI features on Google Search". It is first-party, it is free, and no third-party tool has better access to it than you do.

Then run the vendors' free checkers. Ahrefs runs one with no signup, and SE Ranking gives five checks a day without an account. Neither gives you a series, because one check is one sample, but in ten minutes they will tell you whether the assistants name you at all. If the answer is a flat no, buying a dashboard to watch a zero is an expensive way to feel busy. How AI search decides which local businesses get named is the work list then.

When a spreadsheet is enough

Use this test before you buy one.

You do not need a tool yet if you have one or two locations, you can write down eight prompts your customers would actually type, and you will spend twenty minutes on the first of every month. How to monitor AI search visibility is the procedure: eight prompts, a signed-out assistant, a fresh chat for each, and a seven-column log of whether you were named, which rivals were, what was cited and whether the facts were right. Doing it by hand has one advantage no dashboard matches, which is that you read the answers.

What breaks it is arithmetic. Eight prompts across three assistants is 24 answers a month at one location and 264 at eleven, and nobody does 264 by hand twice. The other thing that breaks it is a person: if the check depends on whoever remembers, it stops in the third month. When either is true you are buying automation, and you should price it as automation.

How to run a trial that tells you something

Two weeks is plenty if you decide beforehand what you are testing. Most trials here get spent admiring the dashboard.

1. Enter your real prompts on day one, not the ones the tool suggests. Discovery questions with no brand name in them, per location. A tool scored on brand-name prompts flatters everybody.
2. Check the geographic unit immediately. Set one prompt to a suburb rather than a country and see whether the answer changes. If it does not, the tool is not asking a local question.
3. Ask the same prompt twice in one day, once through the tool and once by hand. That is the sampling noise you will be reading all year, and you want its size before you start reacting to it.
4. Open one answer in full. If you cannot read the paragraph the assistant actually wrote, you are buying a number rather than a finding.
5. Ask support for the denominator in writing. A vendor who cannot answer that in a sentence has not decided.
6. Add a second location and see whether the price moves the way you expected and whether reporting rolls up or leaves you exporting.

On the last day, ask what you would have done differently because of the tool. If the honest answer is "nothing", the spreadsheet was the right product.

Questions people ask

What is the best AI search visibility tool? There is no single answer, because these tools are priced and scoped for different buyers. The question that narrows it fastest is whether you need per-location measurement. If you do, most of the category is out immediately, because it counts countries or nothing at all.

How do I check my AI visibility for free? Search Console's Generative AI performance report covers Google's AI features for a site you own. For the assistants, several vendors run free checkers, including Ahrefs with no signup and SE Ranking at five checks a day. And you can ask the assistants yourself: eight prompts, signed out, a fresh chat each time.

Are AI visibility tools accurate? They are accurate about what they observed, which is not the same as being accurate about your position. Each answer is one sample of a system that can reply differently minutes later, so a single result is evidence and a trend over months is a measurement.

How much do AI search visibility tools cost? The published monthly prices among the six above run from US$50 for a prompt package to US$1,499 for an enterprise plan, with one vendor priced in euros and ours priced per location, at US$69 a location a month for the tier that carries AI search visibility. Compare what a month of measurement costs at your number of locations, not the headline.

Do I need a GEO tool if I already do local SEO? Not necessarily, and Google's position is that the underlying work is the same. What a tool adds is the observation: whether the assistants name you, at which addresses, and what they are reading when they do.

AI visibility is how often, and how accurately, an AI assistant names your business when a customer asks it a question you want to be the answer to. It counts four things: whether you were named, whether you were cited, whether the facts were right, and how you compared with rivals.

It is a measurement, not a ranking, and none of it is guaranteed. This is the vocabulary page for how AI search decides which local businesses get named: what the term means, how a score gets built, and where the measurement stops being reliable.

The term arrived with the tools, not with anything Google or OpenAI published, which is why it gets used so loosely. Search it and the top of page one is software: a free report, an analytics dashboard, a tracker. SE Ranking's own FAQ answers the obvious question, what counts as a good score, with the line most of its rivals leave out: there is no benchmark yet, because AI search is too new for the industry to have set one.

What AI visibility measures

Four results come out of one answer, and they fail for different reasons.

Named or not. The binary, and the one that matters. A customer asks for the best physio in their suburb and gets a shortlist of five. You are on it or you are not, and nothing else here counts until you are.

Cited or not. Whether your own pages were among the sources shown. Being named without being cited is common and not a fault: the assistant read a directory page, a roundup or your Google Business Profile and never touched your website. The citation list is the more useful half of the screen, because it names the pages deciding your category.

Facts right or not. The phone number, the hours, whether you bulk bill. An answer that names you and hands out a number you disconnected in 2023 is worse than no answer at all, and you never hear about it.

Share against rivals. Of the answers you sampled, how many named you, and who was named instead. This is the only one of the four that comes out as a number, which is why the tools lead with it.

An owner asking whether they are visible in AI means the first. A tool selling AI visibility means the fourth. Most of the confusion here sits in that gap.

What it is not

It is not a ranking. An assistant's list has an order, and that order is real for the answer in front of you, but it is not a position you hold. It moves with the phrasing, with the location the assistant inferred, and on a re-ask. Read it as one sample with an ordinal attached. Google's local ranking guidance closes off the shortcut as well: "There's no way to request or pay for a better local ranking on Google."

It is not a traffic number either. An assistant that answers well has given the customer what they came for, so a rise in AI visibility need not show up in your sessions at all. Your outcome was never a session. It was a call, a booking, someone walking in. Measure it there and treat visibility as the leading indicator.

AI visibility vs SEO

For Google's own AI features the work is the same work, and Google says so itself in the guide quoted in the next section. What differs is the thing being measured, and the two numbers behave nothing alike.

Google provides first-party reporting for its own surfaces and nobody else does. Search Console's generative AI performance report, live for all websites worldwide since 31 August 2026, counts impressions in AI Overviews and AI Mode. For ChatGPT, Gemini and Perplexity there is no equivalent, which is why this gets measured with a log or a tool rather than a dashboard you already own.

AEO, GEO and AI visibility name the same work

Google takes the acronyms on directly in its guide to optimizing for generative AI features, last updated 10 July 2026: "AEO stands for 'answer engine optimization' and GEO for 'generative engine optimization'... From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

The same page closes off the tactics usually sold alongside them. On files and markup: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." On style: "You don't need to write in a specific way just for generative AI search."

AEO and GEO name the work, and Google says the work is SEO. AI visibility names what you measure afterwards. If somebody quotes you a GEO retainer on top of an SEO retainer, ask which line item is not already SEO. What the acronyms mean for a local business has the longer version.

What an AI visibility score is, and how it is calculated

Every score in this category is built the same way, whatever it is called.

1. A prompt set. A fixed list of questions, chosen by whoever set the tool up.
2. A set of assistants, in whatever combination the tool covers: ChatGPT, Gemini, Perplexity, Copilot, AI Mode.
3. A run. Each prompt asked once, or a handful of times, per assistant, per period.
4. A count. How many of those answers named you.
5. A percentage, sometimes weighted by where in the list you appeared, sometimes by whether your own site was cited.

The arithmetic is worth doing once by hand. Eight prompts, three assistants, once a month, is 24 answers. Named in nine of them is a score of 37.5%. Weight a first mention above a fifth and the same 24 answers give a different number. Add a ninth prompt your rivals happen to win and the score drops with nothing changed about your business.

That is the limit of what it carries: a direction, for one business, over several months, as long as the questions stay identical. It cannot carry a level, because the denominator is yours. A vendor quoting you an industry benchmark has invented it.

Cloutly's AI search visibility ranks each location by share of voice, counted from the standard questions only, since custom queries do not feed it. Movement between sweeps is reported as flips, meaning locations that improved and locations that lost, never as a change in share points.

Why two tools give you two different scores

Run two trackers on the same business in the same week and the scores disagree. Four reasons, none of them a bug, and what the tools actually measure covers how to choose between them.

The prompts are different. There is no standard set. One tool asks discovery questions, another asks about your brand by name, and brand-name prompts score far higher, because an assistant asked about you will talk about you. Two tools asking different questions are measuring different things and calling both visibility.

The sample size is different. One answer per prompt per month, or ten. Every extra sample costs the vendor money and steadies the number, so each picks a point on that trade-off and few publish where they landed. Ask before you compare anything.

The model is different, even inside one assistant. Google's AI Mode help page describes a Pro option running Gemini 3 Pro next to the Fast model that handles most questions, and you have to be signed in to a personal Google account to choose it. A tool querying an API is not necessarily reaching the model your customer sees in the app.

The system is not deterministic. The same help page tells searchers to "ask multiple versions of your question to get the best answers", because AI Mode splits a question into subtopics and searches each of them separately. Read that as an admission: the answer to one phrasing is not the answer to the question. A tracker that asks once has sampled once, and the location and the account it asked from move the result again, which how to monitor AI search visibility works through in detail.

Google's warning about the tools built on all this applies to every vendor here, us included: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems."

How to measure AI visibility without a tool

With one or two locations, use a log rather than buy a tool.

1. Write eight questions a customer would type, covering plain discovery, a constraint, a problem, your business name and you against a rival. Then stop changing the sentences.
2. Ask each in a fresh chat, signed out where the assistant allows it, with the suburb in the wording.
3. Record four things: were you named, who else was, what was cited, were the phone number and hours right.
4. Same day each month. Read the series, not the run.

The full procedure, with the eight prompt shapes, the private-mode setting each vendor documents and a seven-column log, is in how to monitor AI search visibility.

What moves AI visibility

AI visibility moves with ordinary local-search inputs: a complete Google Business Profile, a review count and rating that keep growing, reviews whose text describes the actual job, the same phone number and hours everywhere you are listed, pages elsewhere that mention you, and a site the crawlers can reach. The parent piece sources each item and gives the order to work in.

The one a review vendor can speak to rather than an SEO is the text. Nobody writes "great service" in an answer about rotator cuff rehab, so a profile whose reviews all say "highly recommend" gives an assistant nothing to match against a specific question. Whether reviews change what AI says separates what is documented from what is assumed.

Frequently asked questions

What does AI visibility mean in plain terms?

Whether an AI assistant names your business when a customer asks it for one, and whether what it says about you is correct. Every metric in the category counts those two things across more questions and more assistants.

Is AI visibility the same as SEO?

The work is. Google says optimising for its generative features is still SEO, and that no separate files, markup or writing style are needed. The measurement is not: a ranking is a position in an index you can check, while AI visibility is whether a freshly generated sentence names you.

What is a good AI visibility score?

There is no benchmark, and a vendor quoting one is quoting their own arithmetic. Compare a location against itself over several months with the questions held identical, and against the rivals named in the same answers. Ignore the absolute figure.

Can you pay to improve AI visibility?

No. Google's local ranking page says there is no way to request or pay for a better local ranking. OpenAI's help centre is just as plain about ChatGPT: it "ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." Anyone selling a guaranteed mention is selling something they cannot deliver.

Yes, and only one vendor documents it plainly. Google's Gemini API says its Maps grounding step "queries Google Maps for relevant information (e.g., places, reviews, photos, addresses, opening hours)", and answers questions about a place "based on Google user reviews and other Maps data". Everything else written about AI and Google reviews is inference from that, or somebody's guess dressed as a ranking factor.

Every quote below has a link and a date. Where a vendor has published nothing, this page says so instead of filling the space. AI search optimization for local businesses explains how these systems answer local questions; this page covers the reviews input.

AI and Google reviews: which assistants document reviews as an input

Four surfaces, and they are not equally forthcoming. Only Google names reviews as something retrieved, and only in its developer documentation.

Two things follow that most of the advice on this topic skips.

The Gemini quote comes from the developer API, which is what an application gets built on rather than a description of every answer the Gemini app gives. Read it as the clearest published statement that review text gets retrieved and read, not as proof that it happens on every query.

