How to check if AI mentions your business

Why it is worth twenty minutes

In BrightLocal's Local Consumer Review Survey 2026, a panel of 1,002 US adults, 45 percent said they had used ChatGPT or other generative AI tools for local recommendations, up from 6 percent the year before. That is a survey of what people say they do, not a measurement of your customers, but it is enough to justify a look. The same survey reports that AI tools became the third most popular source of business recommendations in that panel. Whether any of that is true for your customers is exactly what a check cannot tell you. It can only tell you what one product said about your category in your city on one day, and the look should be small, recorded and clear about that limit.

Minutes 0 to 3: choose one service and one place

Write down one service you actually provide, in the words a customer would use, and one city you actually serve. Water heater repair and water heater installation are different jobs; a general dental office and an orthodontic practice take different questions. Leave out work you do not do, however valuable it sounds. The question has to match a service you offer in a place you serve, or the result means nothing.

Minutes 3 to 6: write three questions

Use the three patterns from the free checklist and replace the brackets with real details. One discovery question that does not name you: Who is the best [category] in [city, state]? One question about a specific job: Which [category] in [city] handles [service]? One branded question that supplies your name: What services does [business] in [city] offer? Keep the branded question separate in your notes from the start, because recognizing a name you supplied is a different result from being named unprompted.

These are example patterns, not measured popular searches. If you want a longer set, the audit builds 40 to 60 of them from your category, city and services, and you can preview that plan at checkout without paying. It groups them by the job they do: who is best, who to hire, what it costs, what reviews say, how providers compare, a specific need, and questions that name you.

Minutes 6 to 16: ask, and record everything

Open one AI product and ask the three questions one at a time. For each one, record the product and model if shown, the date and time, the exact wording you typed, the location you asked about, the full answer, every business named in the order it appeared, and every source the answer cited. Copy, do not summarize. The answer you did not keep is the one you will want later. If the product shows sources, note the domain and the page for each one, because a source tells you which page an answer was built from, and that is the first thing you will want to compare with your own site. Expect a short list rather than a ranking: in BrightLocal's Local AI Visibility Study, published in September 2026, ChatGPT named about four businesses per answer, with a range of one to seven.

Keep questions that contain your name separate from discovery questions that do not. If an answer never arrives or the product declines to answer, write No answer returned. That is a different result from an answer that names other businesses and not yours, and treating the two as the same is an easy mistake to make in a manual check.

Do not type customer names, patient details, case facts or anything private into the questions. Public business information is all this method needs.

Minutes 16 to 18: classify each answer

Give each question one of three labels: Mentioned, Not mentioned, or No answer returned. Then add a second note for each: did the answer cite your website, cite someone else's, or cite nothing you could see? A mention is not a citation, and either can happen without the other. If another business was named, write the name down and stop there. The answer alone does not say why.

Minutes 18 to 20: choose one check, not a conclusion

Pick one thing you can verify today. If a question about a service you offer named others and not you, read the page on your site that should answer that question, on a phone, without signing in. If a branded question got your services wrong, compare the answer with your current records. Write down the page, the evidence, one proportionate action, an owner and a date to check the work. The free on-page worksheet has a line for each of those, and the blank two page PDF prints the same structure. If the branded question did not recognize you at all, compare the forms of your name on your site and your profiles before you check anything else.

What you must not do with three answers: build a page to chase one result, conclude that a competitor paid for placement, or count a later mention as proof that your change worked. Three answers are a starting point.

The caveat that governs everything above

Answers can vary between requests, models, locations and products. If you run the three questions again tomorrow, the list of names can change with no change on your side. BrightLocal's Local AI Visibility Study measured this: repeat runs of the same local prompt overlapped only 20 to 33 percent on average across the three platforms it tracked, and running the same prompt from different points on a map changed the result as well. So compare like with like: the same product, the same wording, the same place, with the dates recorded, and compare the question sets and the number of answers that came back before you compare anything else. A single check cannot establish a trend, a cause or market share. It can establish whether there is something worth investigating. Our own audit reports coverage gaps separately from mention rates for the same reason, and every check in the sample report begins with verification, never with a cause.

Or let the free check run it for you

The free check asks three local buyer questions about your category in your city through one available answer engine, labels each one Mentioned, Not mentioned or No answer returned, lists the other names in the sample and shows the engine that answered. The result appears on screen and arrives by email. It needs your business details, an email address and a quick verification, and it runs once per email per day with no card and no sales call.

The audit is the longer version of the same method: 40 to 60 questions for one business in one US city, retained evidence, and a prioritized list of checks in a PDF, with a delivery target of 48 hours and a full refund if you cancel before work starts. Preview the questions and the configured source before paying. If you want to see the whole question set for your own category and city before you run anything, the free question planner at topofanswers.com/tools/question-planner builds it in your browser, grouped by what the buyer wants, with nothing sent to us.