Answer: Ask once and you learn nothing. In one test, 600 people ran the same questions 2,961 times. Two answers named the same businesses less than 1 time in 100 (Source: SparkToro, 2026). So sign out. Ask ChatGPT, Perplexity, Gemini, and Claude what a customer would type, five times each. Count how often they say your name.
Doing this properly costs nothing but patience. Doing it badly is worse than skipping it, because one flattering answer feels like proof and is not.
What should you actually ask?
Ask what a stranger with a problem would type, not what an owner would type. "Who is the best emergency plumber in Leeds" is a buyer question. "What do you think of [business name]" is a leading question, and it gets a polite answer whether or not the business would ever have come up on its own.
There is a free way to find the real phrasings. Start typing the question into a Google search box and read the predictions underneath. Google says plainly that "Autocomplete predictions reflect searches that have been done on Google" (Source: Google, 2020). Those predictions are search phrasings rather than AI prompts, and that gap is worth being honest about, but they come from real people and they beat guesswork.
Ten to twenty questions is a sensible list. Cover the different ways someone might describe the problem, and put the town or area in the question if the customers are local.
Why does asking once prove nothing?
Because the same question does not return the same answer twice. In the largest public test of this, about 600 volunteers ran the same prompts through the real consumer apps 2,961 times. The chance of seeing the same list of brands in any two responses came out under 1 in 100, and the same list in the same order closer to 1 in 1,000 (Source: SparkToro, 2026).
Researchers studying the same problem put it in one line: answers vary across runs, prompts, and time, "making one-off observations unreliable" (Source: Schulte et al., 2026).
So ask each question at least five times on each assistant, and spread the asks across a few days rather than firing them all off in one sitting. Five answers to one question tell you something. One answer tells you what the weather was like in the model that second.
Why is your own account the worst place to check?
Because it already knows you. ChatGPT shapes "its answers to users based on the contents of their previous conversations" (Source: TechCrunch, 2025). If you have spent months pasting your own words into that chat, it has been coached. Sign out, or use a private window.
Check in a signed out window or a temporary chat, ideally on a device that never gets used for the business. Location matters too, because answers to local questions follow wherever the app thinks the person is sitting.
What should you write down?
Four things each time you ask. The exact question. Which AI. The date. Whether it said your name. Then add a fifth if you can: which businesses it named instead. That last list is usually the most useful part, and it is the one you never see from the inside.
Then count. Named in 3 of 20 answers is a finding. Named once, in one answer, on one afternoon, is a mood.
What can a check like this never tell you?
Where you "rank". The order of names changes every time you ask, so a spot in the list is a made up number. The researcher behind that 2,961 run test calls any tool selling a rank in AI "full of baloney" (Source: SparkToro, 2026).
It is also not what ChatGPT tells your customer. That customer has a different screen, a different country, a different history, and a different afternoon. The honest description of a check like this is what the assistant returned, to these questions, on these dates.
And absence is not a zero. Coming up in none of twelve answers means the true rate is probably small, not that it is nothing.
Is this worth the trouble at all?
The numbers say yes. Half of US adults now use AI chatbots, 49%, up from a third two years earlier (Source: Pew Research Center, 2026). And in a survey of 1,002 US adults, 45% said they had used AI to find a local business in the past year, up from 6% a year earlier in the same survey (Source: BrightLocal, 2026).
A business can be perfectly findable on Google and still never come up in any of it. Checking is how that gets discovered before a competitor's name does the discovering.
Where does that leave a busy owner?
Done right, this is an afternoon of typing and counting, done again often enough to mean something. Most owners try it once, see a good answer, and stop. That is the one way to learn nothing.
We ask ChatGPT, Perplexity, Gemini, and Claude the questions your buyers really ask, many times each, and count how often each one names you, with the range printed next to every number. The first check is free.
Free. No card, no call.