And for ChatGPT and Perplexity, your reviews are not a signal the assistant holds an opinion about. They are content on a page. If a Google Maps place page, a directory listing or a local roundup carrying your reviews comes back in the retrieval, the text of those reviews is in front of the model. If nothing carrying them comes back, they may as well not exist. OpenAI is blunt about the ceiling: "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed."

How AI reads reviews: the nouns are the point

Reviews are the longest piece of writing about your business that you did not write. That is the whole argument. A model answering "who can see a rabbit in [suburb] on a Saturday" is looking for text that matches "rabbit" and "Saturday", and the only place those words are likely to appear about your clinic is in what your customers wrote.

Google demonstrates this on its own surface. The Places API's AI-powered review summaries (updated 1 September 2026) are "AI-generated summaries of places based solely on user reviews", produced "by synthesizing key elements of user reviews, such as place attributes and reviewer sentiment". Solely. Not your description, not your website copy. It is a developer feature, and Google says the summaries are not guaranteed for every place, so do not read it as something every listing gets. Read the principle: given a place, Google will write a description of it out of review text alone.

Take Rowan Street Vet, a fictional two-vet clinic with 96 reviews at 4.8 stars. Eighty of them say some version of "lovely staff, highly recommend". Those eighty are not useless. They hold the rating up, and the rating is real. But there is no sentence in them about rabbits, about after-hours, about a nervous dog or the cost of a dental. Nothing in them matches a question.

The other sixteen say things like "took our rabbit in on a Saturday morning when nobody else would look at an exotic" and "explained the dental quote line by line before doing anything". Those sixteen name a species, a day, a service and a behaviour. When somebody asks a question shaped like any of those, there is text to match.

Nothing about that requires you to script anything, and scripting it breaks Google's rules anyway. It follows from when you ask. A customer asked the same afternoon writes about the thing that just happened to them. A customer asked six weeks later writes "great service", because that is all they still remember.

Do review volume and recency change what AI says?

Volume and rating have documented weight in one place, and it is not an AI system. Google's local ranking page describes prominence as "how well-known a business is", then: "This factor's also based on info like how many websites link to your business and how many reviews you have. More reviews and positive ratings can help your business's local ranking." That is a statement about local Search and Maps. Gemini grounds on Maps, and a rewritten ChatGPT search returns local results, so the effect reaches the assistants second-hand. Second-hand is still real. It is also still second-hand, and most of the pages on this topic quietly drop that word.

Recency is different. Nobody documents it. Not Google, not OpenAI, not Perplexity. What can honestly be said is an observation: a profile whose most recent review is from 2023 reads as abandoned to a person, and retrieval systems generally favour current pages. Both are reasonable. Neither is published. So when somebody hands you a freshness window or a monthly review quota, the useful question is which document it came from, and the answer will be that it came from them.

Weighting is the same story. Nobody publishes what a review is worth against a link, whether Google reviews count differently from a trade directory's, or whether any threshold exists at all. Google's position on its own local ranking is that "we do our best to keep the search algorithm details confidential to make the ranking system as fair as possible for everyone", and no assistant has been more generous than that. Whether reviews move your local ranking is the fuller version of the documented half.

Your replies are text too

Everything above applies to what you write under a review. Google puts it in the reader-facing form on the same local ranking page: "When you reply to customer reviews, it shows that you value their feedback. Positive reviews and helpful replies can help your business stand out."

Where a reply gets read is worth being precise about. The Places API's review object carries the review text, the rating, the author, the timestamp, a flag link, a Maps link and a visit date. There is no field for the owner's reply. And the AI review summaries are drawn solely from user reviews. A reply is not reaching a model down either of those routes.

It reaches a model the way any other text does. Your reply is visible on the Maps place page and on any site that renders your reviews, so it is there to be read when one of those pages is retrieved.

What that means in practice is unglamorous. A reply that says "Thanks Sam" adds nothing to the page. A reply that names the treatment, corrects a factual error in the review, or states your actual cancellation policy adds a sentence of your own words to the only page about you that a customer and a model both read. Review response examples covers what a good one looks like for each kind of review.

The tactics that do not work, in Google's own words

Three tactics are being sold against this topic right now. All three break Google's prohibited and restricted content policy, which governs the profile you can least afford to lose.

Filtering who you ask. Sending happy customers to Google and routing unhappy ones to a private form is covered directly: merchants must not "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers". It also defeats everything above. If review text is the raw material an assistant reads, a filtered corpus is one that does not describe your business. Cloutly has not had a rating step in front of the review sites since 2021, and review gating is the longer argument.

Incentives. The policy prohibits merchants who "offer incentives" of any kind, naming payment, discounts and free goods or services, "in exchange for posting any review or revision or removal of a negative review". The permitted version sits in the same document and is the whole job: "solicit or encourage the posting of content that does represent a genuine experience, without offering incentives to do so or attempting to influence the rating or the contents of the review."

Asking for specific words. This is the one that looks clever after reading the section above, and the policy names it twice. Merchants must not attempt "to influence the rating or the contents of the review", and the prohibited list covers merchants requesting that staff "solicit reviews that include specific content, including content that identifies a staff member". The template that says "please mention emergency plumbing in [suburb]" breaks the rule in exchange for an effect no vendor has ever documented. Ask promptly instead and the specifics arrive on their own, because the customer still remembers them.

There is a fourth, which is less a tactic than a category error. Search anything in this area and something will offer to write your Google reviews for you with AI. Nothing that comes out of that is a review. It is a paragraph about a visit nobody made, banned on every site that would host it, and worthless to an assistant for the same reason real reviews are valuable: nobody experienced it.

What to do instead

None of it is new work. It is the same three jobs, done for a reason that has changed. The acronyms sold around it are unpacked in GEO vs SEO, AEO and LLMO, and Google's own surface has SEO for Google AI Mode.

1. Ask everyone, at the moment the job finishes. Not a drive, not a filtered list. Everyone, promptly, so the review describes the specific thing rather than the general feeling. How to get more Google reviews covers the ask and the timing.
2. Reply to everything, with a sentence that carries information. The reply is public text on the page an assistant reads, and Google says helpful replies help you stand out.
3. Keep the listing consistent everywhere. ChatGPT reads whatever the web returns about your hours and phone number, and OpenAI's own warning is that "search results and citations can be incomplete, outdated, or incorrect". Adding your business to Google Maps is the setup guide for the profile all of this rests on.

Then check whether any of it moved. There is no report for this, so the honest method is to ask the assistants the questions your customers ask and write down what came back, month after month, treating each answer as one sample rather than a rank. Does ChatGPT recommend businesses like yours has the prompt set and the log. Cloutly runs that sweep for you: AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were.

No review count guarantees a mention in an AI answer. Anyone quoting one has made it up. Specific, current reviews put something about your business in front of the model in language that matches what somebody asked. Whether it gets used is not yours to control.

Frequently asked questions

Does ChatGPT use Google reviews?

Not directly, and OpenAI does not name reviews anywhere in its documentation. ChatGPT rewrites the question into a web search and builds the answer from the pages that come back. If one of those is your Google Maps listing, a directory entry or a local roundup that carries your reviews, the review text is in front of the model. If not, it is not.

How many reviews do I need before AI recommends my business?

Nobody knows, and nobody has published a threshold. Google documents that more reviews and positive ratings can help local ranking, and stops there. A number quoted at you came from the person quoting it.

Do reviews written by AI help?

No. They are fabricated, they are prohibited on every site that would host them, and they defeat the purpose. What makes a review useful to an assistant is that a real customer described a real thing.

Do old reviews stop counting?

No vendor documents an expiry, a decay curve or a freshness window. What is observable is that a profile with nothing recent looks dormant to a human reader, and that these systems retrieve from the current web rather than a snapshot. Treat steady collection as the sensible default, not as a documented requirement.

Does ChatGPT recommend businesses? Yes. Ask it for the best physio in your suburb and it will name three or four of them, with links. Whether it names yours depends on what it can find about you on the open web at the moment somebody asks, and none of the assistants publishes a ranking formula for local recommendations. What OpenAI, Google and Perplexity do publish is where the answer comes from, and that is enough to work with.

Most of what is written on this is guesswork, including several of the pages ranking for it today. So this piece quotes only what the three vendors have published themselves, says where that runs out, and gives you a test you can run in ten minutes a month.

How does ChatGPT recommend local businesses?

ChatGPT answers a local question by searching the web, working out roughly where you are, and reading what it finds. OpenAI's help centre is direct about it: ChatGPT "can search the web to answer questions with current information and links to relevant sources" and "can also use location information to find local results, including restaurants, news, and weather" (Searching the web with ChatGPT, updated August 2026). Four details on that page matter to a business owner.

The question is rewritten before it is searched. OpenAI's own example: ask "What are some good restaurants near me?" and, if ChatGPT works out from your IP address that you are in the San Francisco area, it "may rewrite your prompt into the search query 'top restaurants San Francisco'". What goes out is an ordinary local search, and the pages that come back are the raw material for the answer. If the person has memory switched on, saved facts get folded into the rewrite too, which is one reason the same question from two people returns different businesses.

Location is approximate unless the customer hands over more. ChatGPT uses "an approximate location based on your IP address", and device location sharing "is optional and off by default". Someone on a home connection three suburbs away can be shown a different set of businesses from someone standing outside your door.

The sources are visible. Responses that use web search "may include citations", with a Sources control that opens the full list, and on iOS and Android "relevant results may also include a map". Read that list. It tells you which pages the answer was built from.

A website can be shut out, and no website is ever guaranteed in. OpenAI's crawler documentation says "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links", and a robots.txt change takes about 24 hours to register. If somebody once dropped a blanket AI block into yours, check that before anything else here. Going the other way, the help centre closes off the ranking speculation in one line: "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed."

One correction, because it appears on most of the pages ranking for this question: ChatGPT does not answer local questions out of training data any more. Web search reached everyone, signed in or not, on 5 February 2025. "It only knows what it was trained on" was true in 2023 and is a bad model of the thing now.

Gemini and Perplexity get their answers somewhere else

The three assistants work differently, so your effort should follow the sources each one uses.

Google's Gemini API documents a tool called Grounding with Google Maps, and it is unusually specific about the ingredients. At the data retrieval step the service "queries Google Maps for relevant information (e.g., places, reviews, photos, addresses, opening hours)", the model answers out of that, and the response carries "inline annotations linking to Google Maps sources". The first use case Google lists is handling place-specific questions: "Ask detailed questions about a specific place to get answers based on Google user reviews and other Maps data."

That is the clearest statement any vendor has made that your reviews are read as text when an assistant answers a question about your category.

For Google's consumer surfaces the guidance is shorter and points at the same place. Search Central says generative AI responses "can include product listings, product information, and information about local businesses", and that "Merchant Center feeds and Google Business Profiles can help your products and services to be visible in both AI responses and other Google Search results" (guide to optimizing for generative AI features). Its AI features guide opens by saying there is no separate game to play.

Perplexity is the simplest of the three. Its help centre says it uses AI to "search the internet in real-time" and that "each answer includes numbered citations linking to the original sources" (How does Perplexity work?). Its crawler documentation adds that PerplexityBot exists "to surface and link websites in search results on Perplexity". For a local query, the answer is assembled from whatever public pages mention your category and your area: directories, review sites, roundups, your own site.

For a business with an address, that is three of five surfaces. AI search optimization for local businesses covers all five, including Google's own AI Overviews and AI Mode, and the inputs they share.

Does ChatGPT recommend businesses based on reviews?

Partly, and reviews are the only input on this list that a vendor has documented reading as text. Nobody publishes weightings for any of it. What follows is what the three vendors say they read, ordered by how much you can do about each one.

A complete Google Business Profile. Google's page on local ranking says local results "are mainly based on relevance, distance, and popularity", and that to improve relevance you should "provide complete and detailed business info". You cannot change how far you are from the person asking. You can finish the profile this afternoon.

Review count and rating. The same page, on prominence: "This factor's also based on info like how many websites link to your business and how many reviews you have. More reviews and positive ratings can help your business's local ranking." Gemini grounds on those local results, and a rewritten ChatGPT search returns them.

