To see what AI says about your business, ask the questions your buyers ask, in four engines, from a clean session, and write down exactly who gets named. You need a spreadsheet, the ten questions below and about an hour. Repeat it monthly with the same questions and you have a real measure of your AI visibility, for free.
This is the same method I use in paid audits, scaled down. Nothing here needs a tool or a subscription.
Why you have to check by hand
Your analytics will not show you this. When Pew Research tracked real Google users in March 2025, people clicked a result link on 8% of visits that showed an AI summary, against 15% on visits that did not. Most of the influence now happens inside the answer, where no click is ever recorded.
Google Search Console began rolling out a generative AI performance report in June 2026, and Bing Webmaster Tools has an AI performance report in preview. Both are worth turning on, but they count impressions and citations. Neither shows you the sentence the assistant actually wrote about you, or who it named instead.
Step 1: Set up a clean session
Assistants personalize. If you ask from your own logged-in account, the one you use to research your business every day, you will get a flattering answer nobody else sees.
- Use a private or incognito browser window.
- Stay logged out where the assistant allows it, or use a temporary chat with memory turned off.
- Put the city or neighborhood in every question rather than relying on your location.
Step 2: Write ten buyer questions
Write questions the way a stranger who needs you would type them, not the way you describe yourself. Most of your buyers do not know your name yet, so only two of the ten mention it.
| # | Type | Template | Example (HVAC company, Houston) |
|---|---|---|---|
| 1 | Discovery | Who is a good [category] in [city]? | Who is a good AC repair company in Houston? |
| 2 | Urgent | I need [service] today in [city]. Who should I call? | My AC stopped working. Who can come out today in Houston? |
| 3 | Condition | [Category] in [city] that [condition] | AC repair in Houston that offers financing on new units |
| 4 | Area and time | [Category] in [neighborhood] open [when] | Emergency AC repair in Katy open on Sunday |
| 5 | Best of | What are the best [category] in [city]? | What are the best HVAC companies in Houston? |
| 6 | Price | How much does [service] cost in [city], and who is fairly priced? | How much does a new AC unit cost in Houston, and who is fairly priced? |
| 7 | Comparison | [You] or [competitor] for [need]? | [Your company] or [competitor] for a new AC install? |
| 8 | Alternatives | Alternatives to [competitor] in [city] | What are some alternatives to [competitor] in Houston? |
| 9 | Brand facts | What do you know about [business] in [city]? | What do you know about [your company] in Houston? |
| 10 | Brand trust | Is [business] legit? What do reviews say? | Is [your company] legit? What do reviews say? |
Step 3: Run every question in four engines
Run all ten in ChatGPT, Perplexity, Gemini and Google Search, where AI Overviews appear on some queries and AI Mode sits in its own tab. That is forty answers. Copy each answer into your sheet word for word, because you will want the exact wording next month.
Then take questions 1, 2 and 5 and run them twice more in each engine. Answers vary between runs, and this tells you whether a mention is reliable or a one-off.
Step 4: Record what matters
Copy this header row into a spreadsheet. One row per answer.
Date,Engine,Question #,Named (Y/N),Position (1st/2nd/3rd),Competitors named,Sources cited,Anything wrong about you
2026-10-04,ChatGPT,1,N,,Company A; Company B,yelp.com; bbb.org,
2026-10-04,Perplexity,9,Y,1st,,yourdomain.com; google.com/maps,Lists old addressThe last column matters most. An assistant stating the wrong hours, an old address or that you have closed is costing you customers right now, and it is usually the fastest thing to fix.
Step 5: Work out your share of answers
Count the answers that name you and divide by the total you ran. If you were named in 6 of 40, your share of answers is 15%. Do the same for each competitor that keeps appearing.
That single number, tracked monthly with identical questions, is more honest than any visibility score a tool can give you. A score hides the transcript; your sheet is the transcript.
Step 6: Read the sources
Perplexity, Google and ChatGPT's search answers usually show where they got their information. List every domain cited across your forty answers and count how often each appears.
Those domains are your map. If one directory, one review site or one local publication keeps showing up, that is where to get your facts right and your name included first.
What your results are telling you
- You are named often and accurately. Protect it. Keep listings consistent and re-test monthly so you notice a slip early.
- A competitor is named and you are not. The answer is choosing someone it can verify. Why ChatGPT Recommends Your Competitor Instead of You covers the seven usual causes.
- Nobody is named, just generic advice. The engines cannot confidently name anyone in your category yet. That is the cheapest position to win from.
- You are named with wrong facts. Fix the listings and pages the wrong fact came from before anything else, then re-test.
Mistakes that make the results useless
- Testing from your own account, which the assistant may already associate with your business.
- Asking once and drawing conclusions. One run is an anecdote.
- Testing only your brand name. Buyers who already know your name are not the ones you are losing.
- Using your own jargon. Use the words customers use on the phone.
- Changing the questions each month. You cannot compare answers to different questions.
Frequently asked questions
Is there a free tool that does this automatically?
Paid trackers exist, and the free reports in Google Search Console and Bing Webmaster Tools are worth enabling. None of them show you the full answer text for your buyers' real questions, which is why the manual check is still worth the hour.
How often should I check?
Monthly, with the same questions, on roughly the same date. Weekly checks mostly measure random variation.
Why do I get different answers on my phone and my laptop?
Location, login state, app version and the model behind the answer all change the result. That is why the method fixes all of them before you start.
What should I do if AI says something false about my business?
Find where the false fact lives, usually an old listing or an outdated page, and correct it there. Reporting the answer inside the app may help, but fixing the source is what lasts.
Can I check Microsoft Copilot and Claude too?
Yes, add them as extra engines if your buyers use them. Keep the same ten questions so the results stay comparable.
Rather not spend the hour?
The $49 audit is this method done properly: 25 buyer questions across ChatGPT, Perplexity, Gemini and Google AI Overviews, the dated transcripts, a crawler and schema check, and a prioritized fix list. It is credited in full against your first month if you go ahead. Get the $49 audit.
Sources: Pew Research Center, July 2025 Search Console generative AI reports, PinMeTo Bing Webmaster Tools AI Performance, Wikipedia