Review recency, which nobody documents. Google says nothing about it. Neither does OpenAI. It is a reasonable assumption, because a profile whose newest review is from 2023 looks abandoned to a person and probably to a model. It stays an assumption. Anyone quoting you a freshness window as a ranking factor invented the number.

What the reviews say, not only how many there are. The input most owners skip. Gemini's documentation makes it concrete: a question about a place is answered out of review text. Nobody writes "great service" in an answer about rotator cuff rehab. If forty of your reviews name the treatment, the technician, the breed or the machine, you have forty sentences an assistant can match to a question. If they all say "highly recommend", you have none.

The same details everywhere you are listed. ChatGPT rewrites the question into a web search and reads what comes back, so whatever the web says about your hours and phone number is what it repeats. OpenAI's warning is worth putting in front of anyone who assumes accuracy: "search results and citations can be incomplete, outdated, or incorrect."

Pages the assistants can read that mention you. Directories, local roundups, a supplier's stockist page, an association member list, a news mention. These are the citations that turn up under a local answer, and they are why a business with a thin website can still get named.

Can you pay ChatGPT to recommend your business?

No, and that is documented rather than assumed. OpenAI began testing ads in ChatGPT in February 2026, on the Free and Go tiers, and the announcement is explicit on the point that matters here: "ads do not influence the answers ChatGPT gives you", and an ad is "always clearly labeled as sponsored and visually separated from the organic answer" (Testing ads in ChatGPT, updated 11 August 2026). Buying the ad buys the ad. It does not buy the recommendation above it.

There is also nothing to install. Google's mythbusting section is blunt: "you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities)". No llms.txt, no special schema, no AI-flavoured rewrite of your pages.

The shortcut people do try is buying reviews, and it is banned in the place the assistants read. Google's contribution policies disallow "reviews or ratings that have been paid for, directly or in kind" and forbid merchants from offering "incentives such as payment, discounts, free goods and/or services" in exchange for them. Filtering is the same problem in a nicer suit: the policy also stops merchants who "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers", and a filtered profile does not describe your business anyway. Ask everyone or ask nobody.

How to test whether ChatGPT recommends your business

Run the check once a month, on the same day, and write the results down. Ten minutes is enough. A single answer is not the point. The trend across months is, and so is the list of sources you keep seeing.

1. Write eight prompts and then stop changing them. Vary the shape, not the wording, month to month:
   • best [category] in [suburb]
   • [category] near [landmark] open [day]
   • who should I see for [a specific problem you solve] in [suburb]
   • best [category] in [suburb] for [customer type]
   • [category] in [suburb] with [the thing people ask for: parking, late nights, a particular insurer]
   • is [your business name] any good
   • [your business name] reviews
   • [your business name] or [a rival you actually lose to]
2. Name the location in the prompt. OpenAI's advice when local results show the wrong area is to "include your city, neighborhood, or postal code in your question". Do not rely on the assistant guessing it from your office connection.
3. Run every prompt in a fresh chat, in all three assistants, signed out where you can be. Memory and chat history change the rewritten query, so testing inside your logged-in account measures your account, not the world.
4. Record six things per run. Whether you were named, where in the list, which rivals were named, which sources were cited, whether your phone number and hours came back right, and the date.
5. Treat each result as one sample. These systems are not deterministic, so a re-ask can differ. One miss is not a verdict and one hit is not a rank.
6. Read the citations, because that is the work list. If the same directory or roundup is cited every month and you are not on it, that is a concrete job. If your Business Profile is the only source cited, everything rests on one page.

A spreadsheet with seven columns holds all of it. Two rows from a first month, one hit and one miss:

If you would rather not run it by hand, Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were.

What changes when you ask every customer for a review

Here is the shape of it, with an invented business so nobody mistakes this for a case study. Coastline Physio has two sites and 31 Google reviews between them, most written in 2023, most of them one line long. Six months of asking every patient at the end of a course of treatment takes that to 180, all current, a good share of them naming the injury and the physio who treated it.

Three things follow, and they are not equally certain. The local ranking has a documented reason to improve, because Google says more reviews and positive ratings can help it. The reviews now carry language that matches real questions, which follows from Gemini's documentation on reading Maps reviews. And the profile stops looking dormant, which is the recency argument: plausible, undocumented.

What does not change is the certainty. There is no number of reviews that buys a mention in an AI answer, and a vendor promising one is selling something they cannot deliver. What reviews buy is candidacy: you become a business the assistant has something to say about. The rest is the unglamorous work it always was. Get more Google reviews steadily rather than in bursts, keep your Google Business Profile complete and accurate, and reply to the reviews you get so the profile reads as tended. Whether reviews move your local ranking is the longer answer.

FAQ

How do I get ChatGPT to recommend my business? There is no submission form and no setting. Finish your Google Business Profile, make your details match everywhere you are listed, collect current reviews whose text names what you actually do, and check your site is not blocking OAI-SearchBot or PerplexityBot. Then test monthly and work the sources the answers cite.

Why does ChatGPT name my competitor and not me? Usually for a reason you can check. They are cited in a source the assistant reached and you are not, or their profile answers the specific thing that was asked and yours does not, or the assistant placed the customer nearer to them. The Sources list tells you which.

How often should I check? Monthly. More often is misleading: a re-ask can return a different answer, so short-interval checking measures noise. Same prompts, same day each month, in a log you keep.

GEO, AEO and LLMO all describe the same work as SEO. Google's guide to optimizing for generative AI features settles it: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." The inputs have not changed. The measurement has.

Proposals often frame GEO and SEO as separate disciplines. They are not. If you run a business with an address, this is the vocabulary page for how AI search decides which local businesses get named: what each letter stands for, where it came from, and which parts of the pitch were invented.

What each acronym stands for

GEO is the one with a paper behind it

GEO: Generative Engine Optimization, submitted in November 2023 by researchers at Princeton University and IIT Delhi and accepted at KDD 2024, is where the term starts. Its own sentence: "we introduce Generative Engine Optimization (GEO), the first novel paradigm to aid content creators in improving their content visibility in generative engine responses". The paper reports that GEO "can boost visibility by up to 40% in generative engine responses".

That 40% is the figure the whole category quotes, so it is worth knowing what produced it. The winning tactics were citing sources, adding statistics, quoting people with names and titles, and writing more clearly. The tactic that lost ground was keyword stuffing. None of that is new advice, and none of it needs a new discipline to deliver it.

The paper is also more careful than its resellers. The same abstract says "the efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods", and the GEO-bench benchmark it was measured on is "diverse user queries across multiple domains", not a test of which physio an assistant names in one suburb. The research is real. The retainer built on top of it is a different object.

AEO and LLMO have no origin story

AEO expands to "answer engine optimization" and LLMO to "large language model optimization". Google's guide gives the AEO expansion in the same breath as GEO's, which is close to the only authoritative definition either term has. Neither has a paper, a standards body or a documented first use.

Search "what is AEO" today and page one is a Reddit thread, a course seller and a run of software companies, several of them selling an AEO subscription underneath the definition. The term is defined mainly by people with a tool to sell, and in ordinary use it means "write things an assistant can lift".

If you have been sold both AEO and LLMO, you have been sold the same thing twice.

GEO vs SEO: what differs for a local business

Very little, and Google is blunt about the part it controls. Its AI features guide says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

The inputs to a local AI answer were already on your list. A Google Business Profile that is finished rather than started. A review count and rating that keep moving, because Google's local ranking guidance says plainly that "More reviews and positive ratings can help your business's local ranking". Review text that describes the job instead of saying "highly recommend". The same phone number and hours on every site that lists you. Pages elsewhere that mention you. A website the crawlers can reach. The parent piece sources each one and gives the order to work in.

One honest complication: ChatGPT and Perplexity do not run on Google's index, so Google's "no additional requirements" line does not cover them. The work list barely moves anyway. Those two assemble a local answer out of the public pages their crawlers can reach, which means directories, roundups, review sites, and your profile before your homepage. Being present and correct on the pages that already rank for your category is what local SEO has meant for a decade. Does ChatGPT recommend businesses like yours walks that half of it end to end.

What is different: the measurement

A search ranking is a position you can look up, and it holds still long enough to be checked. An AI answer is generated fresh every time it is asked, so being named once is a sample rather than a place you hold.

Reporting is split as well. Google gives you a first-party number for its own surfaces: Search Console's Generative AI performance report counts impressions in AI Overviews and AI Mode, and Google rolled it out "to all websites worldwide" on 31 August 2026. Nothing any other assistant does appears in it, so ChatGPT, the Gemini app and Perplexity end up measured with a log or a subscription rather than a dashboard you already own. AI visibility explained covers how those scores get built and why two tools handed the same business return two different numbers.

When a proposal offers GEO on top of SEO, the new line item is almost always measurement, not method.

The parts that were invented

Google's guide ends on a bullet that could stand as this section's heading: "Prioritize effective SEO strategies over 'AEO/GEO hacks'." Five claims recur in proposals.

Proprietary scores. Every AI visibility score is a count of how many sampled answers named you, divided by how many were sampled, and the vendor picks the denominator: their prompt list, their assistants, their sample size. Change any of the three and the score moves with nothing changed about your business. It carries a direction for one business over several months, with the questions held identical. It does not carry a level, because nobody has set an industry benchmark to compare it against. Google's warning applies to every vendor here, us included: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems." AI search visibility tools goes through what a tracker can and cannot tell you.

The GEO agency. There is no separate discipline to staff. Google sends the question back to ordinary due diligence: "If you're considering third-party 'AEO' or 'GEO' advice or services, review our guidance on evaluating third-party SEO advice." Ask for the deliverables list before you sign, then count the line items an ordinary SEO would have done in 2019: fix the profile, fix the citations, get the site crawlable, publish pages that answer real questions. If that is all of them, you are buying SEO with a newer invoice.

Guaranteed mentions. Nobody can sell you one. Google's local ranking page says "There's no way to request or pay for a better local ranking on Google", and OpenAI's help centre says ChatGPT "ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed."

The file and markup upsell. Google's guide names these one at a time. On llms.txt and its relatives: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." Keep one if you like the tidiness. Google says it ignores them. On style: "You don't need to write in a specific way just for generative AI search." On link building rebranded: "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem."

Keyword templates and rating steps. This one does damage. A request that asks the customer to work in a phrase, and any rating question standing in front of the review sites, both run into Google's prohibited and restricted content policy, which stops merchants soliciting reviews that include specific content or selectively soliciting positive ones. They also spoil the one asset an assistant reads as evidence about you. Review gating is the longer version.

What to do instead

1. Finish the Google Business Profile. Primary category, hours, every remaining field.
2. Check the crawlers can reach you, including any blanket AI bot blocks left over from 2023.
3. Ask every customer for a review, promptly and continuously, so the text describes specific jobs rather than general praise.
4. Fix the two or three directory listings carrying an old phone number.
5. Ask the assistants the same eight questions on the same day each month and write down what came back. Read the series, not the run.

Cloutly's AI search visibility does step five for you: it asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were. Read it as a point-in-time sample of a system that is not deterministic, never as a rank.

Frequently asked questions

Is GEO the same as SEO?

For Google's surfaces, yes, and Google says so in its own documentation: optimising for generative AI search is optimising for the search experience, and thus still SEO. For ChatGPT and Perplexity there is no equivalent statement, but the inputs are the same public pages and profiles, so the work list is the same work list.

What is AEO?

Answer engine optimization: writing so that a system generating an answer can find the answer in your page and repeat it accurately. There is no published origin for the term and no body that defines it. Google's guide gives the expansion and then puts it in the SEO bucket alongside GEO.

What does generative engine optimization mean?

It means improving how often a generative engine, meaning any system that answers a query with composed text rather than a list of links, names or cites you. The term comes from a 2023 paper accepted at KDD 2024, whose tested tactics were citing sources, adding statistics and quoting named experts.

Do I need a GEO agency?

Not as a separate line item. If the deliverables are profile hygiene, citations, crawlability and useful pages, that is SEO under a newer name. What you might reasonably buy is the measurement, since the non-Google assistants publish no first-party report.

Is LLMO different from AEO?

Not in practice. Large language model optimization and answer engine optimization describe the same work in different letters, and neither has a definition anybody can point to. Treat both as vocabulary rather than method, and ask what is on the work list.

SEO for Google AI Mode is the SEO you were already doing. Google puts it in one line in its guidance for site owners: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." No submission, no markup, no file to upload. What changes is the question your customer asks. AI Mode is built for the long comparative questions people used to break into four separate searches.

Everything below is quoted from Google's own help pages, developer documentation and announcements, checked on 6 September 2026. Where Google has published nothing, this post says so rather than filling the gap with a number somebody invented. The broader guide to how AI search decides which local businesses get named covers the other surfaces.

What is Google AI Mode, and where can people use it?

AI Mode is a separate search experience with its own page. Google Search Help calls it "Google's most powerful AI search experience", where "You can ask anything and get an AI-powered response, with the ability to go deeper through follow-up questions and helpful links to the web" (Get AI-powered responses with AI Mode in Google Search). People reach it three ways: at google.com/ai, by tapping AI Mode after typing a question on google.com, or from the home screen of the Google app.

The rollout, in Google's own announcements:

• 5 March 2025. AI Mode arrives as an opt-in experiment in Labs, offered first to Google One AI Premium subscribers (Expanding AI Overviews and introducing AI Mode).
• 20 May 2025. Google begins "rolling out AI Mode in the U.S.", with "no Labs sign-up required" and a new tab in Search and in the Google app (AI in Search: Going beyond information to intelligence).
• 21 August 2025. Google says it will "expand AI Mode to over 180 new countries and territories in English" (AI Mode in Search gets new agentic features and expands globally).

The live country list sits on the help page rather than in the announcement. On 6 September 2026 it named Australia, New Zealand, the United Kingdom, the United States, Canada and Ireland. If you serve any of those markets, your customers can already ask AI Mode who to call. Google maintains that list and changes it, so read the page rather than this sentence. One caveat from the same page, before you treat any answer as authoritative: "AI Mode doesn't always get it right."

AI Mode and AI Overviews are not the same feature

They get talked about as one thing. Google says otherwise: "AI Mode and AI Overviews may use different models and techniques, so the set of responses and links they show will vary" (AI features and your website).

The difference that matters commercially is who arrives. Somebody who opens AI Mode, describes the problem and asks two follow-ups has done their shortlisting inside the answer. By the time they click, the comparison is over. A searcher who glanced at an AI Overview on the way to the map pack has not. AI Overviews for local businesses covers that other surface on its own terms.

What query fan-out means when someone asks for a local business

AI Mode does not run the question you typed, and that changes which of your pages can be pulled into an answer. Google's help page: "AI mode uses a 'query fan-out' technique, dividing your question into subtopics and searching for each one simultaneously across multiple data sources." The developer glossary defines the same thing as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query" (Optimizing your website for generative AI features). Google adds that both AI Mode and AI Overviews may use it, so fan-out is not an AI Mode exclusive.

Google's own example is a lawn. Ask "how to fix a lawn that's full of weeds" and the fan-out queries might include "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn". One question, several searches, one assembled answer.

Now put a local question through the same machine. Somebody types: my dog is limping after a run, who should I take her to in Newtown and what will it cost. Google has not published the fan-out for that query and nobody outside Google can see it, so treat what follows as the shape of the thing rather than a leaked list. It is not one search. It is a question about limping in dogs, one about which kind of clinic treats it, one about vets in that suburb, and one about price.

So your service page is no longer competing only for "vet Newtown". It is one of several documents that can satisfy a branch of a question nobody typed that way. A page explaining what a limp usually means and what an assessment costs answers two branches at once. A page headed "Newtown's most trusted vets" answers none of them.

That is not an invitation to spin up a page per fan-out query. Google bans it in the same guide: creating separate content for every possible variation of how people might search, "primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy". Depth on the questions you get asked, not breadth across guessed ones.

The upside is Google's own claim. Its models "identify more supporting web pages" while the answer is written, so AI features "display a wider and more diverse set of helpful links associated with the response than with a classic web search". More links from more angles is an opening for a business too small to hold position three for its head term.

What AI Mode shows when the question is about a place

Business information is in scope. Search Central's generative AI guide says "Where appropriate, generative AI responses can include product listings, product information, and information about local businesses", and that Merchant Center and Google Business Profiles "can help your products and services to be visible in both AI responses and other Google Search results". Its sibling page turns that into a checklist item: "Checking that your Merchant Center and Business Profile information is up-to-date."

AI Mode also reaches Google's place data, not the open web alone. Announcing the agentic version on 21 August 2025, Google wrote that "AI Mode uses the live web browsing capabilities of Project Mariner, direct partner integrations on Search, and the power of our Knowledge Graph and Google Maps to help users take action on the web", starting with restaurant reservations and "expanding soon to local service appointments and event tickets". The fine print matters: that booking behaviour launched as a US Labs experiment, not as the general feature.

There is one published description of how a place answer looks. Writing about travel on 27 August 2026, Google said an AI Mode hotel search returns "a visual list of options, complete with guest reviews and key factors to compare" (3 new ways to plan and book travel in Search). Guest reviews inside the answer, in Google's own account of its own feature.

What Google has published nowhere: which fields of a Business Profile appear in an AI Mode answer, how a set of businesses is chosen, or any weighting given to reviews inside AI Mode. If a vendor tells you AI Mode reads your last thirty reviews, or that 4.6 stars is the threshold, they made it up. What Google does publish about ranking a local business sits on the local ranking page: results "are mainly based on relevance, distance, and popularity", and "More reviews and positive ratings can help your business's local ranking". That page is about local results, not about AI Mode, and the honest position is that the two overlap without Google having said how much. Whether Google reviews affect your ranking works through the documented half.

SEO for Google AI Mode: the four things you can change

None of them is new. That is Google's point, not a rhetorical flourish.

1. Be indexed and snippet-eligible. That is the whole technical requirement: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements." Businesses fail it by accident, usually a noindex, nosnippet or max-snippet left behind after a redesign. Search Console also carries a setting that excludes a site from generative AI features outright. Check both before you buy anything.
2. Finish the Business Profile and keep it current. Google names it on both guidance pages, it costs an afternoon, and it is the input you control most directly. Our guide to adding your business to Google Maps has the setup and verification steps.
3. Make your details identical everywhere you are listed. AI Mode reaches Maps and the Knowledge Graph as well as the open web, so an old phone number on a directory is a number Google has seen. Dull work, and the highest-yield hour on this list.
4. Write the page that answers a branch of the question. The questions your customers ask in the first ten minutes of an appointment are the pages: what a symptom usually means, what the options cost, what you do that the place down the road does not.

There is nothing to install. Google's generative AI guide is blunt about it: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." Any quote you are handed for an llms.txt file or an AI-flavoured rewrite of your service pages is a quote for work Google says it ignores.

Reviews sit underneath all of it, and the way to get them is to ask every customer rather than the ones you expect to be pleased. Google's contribution policies bar merchants from offering "payment, discounts, free goods and/or services" in exchange for a review, and from moves that "Discourage or prohibit negative reviews, or selectively solicit positive reviews from customers". A filtered profile describes a business that does not exist, which is a poor thing to hand a model. How to get more Google reviews covers the timing and the follow-up.

What AI Mode does to your clicks

Google is the only party with the aggregate data, so this section quotes Google and stops. On 6 August 2025, Liz Reid, VP and Head of Google Search, published the company's line: "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year", along with the concession underneath that average, that "user trends are shifting traffic to different sites, resulting in decreased traffic to some sites and increased traffic to others" (AI in Search is driving more queries and higher quality clicks). Google has published nothing about clicks from AI Mode specifically. Read the rest for what it is: a company describing its own product in aggregate with no method attached, not a finding about your website. The figures circulating elsewhere are not Google's and do not agree with each other.

How to check what AI Mode says about your business

Search Console covers the website. A manual check covers the business itself.

Your website. Search Console's generative AI performance report "shows data about how your site performs in generative AI features on Google Search", counting impressions, which Google defines as "how many times links to your site were shown to a user in a generative AI feature on Google Search". As of 31 August 2026 Google has "rolled out these insights to all websites worldwide". Two limits to know before you open it: the report covers AI Overviews and AI Mode together with no way to separate them, and it excludes anything a person saw inside a Search Labs experiment. An empty report usually means too few impressions, or a site excluded from generative AI features.

Your business. Search Console reports on your site, not on your Business Profile, and an impression count says nothing about whether the answer named you, described you correctly or sent the customer to a rival. For that you run the questions your customers run, from the area they are in, and write down what comes back. How to check whether AI assistants recommend your business has the prompts, the sign-out settings and the log, including how to run AI Mode cleanly.

Do the other assistants while you are there, because they answer local questions from different sources and cite different pages. Does ChatGPT recommend businesses like yours covers where each one gets its local answer. Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were. It does not sample AI Mode, so for Google's own surfaces the hand check and Search Console are what you have.

Frequently asked questions

Is Google AI Mode available in Australia and New Zealand?

Yes. On 6 September 2026, Google's AI Mode help page listed Australia and New Zealand, along with the United Kingdom, the United States, Canada and Ireland. Google announced the expansion to "over 180 new countries and territories in English" on 21 August 2025. It maintains the list and changes it, so check the page rather than this sentence.

Do I need schema markup or an llms.txt file for AI Mode?

No. Google says you don't need new machine readable files, AI text files, markup or Markdown to appear in Google Search including its generative AI capabilities, "as Google Search itself doesn't use them", and that structured data "isn't required for generative AI search". Keep your structured data for rich results in ordinary Search. That is what earns its place.

Can I see AI Mode traffic on its own in Search Console?

No. The generative AI performance report combines AI Overviews and AI Mode and does not split them, and both are also counted inside the ordinary Performance report under the Web search type. There is no AI Mode filter.

How many reviews do I need to be named in AI Mode?

Google has never published a number, for AI Mode or for local results generally. It documents reviews as an input to local ranking, in the words "More reviews and positive ratings can help your business's local ranking", and it has documented nothing about how reviews are used inside an AI Mode answer. Anyone quoting you a threshold invented it.

To monitor AI search visibility, run a fixed set of prompts through the same assistants on the same day each month, signed out, with your suburb written into the question, and record whether you were named, which rivals were named and what the answer cited. The measurement is the log, not any one answer. A screenshot of one good reply proves about as much as one good day of trading.

For the wider measurement, see how AI search decides which local businesses get named. The guide to does ChatGPT recommend businesses like yours covers where each assistant's local answers come from and what the vendors document.

Five shortlists from eight prompts

I ran the eight prompts below through Perplexity, signed out and with no account attached, on the morning of 6 September 2026. Eight physiotherapy prompts, one Sydney suburb, six minutes end to end. The clinics are not customers of ours and did not ask to be in a blog post, so they are not named here.

"Best physio in [suburb]" came back with ten clinics and their star ratings. "Top rated physio near [the local shopping centre]" came back with a different set. One clinic appeared in both. Nine did not.

The Saturday-opening prompt returned roughly the first list again, with nothing to suggest the constraint had filtered anybody out. "Who do I see for shoulder pain in [suburb]" did not lead with a physiotherapist at all: it told me to start with a GP, then listed orthopaedic surgeons. "Physio in [suburb] that bulk bills" produced a third set headed by a medical centre and a clinic whose trading name contains the words "bulk billing", and none of the highly rated clinics from the first prompt survived into it.

Then two prompts about one named clinic. "Is [clinic] any good" answered warmly and in detail, naming individual practitioners, out of a third-party aggregator page rather than the clinic's own Google profile. "What do people say about [clinic]", asked two minutes later, said user reviews were not in the results and answered off the clinic's own website.

Five shortlists from eight prompts. One answer tells you what one phrasing returned once. It does not tell you whether you are a candidate. For how widely this is happening rather than what happened here, AI search statistics carries a source on every figure.

Why the answer moves between runs

Four things change the answer, and each one is documented by the company that built the assistant.

Memory rewrites the question before it is asked. OpenAI's help centre is direct about it: "If memory is enabled, ChatGPT may use relevant saved memories when rewriting a search query", and its own example is a user who has mentioned being vegan and living in San Francisco getting a search for "good vegan restaurants San Francisco" (Searching the web with ChatGPT). All five surfaces have a version of it. Gemini personalises "by referencing your chats in Gemini Apps Activity", Perplexity's memory article says it may draw on stored memories and past searches, and AI Mode's Personal Intelligence "references previous searches and activity saved in your Search Services History". Test inside your logged-in account and you are measuring your account, not what a stranger sees.

Location decides the shortlist. ChatGPT "may use an approximate location based on your IP address to provide relevant local results", and device location sharing "is optional and off by default". Google is blunter about Gemini: "Location data is always collected if you use Gemini Apps so that they can provide you with a response that is relevant to your query", and by default that means the general area from your IP address or the Home or Work addresses in your Google Account. Google Search estimates it from your device, your saved addresses, your past activity and your IP (manage your location on Google). Run the test on the shop wi-fi and you have measured your own front door, which is the one place your customers are not standing when they ask.

Phrasing is not one thing. Google says AI Mode "uses a 'query fan-out' technique, dividing your question into subtopics and searching for each one simultaneously across multiple data sources", and its advice to searchers on the same page is "Ask multiple versions of your question to get the best answers" (AI Mode in Google Search). When the company running the system tells people one phrasing is not enough, one phrasing is not a measurement either.

The model underneath changes too. A Perplexity Pro subscription "includes all our latest and most powerful AI models", and in AI Mode a signed-in user can pick Pro from the menu and get Gemini 3 Pro instead of the Fast model on the same question. None of this is deterministic. A re-ask can differ, so one miss is not a verdict and one hit is not a rank.

The eight prompts a customer would actually type

Write eight prompts once, then stop changing the sentences. Change the values inside them when a location or a service changes. Never change the shape, or you lose the ability to compare one month to the next. They fall into five kinds of question, and you want all five.

Plain discovery, with no business name in it. This is the one that tells you whether you are a candidate at all.

best {category} in {suburb}
top rated {category} near {landmark}

Constrained discovery, with the detail people care about: opening hours, parking, a health fund, a brand you stock, whether you see children. This tests whether anything on the open web says that about you.

{category} in {suburb} open on a {day}
{category} in {suburb} that {constraint}

Problem first, because customers rarely search the category name. They search the thing that is wrong.

who do I see for {problem you solve} in {suburb}

Named, which is what the assistant says when a friend has mentioned you and someone is checking.

is {your business} any good
what do people say about {your business}

Comparison, the one that decides a booking.

{your business} or {a rival you actually lose to}, for {the job}

These overlap with the shorter ten-minute test in does ChatGPT recommend businesses like yours on purpose. That post asks the question once. This one is built to be run again in October, and again in November, against the same eight sentences.

With more than one location, run the discovery prompts per location rather than once for the brand. A flagship being named tells you nothing about the suburban site, and the suburban site is usually where the gap is.

How to monitor AI search visibility across ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode

Same day each month, fresh chat for every prompt, and the location written into the wording every time even when the assistant already knows where you are: OpenAI's own fix for local results showing the wrong area is to "include your city, neighborhood, or postal code in your question". Set the assistants up first, because each vendor documents a different control.

Run the prompts this way.

1. Open a fresh chat or a new incognito thread for each prompt. No follow-ups in the same thread: the first answer is now context for the second.
2. Paste the prompt exactly as written in your list.
3. Fill in the seven columns below before you move on. Doing it afterwards from memory defeats the point.
4. Open the sources. All five assistants show what they read, and that list is the most useful thing on the screen.
5. Do not re-ask a prompt that missed until you like the answer. Log the miss.

The asking is quick. Every answer in my run landed in well under a minute. Writing down what came back takes longer, and it is the part that is worth anything.

The log: seven columns, every month

A spreadsheet, a notes file, anything you will still have in a year. Seven columns, the same ones as the shorter test in the ChatGPT piece so the two are comparable.

Coastline Physio and its two rivals are invented, so nobody reads those rows as a case study. The last column earns its place: OpenAI warns that "search results and citations can be incomplete, outdated, or incorrect", and an assistant reading a stale phone number off a directory is a booking you never hear about.

Three months in, the log answers questions a screenshot never can. Are you named more often than in July, or only on the named prompts where the customer already knew you existed? Is the same rival in every assistant, or only in Gemini? Is the same directory cited in six answers out of eight? That last one is the work list.

How often should you check AI visibility?

Monthly. The tool vendors writing about this recommend daily scans and weekly audits, which makes sense when you are tracking a national brand across hundreds of prompts and can average the noise away. With eight prompts and one business, daily checking measures the noise: my own run threw up five shortlists inside six minutes without a thing changing about the businesses in them. Pick a day, put it in the calendar, and treat a month where you did the run and learned nothing as a good month. The value is in the series.

What to do when the answer does not name you

A miss is a starting point, and the log usually tells you which of three things went wrong.

You were not in the sources. If the same roundup, directory or association page is cited every month and you are not on it, that is a job with a name and a deadline. It is the most common finding and the easiest to fix. In my run, the warmest answer about a named clinic came off an aggregator page the clinic almost certainly never thinks about.

Your profile did not answer the question. The constrained prompts are the diagnostic. If "physio in {suburb} that bulk bills" names three rivals and not you, and you do bulk bill, the problem is that nothing public says so. Finishing your Google Business Profile is the first hour of work, every time.

You had nothing worth reading. Google's local ranking guidance says "more reviews and positive ratings can help your business's local ranking", and its Grounding with Google Maps documentation says a place question is answered "based on Google user reviews and other Maps data". A profile with 30 one-line reviews from 2023 gives an assistant nothing to match against a specific question. Getting more Google reviews steadily, and replying to the ones you get, is the slow half and the half that compounds. Whether reviews move your local ranking is the longer version of that argument.

One thing not to do: filter who you ask. Google's contribution policies stop merchants who "discourage or prohibit negative reviews, or selectively solicit positive reviews from customers", and a screened profile does not describe your business anyway, so it gives an assistant less to work with rather than more. Ask everyone or ask nobody.

None of this buys a mention. OpenAI's own wording is "Placement is not guaranteed", and no vendor publishes a formula. What the work buys is candidacy, and the full list of what a local business can change is the parent piece to this one.

Where AI visibility tools fit

Tools do this at a scale a person cannot. Fifty prompts across twenty locations and five assistants is thousands of runs a month, and it has to be automated or it will not happen. What a tool cannot do is make the system underneath deterministic. It is still sampling, one sample at a time, and any product reporting a stable rank is presenting a sample as a fact. Read whatever you buy for how it describes its own sampling. With one or two locations, the log above is the same measurement done by hand.

Cloutly's AI search visibility asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and records whether each location was named and which rivals were named; it does not sample Copilot or Google AI Mode, and its results are a point-in-time sample of a non-deterministic system, one sample each, so a re-ask can differ.

FAQ

Should I test signed in or signed out? Signed out where the assistant allows it, because a signed-in account carries memory and saved locations into the query. OpenAI documents that web search works for people who are not signed in, and Google documents signed-out use of Gemini. Where you must be signed in, use the documented private mode: a Temporary Chat in ChatGPT or Gemini, incognito in Perplexity.

Why did the same prompt give me a different answer an hour later? Because these systems are not deterministic, and because the web underneath them changed. That is the reason for the log, and the reason for monthly rather than daily. If you got two answers, log both.

Is this the same thing as tracking my Google ranking? No. Your local ranking is one input the assistants read, not the output they produce, and an assistant can name a business that is nowhere near the map pack. Track both, separately.

NAP consistency means your business name, address and phone number read the same everywhere they are published: your Google Business Profile, your own website, your Facebook Page, Apple Maps, Bing and whatever directories hold a record of you. It is worth doing and worth doing early. It is not the ranking lever a decade of local SEO writing made it out to be, and knowing which of those two things you are buying decides how much of your week it deserves.

What is NAP consistency?

NAP stands for name, address and phone number. Add the website and you get NAP+W, which is the version most people mean now. A citation is any place online that publishes some or all of that set, whether or not it links to you.

NAP consistency is those values matching across all of them. Values, not characters. "Suite 4, 12 Marine Parade" and "12 Marine Parade, Ste 4" are the same address. "(07) 3040 1188" and "07 3040 1188" are the same phone number. What Google's own guidelines ask for is that you "Represent your business as it's consistently represented and recognized in the real world across signage, stationery, and other branding" and "Make sure your address and/or service area is accurate and precise" (Guidelines for representing your business on Google). That is a rule about the facts, not about punctuation. Chasing commas is not the job. Finding the records that are genuinely wrong is.

What NAP consistency does for local rankings in 2026

Whitespark publishes the Local Search Ranking Factors study, and the 2026 edition came out on 6 November 2025. Forty-seven local search practitioners are invited to score 187 factors in a two-hour survey, across four axes: local pack and Maps, local organic, conversion, and AI search visibility, which is new this year. Here is where the consistency factors landed, out of 187.

The site names in brackets are Whitespark's own factor labels, not a shopping list.

Read it as weighted opinion rather than a controlled test, because that is what it is. The shape is still clear. One consistency factor sits in the top twenty and the rest sit in the back half, below the primary category, the searcher's proximity, whether the address is in the city searched, whether you are open at the moment of the search, and your Google rating. The one that ranks is the one you own outright: your own website agreeing with your own Google profile.

Google has not published a ranking factor called NAP. Its account of local ranking is relevance, distance and prominence, and the guidelines are a rule against lying, not an instruction to standardise 200 directory records.

Consistency is hygiene. Being consistent will not push you up the map pack. Being inconsistent can hold you back, and it costs you customers either way. A wrong number is a lost call, and somebody who rings a disconnected line does not ring back and try the other listing.

It also decides who gets the last word on your profile. Google "gathers info from different sources, like user reports and licensed content", shows what it has changed in blue in the profile editor, and states plainly that "You can't manage all Google updates through your Business Profile" (Understand Google updates on your Business Profile). When an old directory keeps publishing your 2019 number, that directory is what keeps correcting you. Assistants behave the same way, a repair job covered in when AI gets your business details wrong.

What are NAP citations?

Citations come in two shapes, and they are no longer equally useful.

A structured citation is a directory record with fields: Yelp, Yellow Pages, an industry association's member list, an aggregator's database. An unstructured citation is a mention in prose, such as a local news story, a supplier's stockist page or a sponsor list on a sports club site.

The same study scores AI search visibility separately, and three of its top five are citation factors: presence on expert-curated best-of lists at number one, prominence on key industry-relevant domains at three, and the quality and authority of unstructured citations at four. Note which citations those are. Mentions on domains that matter in your industry, not another row in a directory nobody opens.

Whitespark's own reading of it: "Citations as a local search strategy have been decreasing in importance for the past decade, but AI SEO is breathing new life into them."

That matters because people now ask assistants for recommendations. BrightLocal's Local Consumer Review Survey on AI and trust, published on 10 March 2026 from 1,002 US adults, found 45% had used an AI assistant to get a local business recommendation, up from 6% the year before.

The old game was 300 directory submissions. The current one is correcting the records you already have.

Where NAP inconsistencies come from

Nobody sets out to publish two phone numbers. Inconsistency arrives as a side effect of something else.

A rebrand. The new name goes on the signage, the website and the Google profile. It does not go on the seventeen directory records an agency created in 2019, and nobody has the logins. Google also treats a name change differently depending on how big it is, so read how to edit your Google Business Profile before you type the new one in.

A move. The address changes on the profile in five minutes and stays in the website footer, the contact page, the booking confirmation email, the schema block and every directory. A move invalidates every published record at once, which makes it the largest single source of drift there is.

Suite and unit numbers. Google's instruction is conditional and specific: if you must give a mailbox or suite number, the physical address goes in "Address Line 1" and the mailbox or suite number in "Address Line 2". Most directories have one address field, so the suite number gets appended, prefixed or dropped. That is formatting rather than fact, and usually harmless. A virtual office is neither. Google says that if you rent a mailing address but do not operate out of it, "that location isn't eligible for a Business Profile", and a desk in a co-working space only qualifies with signage and staffed hours.

Tracking phone numbers. Dynamic number insertion usually rewrites the number in the visitor's browser after the page loads, leaving your served HTML alone. Confirm that with your vendor rather than assume it. Typing a tracking number into a directory listing is the version that definitely breaks things, and Google's guidelines rule it out: "Use a local phone number instead of a central call center helpline number whenever possible", and do not "provide phone numbers or URLs that redirect or 'refer' users to landing pages or phone numbers other than those of the actual business, including pages created on social media sites." The primary number has to ring at that location. Extras are allowed: "Additional phone numbers can be used on Google Business Profile websites and other local surfaces."

Franchise and multi-site websites. Tallowwood Dental has six practices and one website template whose footer carries head office's number, so five location pages publish a number that does not ring at that practice while all six Google profiles carry the right one. The website is the highest-weighted consistency factor in the table above, and a template is how it goes wrong at scale. Franchisees ordering their own signage and answering service produce the same result from the other direction.

How to audit NAP consistency in an afternoon

1. Write the master record. One row per location: trading name exactly as it appears on the signage, address line 1, address line 2, suburb, state and postcode, the phone number that rings there, and the page on your site that represents it. This spreadsheet, not any profile, is what you correct back to. If you cannot fill it in without asking someone, that is the first finding.
2. Search the way a machine sees you. In Google, search your phone number in quotation marks, then your old phone number in quotation marks, then your business name plus the suburb. The old-number search finds the problems, because it surfaces every record still publishing a value you retired.
3. Open the eight places below and compare each field to the record. Do not fix anything yet. Fixing while auditing is how you lose track of what you have covered.
4. Sort the findings into what you control and what you have to ask for. The second list takes weeks, so start it first.

Compare name, address, phone and website on every row, plus hours wherever the source carries them.

That last row is vague on purpose. The aggregators Whitespark names, Infogroup, Localeze and Foursquare, are the US set; the Australian, New Zealand and UK equivalents are different, and the old-number search in step 2 is the reliable way to find yours.

Anything beyond those eight is optional. A directory with no traffic and no authority in your industry is not worth an afternoon, and a paid tool reporting 47 "inconsistencies" is mostly reporting punctuation.

How to fix each source

Order matters, because directories and assistants read the first two.

Google Business Profile first. Change the field and save. Hours, phone and website usually reappear within the hour. Name, address and category can put you back through verification, so change them one at a time and only when the public record supports the new value. The sequence is in how to edit your Google Business Profile.

Your own website second. The footer is the easy part. The rest is the contact page, every location page, the LocalBusiness schema, the booking widget, any PDF you still link to and the email signature template. Search your own site for the old number instead of trusting your memory of it.

Facebook, Apple, Bing and Yelp third. Facebook is edited on the Page itself, in its own settings. Apple is where most guides are now wrong: Apple Business Connect became Apple Business in April 2026 and businessconnect.apple.com redirects to business.apple.com, so a 2024 walkthrough describes screens that are gone. Adding your business to Apple Maps has the current path. Bing's old bingplaces.com redirects to bing.com/forbusiness, and Yelp needs claiming before you can edit it.

A directory you cannot log into. Most have a "suggest an edit" control that needs no account. The rest have a support address. Keep a list of the ones that never respond, because that list is where a resurfacing wrong number comes from.

A location that has closed. Do not delete the profile. Deleting it takes the reviews and leaves the listing on Maps. Mark it closed, as set out in how to remove a business from Google.

What changes at ten locations

At one site this is an afternoon and then fifteen minutes a quarter. At ten it is a different job, because the failure is not one wrong record. It is one wrong record per site, each appearing at a different time.

The rule that holds at any size: decide the value once, hold it somewhere that is not a profile, and publish from there. Editing profiles by hand makes the profile the record, and a record you cannot read in one place is a record nobody checks. If two sites share a name, multiple Google locations with the same name covers keeping them distinct.

Cloutly's listings management is built on that shape: one record per location covering name, phone, website, description, address and hours, published to Google Business Profile and to the Facebook Page only after you approve a preview of the exact change, then checked weekly, read-only, against Google's public view, Yelp, Facebook and TripAdvisor. The wider picture for a network is on the multi-location page.

How often to check

Quarterly suits a single site and monthly suits a network, but a schedule is the weaker instinct. Check on events instead: a move, a rebrand, a new phone system or answering service, a website rebuild, a franchisee joining or leaving, a location closing. Those causes account for nearly all of it, and every one is a date you know about in advance.

For the layer above this one, landing in the Google map pack has the factors that do move you, and adding your business to Google Maps covers getting the profile right in the first place.

FAQ

What does NAP stand for in SEO?

Name, address and phone number. NAP+W adds the website. Both terms come from local SEO rather than from Google, whose own account of local ranking uses relevance, distance and prominence.

What is a NAP citation?

Any online mention of your business name, address or phone number. A structured citation is a directory record with fields. An unstructured citation is a mention in prose, like a news article or a sponsor page. Neither has to link to you to count.

How do I check my NAP consistency?

Write down the correct details for each location, then compare them against your Google Business Profile, your own website, Facebook, Apple Maps, Bing, Yelp, your main industry directory and the data aggregators. Searching your old phone number in quotation marks finds forgotten records faster than any tool.

Does NAP consistency still matter in 2026?

As hygiene, yes. In the 2026 Whitespark study one consistency factor ranks 15th of 187 for the local pack and the rest sit in the back half, so it is not what moves you up. It is what stops a customer ringing a dead number, stops Google correcting your profile from an old source, and stops an assistant repeating a detail you retired two years ago.

Review schema is structured data that tells a search engine what a rating on your page means: who wrote it, what it is about, and how many people it represents. It will not put star ratings under your own business in Google's results. Google stopped showing those in September 2019 for any business that hosts reviews about itself, and has not brought them back.

That fact is missing from most of the pages ranking for this term, which is why people add the markup, wait a month, and decide they did it wrong. They did it right. The feature was retired.

Everything below was checked on 6 September 2026 against Google's own documentation. Both code blocks were run through the Schema Markup Validator and the Rich Results Test, and the screenshots are what those tools gave back.

What is review schema?

Review schema is a pair of schema.org types you add to a page as JSON-LD. Review describes one person's verdict on one thing, with an author and a rating. AggregateRating describes the average of many, with a rating value and a count. Both attach to the thing they are about, which schema.org calls itemReviewed.

Two terms get used interchangeably and should not be. Review schema markup is the code: a <script type="application/ld+json"> block in your page's HTML. A review snippet is the search feature that code can make you eligible for, the row of gold stars under a result. Eligible is the operative word, and most of this piece is about that gap.

Will review schema put stars under my business in Google?

No, not for reviews you host about yourself. On 16 September 2019 Google published Making Review Rich Results more helpful, which did two things at once. It cut the list of schema types that can trigger a review rich result, and it banned what it called self-serving reviews.

Google's own definition, from that post and in the screenshot at the top of this page:

We call reviews "self-serving" when a review about entity A is placed on the website of entity A - either directly in their markup or via an embedded third-party widget.

The rule survives in the current review snippet documentation, worded for developers: if the entity being reviewed controls the reviews about itself, its pages using LocalBusiness or any other type of Organization structured data are ineligible for the star review feature. Google's example is a Google Business reviews or Facebook reviews widget, which covers most of what a small business actually installs.

That documentation now qualifies two entries in its own list of supported features. Local business works "only for sites that capture reviews about other local businesses", and Organization carries the same caveat. A directory that reviews plumbers is eligible. A plumber is not.

The 2019 post answered the obvious follow-ups, and the answers still hold:

• You do not have to remove markup you already have. Google will not show a snippet for it. That is all.
• You will not get a manual action for self-serving review markup on its own.
• Your Google Business Profile is unaffected. The change applies to organic results, not the map pack.
• It covers AggregateRating as well as individual Review items.
• Sites that gather reviews about other organisations were not touched.

The same post made name a required property, so every marked-up review has to say what it is a review of.

What review schema is still good for

Three things, and stars under your own homepage is not one of them.

First, eligibility for the types that do still show snippets. The current feature list is Book, Course list, Event, Local business, Movie, Product, Recipe and Software App, and Google separately supports reviews on CreativeWorkSeason, CreativeWorkSeries, Episode, Game, MediaObject, MusicPlaylist, MusicRecording and Organization. If you sell a product or run a course, reviews of that product or course are eligible on your own site. It is the business-level rating that is not.

Two traps live in that list. Service is absent from it and from the list of valid itemReviewed types, so a rating hung off a Service will not produce a snippet however tidy the code is. HowTo goes the other way: still a valid itemReviewed type in the property definitions, with no entry at all in Google's structured data feature gallery. Google's pages have drifted apart, and the gallery names six types where the reference page names eight. Work from the reference page.

Second, telling a machine what the rating means. A visitor reads "4.8 from 213 reviews" and understands it instantly. A parser sees three numbers next to some text. Markup takes the guess out.

Third, do not assume it does more than that. Nobody at Google, OpenAI or Perplexity publishes review schema as an input to an AI answer, and anyone quoting you a weighting has invented it. The honest version is narrower: the rating is on the page in a form that does not depend on a JavaScript widget rendering first. Our post on what moves AI answers for a local business covers the inputs vendors do document, and whether reviews change what AI says covers the one case a vendor puts in writing.

The rules, in Google's words

These come from the review snippet guidelines and the general structured data guidelines. They are short, and people break them anyway.

1. The review has to be on the page. "Don't mark up content that is not visible to readers of the page." A 4.8 that only exists in your CMS is a violation, not a shortcut, and users have to be able to see any aggregate rating you mark up.
2. It has to be reachable. Review content must be readily available to users from the marked-up page, and you cannot block that page from Googlebot with robots.txt, noindex or any other access control.
3. It has to be about one specific item, not a category or a list of items.
4. Ratings must come from users. Google's wording for local business markup is blunt: do not rely on human editors to create, curate or compile ratings information.
5. Do not aggregate reviews or ratings from other websites. Blending your Google, Facebook and Yelp ratings into one number breaks the rule directly, and it is one of the most common things review widgets get asked to do.
6. No fake reviews, and no undisclosed incentivised ones. A review written in exchange for money, a discount or a free product has to disclose that clearly and prominently, or it stays out of your markup.
7. The author's name has to be a real name, under 100 characters. Google's example of an invalid one is "50% off until Saturday".

A review schema markup example you can copy

Two blocks for a fictional practice, Coastline Physio in Newcastle. Change the values, keep the shape.

The first is the one most local businesses want. It is accurate, it is allowed, and it will not produce stars. Put it on a page where that 4.8 and that count are visible.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Coastline Physio",
  "url": "https://coastlinephysio.example/",
  "telephone": "+61 2 5550 0100",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "14 Marine Parade",
    "addressLocality": "Newcastle",
    "addressRegion": "NSW",
    "postalCode": "2300",
    "addressCountry": "AU"
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.8",
    "reviewCount": "213",
    "bestRating": "5",
    "worstRating": "1"
  }
}
</script>

When aggregateRating is nested inside the thing it rates, you leave itemReviewed out. The parent's name is the name of the item being reviewed, which is what the 2019 change made compulsory.

The second is the eligible case. Coastline Physio also sells a six-week running rehab programme, on its own page, and asks the people who finish it what they thought. Those reviews are about that programme. Product is on the feature list, so this one is eligible.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Six-Week Running Rehab Programme",
  "description": "A supervised six-week programme for runners returning from injury.",
  "image": "https://coastlinephysio.example/img/running-rehab.jpg",
  "brand": { "@type": "Brand", "name": "Coastline Physio" },
  "offers": {
    "@type": "Offer",
    "price": "540.00",
    "priceCurrency": "AUD",
    "availability": "https://schema.org/InStock",
    "url": "https://coastlinephysio.example/running-rehab"
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.7",
    "reviewCount": "38",
    "bestRating": "5",
    "worstRating": "1"
  },
  "review": [{
    "@type": "Review",
    "datePublished": "2026-08-19",
    "author": { "@type": "Person", "name": "Emma T" },
    "reviewRating": {
      "@type": "Rating",
      "ratingValue": "5",
      "bestRating": "5",
      "worstRating": "1"
    },
    "reviewBody": "Six weeks and I ran the whole loop without stopping. The load plan was the part I would never have got right on my own."
  }]
}
</script>

The catch is the one nobody writes down. Those 38 ratings have to be ratings of the programme, from people who did the programme, shown on that page. Taking your business's Google rating, renaming it a product rating and hoping the type change earns you stars fails two guidelines at once: the rating is not about that item, and the markup is not a true representation of the page.

Aggregate rating schema: the fields that matter

Schema.org itself marks nothing as required, so "required" here means required by Google for a review snippet.

How to test review schema markup

Use both tools. They answer different questions, and the difference is where most of the confusion lives.

The Schema Markup Validator answers "is this valid schema.org". Run by schema.org, descended from Google's retired Structured Data Testing Tool, it parses your JSON-LD whether or not a search feature exists for it. Use it to catch syntax: a trailing comma, a stray quote, a property hung on a type that does not take it. Both blocks above come back with no errors and no warnings.

The Rich Results Test answers a narrower question: which of Google's own rich result types this markup could generate. The result catches people out because it is the opposite of what most guides say. Paste that Coastline Physio LocalBusiness block in and the tool reports three valid items, one of them a review snippet. It says nothing about the self-serving rule.

That is neither a bug nor permission. Google splits its requirements into technical guidelines, which the tools check, and quality guidelines, which it says are "not easily testable using an automated tool". The self-serving rule sits on the quality side, and a tool reading a snippet of code cannot know whether the business in the markup owns the site it sits on. So the test goes green, the stars never appear, and the owner spends a fortnight hunting for a bug in their JSON. There is no bug.

Two smaller things it flagged. On the LocalBusiness block, two missing optional fields, priceRange and image: worth filling in, not errors. On the Product block it failed outright at first, "Missing field 'image'", because Product markup is also measured against Google's merchant listing requirements, where an image is required. That is why the block above carries one.

URL Inspection in Search Console answers a third question: what Google saw when it crawled the live page. Run it after you deploy. A widget that injects markup client side can pass both tests on your machine and give a crawler nothing.

Do you need a schema markup generator?

For one block, no. Copy the example above, change the values, and you are done in five minutes. A schema markup generator earns its place when you want a type you have not written before, and Google's Structured Data Markup Helper will walk you through tagging a page by clicking on it.

The risk with any generator is that it will happily produce a rating for content that is not on your page, because it cannot see what your page says. That is the guideline you are most likely to break, and the one Google can act on.

If the rating already comes from a widget on the page, let the widget emit the markup rather than hand-write a second copy that drifts out of date. Cloutly's review widgets can inject AggregateRating markup so the rating is on the page in a form search engines can read.

Common review schema mistakes

• Two AggregateRating blocks on one page. An SEO plugin adds one, a review widget adds another, and they disagree. Pick one source.
• A rating in the markup that is not on the page. The most common violation, and the easiest to fix.
• A blended rating from several review sites. Against the guidelines. Show your Google and Facebook ratings as separate figures instead.
• Service instead of Product. Not on the feature list, not a valid itemReviewed type, no snippet.
• Expecting stars from a homepage Organization block. The 2019 rule again, and the most common reason people think their markup failed.
• Reading a green Rich Results Test as a promise. It tests the code, not the policy.
• Client-side injection nothing can see. Check with URL Inspection as well as a local test.

Once the markup is right, the next question is what to put on the page in the first place. Our guide to embedding Google reviews on your website covers the four routes and the terms attached to each, including what the Places API lets you store.

Frequently asked questions

Does review schema still work in 2026? Yes, for the types Google shows snippets for. It does not work for a business rating a business hosts about itself, and has not since September 2019.

Can a local business get star ratings in Google search results at all? Not under your own organic result, from reviews you host. A third-party profile Google surfaces can still carry a rating, and your Google Business Profile keeps its own rating in the map pack, a separate system the 2019 change did not touch.

Should I remove the review markup I already have? No. Google said plainly that you do not need to remove it and that it will not trigger a manual action on its own. It is accurate information about your page. It just will not earn a snippet.

Does review schema help with AI search? It puts the rating on the page in machine-readable form, which is worth doing. No AI vendor documents review schema as a ranking input, so treat anyone selling it as one with suspicion.

If ChatGPT hands out a phone number you gave up two years ago, or Gemini says you close at 4pm when you close at 6, or an AI Overview has you permanently closed, the assistant did not invent any of it. It retrieved a wrong detail from somewhere that still publishes it: your Google Business Profile, an old page on your own site, or a directory nobody has logged into since 2021. You cannot edit the answer. You fix the source and wait for the next retrieval to pick it up.

Fix wrong details before any other work in how AI search decides which local businesses get named. Visibility you have earned is worth nothing if the answer sends the customer to a disconnected line.

Why AI assistants get your business details wrong

Assistants build a local answer at the moment you ask, out of whatever they can retrieve about you. OpenAI says so in its own help centre: "Search results and citations can be incomplete, outdated, or incorrect", and its advice is to open a cited source and "review when it was published or updated" (Searching the web with ChatGPT). Google documents the same shape from the other end: Gemini's Maps grounding step "queries Google Maps for relevant information (e.g., places, reviews, photos, addresses, opening hours)" (Grounding with Google Maps).

So the mistake has an address. It is nearly always your Google Business Profile, which feeds Google's own surfaces and gets copied into half the directories that feed everyone else's; your own website, usually a contact page that was right when the site was built; a listing you have never logged into; or a page about you that you do not control at all, like a booking aggregator or an old press release.

Stale data outlives corrections because those sources copy each other. Fix the profile, leave the aggregator, and the aggregator keeps republishing the old number for the assistant to find.

Is it an AI hallucination, or stale data?

Almost always stale data. "AI hallucination" means a model generating something it never retrieved, and that does happen. Opening hours are the wrong place to go looking for it. Ask the assistant where the hours came from, then open that page. Nine times out of ten the page says exactly what the assistant said, and has done since your last renovation.

The distinction matters because the two fixes point in opposite directions. A fabrication is a vendor problem you can report and cannot repair. Stale data is a listings problem you can repair this afternoon.

When AI says your business is permanently closed

This one is not the assistant's doing at all. Anyone signed in to a Google account can propose that your shop has shut, and Google publishes the route: open Google Maps, search for the place, select Suggest an edit and then Place closed or not here, select Permanently closed and Submit (Mark a place reopened or closed on Google Maps). A competitor can do it. So can a customer who turned up on a public holiday.

Once Google carries the status, the assistants inherit it, because it travels with the profile into every surface that reads the profile, including the AI Overview above your own category search.

To undo it as the owner: go to your Business Profile, select Edit profile and then Hours, select Edit and then Open with main hours, tick the days you trade, set the times and Save. For several sites at once, sign in to Business Profile Manager, select the locations, then Actions, Open and Mark as open (Mark your business as reopened). Google's caveat on public edits applies to yours too: "We review your edits, so it can take some time for your changes to show on the map."

How to find where the wrong detail came from

Use these three checks.

1. Read the citation the answer already gave you. ChatGPT shows them inline ("Responses that use web search may include citations. Select a citation to open its source") and has a Sources control. Perplexity numbers every answer. AI Overviews carry supporting links, under Google's own instruction to searchers to "Always check important info in more than one place" (AI Overviews in Google Search). The culprit is usually sitting there in plain sight.
2. Ask the assistant where the detail came from. "Where did you get that phone number for Coastline Physio?" often names a page. Treat it as a lead rather than proof, because a re-ask can produce a different explanation. OpenAI's own documented move is more useful: when a result looks wrong, ask ChatGPT "to search again using a specific source, date, or location".
3. Search for the wrong value itself. Put the old phone number in quotation marks into Google and look at everywhere it still lives. Do the same with the old suite number. That result list is your work list, and it beats anything an assistant tells you about its own reasoning.

Write down what was wrong, which assistant said it, what it cited and the date. You need that record to tell whether the fix worked.

The fix order: how to correct business information in AI search

Work outward from the source with the most reach, so the later fixes have less left to correct.

1. Google Business Profile first

Google says plainly that the profile feeds its AI answers: "Using products like Merchant Center (such as Merchant Center feeds) and Google Business Profiles can help your products and services to be visible in both AI responses and other Google Search results" (Google's guide to optimizing for generative AI features). It also feeds Gemini through Maps grounding, and it is the record most directories copy.

Go to your Business Profile, select Edit profile, change the field and Save. On mobile, open the Google Maps app, tap Business at the bottom right, then Edit profile. Then make the same change everywhere else it is published, which is NAP consistency.

Nothing goes live the moment you save it. "We review your changes before we update them live on your profile", and some businesses have to verify the profile before they can change anything on it at all (Edit your Business Profile).

If the profile itself is a mess, or there are two of them, start at how to add your business to Google Maps rather than editing around the problem.

2. Your own website second

This is the page an assistant can read without a login, and the one you can change in five minutes. Contact page, footer, every location page, the structured data. Put the phone number, address and hours in text, not in an image or a script that only paints after the page loads.

Then go hunting for the old value on your own domain: old landing pages, a PDF price list, the "our new premises" post from the move. If it is on your site it is fair game for retrieval, and no assistant knows which of your pages you consider current.

One caveat. ChatGPT will not use a page it cannot crawl: OpenAI's eligibility rule is to "allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses". Blocking the crawler does not take the wrong answer away. It takes away the page that would have corrected it.

3. Then the directories the answer cited

Do the ones the citation named first: those are demonstrably being read. Then do these two anyway, because they sit under map apps and assistants plenty of people use and almost nobody in a service business logs into.

Apple Business (formerly Apple Business Connect). Sign in at business.apple.com, choose Brands and then Locations, and select the location to edit it. Apple is blunt about why: "It's important to keep your location data up to date in Apple Business to ensure customers discover your business on Maps and know when to visit your business" (Maintain location information accuracy). The walk-through is in how to add your business to Apple Maps.

Bing Places for Business. The old bingplaces.com address now redirects to bing.com/forbusiness, where the offer is still "Claim your business listing on Bing Places for Business" behind Get started and Sign in. Bing's index sits under Copilot, so a wrong record here is a wrong answer there.

How long a correction takes to reach an AI answer

Nobody publishes it, and anyone who gives you a number is guessing.

The vendors document the first leg only. Google reviews Business Profile changes before they go live and puts no clock on it. Google reviews suggested Maps edits and says only that "it can take some time for your changes to show on the map". Apple says a contested location name change takes "up to five working days". That is the whole of the published record.

No vendor documents the second leg, from a corrected listing to a corrected answer, and it is not really one lag either: the answer is rebuilt from a fresh retrieval every time somebody asks, with caches, crawl schedules and the assistant's choice of source in between.

So fix the source, re-ask a week later, re-ask a month later, and read the trend rather than the single run. Signed out, fresh chat, so you are measuring what a stranger sees.

How to report a wrong answer in each assistant

Every assistant has a feedback control. None of them is a correction channel, and no vendor promises that a report changes a future answer. Use them anyway, because they cost nothing, and because in ChatGPT's case there is a documented consequence.

OpenAI's is the one lever with a mechanism behind it (Reporting content in ChatGPT and OpenAI platforms), and it points at the source domain rather than at the answer. If an aggregator with a dead phone number keeps getting cited about you, reporting that domain is a better move than thumbing down the reply.

One wrong location poisons the brand answer

With several sites, the failure is rarely spread evenly. Say Coastline Physio has four clinics and the Marrickville site moved in 2024, but an old aggregator page still carries the previous address. Ask "where is Coastline Physio" and a customer can be handed that address for a brand whose other three are perfectly correct.

Check per location and check the brand: one prompt naming each suburb, one naming only the business. They fail differently, and the brand-level prompt is the one that catches a single bad record contaminating everything. Clear duplicate profiles before you audit anything else, though, because two profiles for one shop give the assistant two versions of you and it is not obliged to pick the right one. Google Business Profile: multiple locations with the same name covers merging one.

Do you need an AI search visibility checker?

A checker identifies a wrong answer. You still repair it on Google, on your website and on the directories, not inside a dashboard.

Use a checker to keep the record: the same prompts at the same interval, with what each assistant said and what it cited, so you can tell a fix from a fluke. Judge one on whether it stores the answer text and the citation and keeps each location separate. A score on its own cannot be audited. You can also do the whole thing by hand, for nothing: how to monitor AI search visibility has the prompt set and the seven-column log.

A monthly fact-check that takes fifteen minutes

1. Open a fresh, signed-out chat in ChatGPT, Gemini and Perplexity. Signed in, you are measuring your own account, not what a customer sees.
2. Ask one question per location that a customer would actually ask: "what time does Coastline Physio in Marrickville close", "what is the phone number for Coastline Physio Marrickville".
3. Ask one brand-level question with no suburb in it.
4. Check every phone number, address and set of hours against your own record, not against your memory.
5. For each mistake, open the citation and note the domain.
6. Fix Google first, your site second, the cited directory third. Log the date, the assistant, the wrong value and the source, so next month you are comparing rather than starting over.

Cloutly's AI search visibility sweep asks ChatGPT, Gemini and Perplexity discovery questions such as "best {category} in {locality}" once a month and flags a wrong phone number in an assistant's answer ("{Assistant} gives out the wrong phone number for {venue}"), reported as a point-in-time sample of a non-deterministic system: one sample each, and a re-ask can differ. It checks the phone number and nothing else; website, address and hours drift is listings management, which watches Google's public view, Yelp, Facebook and TripAdvisor against the record you set.

FAQ

ChatGPT has the wrong information about my business. Can I edit it directly?

No. There is no business account inside ChatGPT and no field to correct. You can report the answer, and OpenAI says a reported domain may draw filters or other mitigations that stop ChatGPT relying on it in future. The repair itself happens on the source the answer cited.

Will fixing my Google Business Profile fix every assistant?

It fixes the Google surfaces and, over time, most of what copies from Google. It does not reach a directory that holds its own record, or a page about you on somebody else's site. Those get corrected where they live.

A competitor marked us permanently closed. Can I stop it happening again?

Not preventatively. Google accepts closure suggestions from the public and reviews them. What you can do is notice quickly, which is the argument for a monthly check rather than an annual one, then reopen the profile through Edit profile and Hours.

Should I block AI crawlers to stop the wrong information spreading?

No. Blocking a crawler removes your corrected page from the pool the assistant draws on, and leaves the third-party listing with the old number as the only thing it can find. OpenAI's own eligibility requirement is that OAI-Searchbot can crawl the site.

Do Google reviews affect ranking? Yes, and Google says so in writing. Its help page on local ranking puts reviews inside prominence, one of the three factors behind local results, and states that "More reviews and positive ratings can help your business's local ranking." What Google has never published is how much they help, how many you need, or how fast it happens.

Almost every page on this question asserts the effect and stops there. The useful version separates what Google has published from what experienced local SEOs believe but cannot prove, and then from the thing reviews do after the ranking is settled, which is where most of their value sits.

What is the Google Map Pack?

The Google map pack is the block of three business listings, shown under a small map, that Google puts above the ordinary web results for a local search. Google calls this section the local results. Marketers call it the map pack or the 3-pack. It is ranked separately from the blue links beneath it, on a different set of signals, so a business can own the pack and be nowhere on page one otherwise.

What Google says decides local ranking

Google publishes one page on this: Tips to improve your local ranking on Google. It names three factors and gives each of them a sentence.

Relevance is how well a Business Profile matches what someone is searching for.

Distance refers to how far each business is from the customer who's searching.

Prominence means how well-known a business is. Prominent places are more likely to show up in search results.

Reviews sit inside the third one. The two sentences that follow are the whole documented basis for the question in the title:

This factor's also based on info like how many websites link to your business and how many reviews you have. More reviews and positive ratings can help your business's local ranking.

That is it. No numbers, no threshold, no timeframe. The same page carries a line worth reading before you buy anything from anyone: "There's no way to request or pay for a better local ranking on Google."

One quirk, because you will see both words quoted around the web. The page's opening line says local results are "mainly based on relevance, distance, and popularity", while the headings underneath it read Relevance, Distance and Prominence. Same idea, two labels, one page, today.

The honest reading is not flattering to review software, mine included. Distance is the factor you cannot move without opening a location. Relevance is mostly your category and what your profile says. Reviews are one input into the third factor, sitting alongside links from other websites. They matter. They are not the lever, and no review tool, Cloutly included, moves a map pack position on its own.

Do Google reviews affect ranking, or the click?

Both, and the second effect is larger and much faster.

Your rating does not put you in the pack. Your category, your address and a complete profile do most of that work. What the rating decides is which of the three listings gets tapped once all three are showing, and that split is wider than owners expect. BrightLocal's Local Consumer Review Survey 2026 (published 11 February 2026, 1,002 US consumers) found 68% of people will only use a business rated four stars or higher, up from 55% the year before, and 31% will only use one rated 4.5 or higher. In the same survey, 97% read reviews for local businesses.

Take Coastline Physio, a fictional clinic. Three physios show in the pack for "physio near me" in the same suburb, all within a kilometre of the searcher, all with the right primary category. Ranking has done its job and stopped. The one showing 4.9 from 180 reviews takes the calls. The one showing 4.2 from 11 does not, and no further ranking work changes that.

So: do Google reviews affect ranking? A little, indirectly, over months. Do they affect whether anyone rings you? Far more than that, and this week.

What local SEO practitioners believe about reviews

The best available read on the parts Google does not document is the Local Search Ranking Factors report, published by Darren Shaw of Whitespark on 6 November 2025. It asks 47 local search practitioners to score 187 factors from 0 to 5 for their impact on local pack and Maps rankings, then ranks them by total score.

Whitespark is direct about what the survey is: "None of the experts have special access to the internal workings of Google's local search algorithm." It is informed opinion from people who move local rankings for a living. It is not a leak.

Here is the top of that list, with every review factor that appears in it.

Two things stand out. The highest-scoring review factor is sixth, behind two profile settings, a distance factor and your address. And rating outranks review count, which is the reverse of how most businesses treat the problem.

Why review recency beats a bigger total

Recency comes in at 11 and a steady flow of reviews at 14, both inside the top 15 of 187. Neither appears anywhere in Google's documentation. Treat them as observation.

The consumer side of recency is not observation. In the same BrightLocal survey, 74% of people said they look for reviews written in the last three months, and 32% look for reviews from the last two weeks, up from 20% a year earlier. A profile carrying 300 reviews where the newest is from 2024 reads as a business that has stopped, whatever the average says.

That is why a review drive works poorly as a one-off event. Forty reviews in a fortnight and then nothing for a year gives you a good number and a stale page. Asking every customer, from the job, produces the pattern the practitioners and the consumers both respond to. Cloutly's review campaigns are built for that shape: a repeat customer is one contact, asked once, anyone who reviewed you in the last 90 days is skipped, and follow-ups stop the moment the review lands.

If you have not set the asking up yet, start with how to get more Google reviews, which covers the timing and the follow-up in detail.

Keywords in review text, and why not to engineer it

"Keywords in Native Google Reviews" ranks 36th of 187 in Whitespark's list. The effect the industry believes in is real and small. The reason to leave it alone is not that it does nothing. It is that asking for it breaks Google's rules.

Google's prohibited and restricted content policy tells merchants they "should not require or pressure users to leave ratings or write reviews while on the premises, nor should they request that specific content be included". The same policy bars merchants from offering "incentives ... in exchange for posting any review or revision or removal of a negative review", and from any move to "Discourage or prohibit negative reviews, or selectively solicit positive reviews from customers".

So the template that says "please mention emergency plumbing in Newcastle" is not a clever tactic. It is a policy breach on a profile you cannot afford to lose. The same page states what is allowed, in one sentence: "Solicit or encourage the posting of content that does represent a genuine experience, without offering incentives to do so or attempting to influence the rating or the contents of the review." Ask everyone, promptly. A customer asked the same day writes about the job they just had, and the language turns up on its own.

The parts of map pack ranking that are not about reviews

Four of the five factors above reviews are things you set once and then leave alone, and none of them are a review problem.

1. Your primary category, the highest-scoring factor in the survey and the field most often wrong on a profile someone set up years ago.
2. Complete and accurate information. Google's own words: "Businesses with complete and accurate info are more likely to show up in local search results." Hours are part of that, and "Business is Open at Time of Search" ranks fifth in Whitespark's list, so wrong or missing holiday hours cost you placement at exactly the hours people are searching.
3. Photos and videos, which Google asks for on the same page.
4. One address, written the same way everywhere. If you run several sites, more than one business on Google with the same name is its own problem and worth fixing before anything else.

The setup and verification steps are in our guide to adding your business to Google Maps. Past that list, local SEO is a specialism, and a review-management company is not the right source for the link building and landing page work sitting under the rest of it.

Do reviews affect AI Overviews and AI Mode?

They feed the same profile, so there is no separate lever to pull. Google's guide to optimizing for generative AI features says "our generative AI features on Google Search are rooted in our core Search ranking and quality systems", that "Where appropriate, generative AI responses can include product listings, product information, and information about local businesses", and that "Using products like Merchant Center ... and Google Business Profiles can help your products and services to be visible in both AI responses and other Google Search results".

The same page is blunt about the industry that has grown up around this: "many suggested 'hacks' aren't effective or supported by how Google Search actually works". The profile you fix for the map pack is the profile AI Overviews and AI Mode read. Whitespark's current edition scores every factor for AI search visibility as well as local pack impact, which tells you how new the question is and how little anyone can prove yet.

ChatGPT and Perplexity are a different matter. They do not run on Google's local index and they do not read your profile the same way. We tested what they name in does ChatGPT recommend businesses like yours. AI search optimization for local businesses covers all five surfaces together, and what each one documents about reviews.

What to do first

Fix your category, your address and your hours before anything else, because those outrank everything a review programme can do. Then ask every customer, every week, because that is what decides who rings you once the ranking has done its part. The numbers behind the second half are in our online review statistics for 2026, which carries a source on every figure.

Frequently asked questions

How many Google reviews do I need to rank in the map pack?

Google has never published a number and there is not one. The useful version of the question is comparative: open the pack for your own search, note the review count and rating on each of the three listings, and you have your target. How to get more Google reviews has a method for working out the gap.

Does replying to reviews improve ranking?

Google does not say so. What it says is narrower: "Positive reviews and helpful replies can help your business stand out." Replies are worth writing for the person reading them. The ranking claim attached to them elsewhere is not something Google has confirmed.

How long after getting reviews does ranking change?

Nobody outside Google can answer this, and anyone quoting a number is guessing. Reviews appear on your profile within a day or two in most cases. Any ranking effect is gradual and mixed in with everything else moving at the same time.

Do reviews on Yelp or Facebook affect my Google ranking?

Google's prominence wording covers "how many websites link to your business and how many reviews you have" and does not name third-party review sites. Practitioners generally treat them as a weaker signal than native Google reviews, which is consistent with Whitespark's list. They matter for the people reading them either way.

Can I pay to rank higher in the map pack?

No. Google states it plainly: "There's no way to request or pay for a better local ranking on Google." Ads are a separate thing, and Google labels them where they appear.

Is the map pack the same as Google Maps?

Not quite. The map pack is the three-result block on the search page. Tapping "More places" opens the local finder, and the Maps app is a third surface. They run on the same signals but show different numbers of results, so your position can differ between them.