Answer: Use the questions your customers really search for, not ones you guessed. Google says its search suggestions come from real searches people did (Source: Google, 2020). Type the start of a customer question into Google. Write down what it offers. Then put that list to the AI chat apps ChatGPT, Perplexity, Gemini, and Claude. See if your business is named.
A question list is just the set of questions you put to an AI chat app, and it exists to answer one thing: when a customer asks one who to call, does your business get named? That is worth knowing because the asking went mainstream. 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). You cannot find that out without knowing which questions to put to it. Nobody can look up what people really type into AI, so the list has to be built out of real demand that somebody can actually see. Building it is free, and the limits of it are easy to say out loud. Both are below, with the documents that say so.
Why can you not just look up the questions people ask AI?
Because nobody publishes them. OpenAI, Perplexity, Google, and Anthropic each hold the words typed into their own chat app, and none of them hand those words out. Otterly, a company that sells AI answer tracking, says the same in print: until there is a report from OpenAI showing what people type, nobody knows for sure (Source: Otterly, 2026). So any list of questions used to check AI was built by somebody. What it was built from is the only thing that separates a real list from a guess.
What happens if you let AI invent the question list for you?
You get questions that sound right and read nothing like the ones real people type. Otterly compared the questions its own software had guessed against questions that marketing people said they would really type. The guessed ones ran 8.8 words on average. The real ones ran 15.1. Nearly one in five of the guessed questions opened with the word "best", at 18.8%, against 11.3% of the real ones (Source: Otterly, 2026).
Read that with its limits. It covers one market, software for tracking AI answers, and the real side came from people saying what they would type rather than from typed words anyone recorded. It is still the only public test of guessed questions against stated ones, and the company that published it sells the guessing. The direction it points is the useful part. Ask software to imagine your customers and you get a shorter, blunter version of them. So do not ask a chat app to write the question list for you.
What does a real question actually look like?
A real question is longer and more personal than something typed into a search box. In January 2026 the research agency Stella Rising asked 524 people who use AI to write out the words they would type when buying something ordinary. Only one in five wrote a short string of keywords. Eight in ten wrote six words or more. Of those 524 written questions, 28% mentioned price or budget, and 16% named a place (Source: Stella Rising, 2026).
So a question worth checking usually reads like a spoken sentence, and often carries a price or a town inside it. "How much does a new boiler cost in Leeds" is closer to the real thing than "boiler installation". Those answers were written out by people in a survey rather than taken from what they typed, so treat the shape as the finding and not the exact wording.
The same person asks in two different shapes, depending on one setting. A study of 80 million lines of browsing records found that with the web search setting switched on, people typed 4.2 words on average, about the length of a Google search. With it switched off, the same tool saw 23 words (Source: Semrush, 2025). A list of short category questions is not wrong, then. It is half the picture.
Where do you get real questions without paying for them?
From the list Google drops down while you type. Start typing the opening words of what a customer would search for, such as "best plumber in" or "how much does a boiler", and read what Google offers to finish the line with. Google calls that drop-down autocomplete, and it says those suggestions reflect searches that have been done on Google (Source: Google, 2020). Copy them down as they come, in the customer's words rather than your own.
That is free, it needs no account, and it works in any trade. Add the phrasings you already hear at the counter or on the phone, because those are the closest thing to a real spoken question that any business has on hand.
What is the honest catch?
Those suggestions are Google searches, not AI questions. The two overlap. Nobody has published how much. Google also never says the list is ordered by how many people searched. It names where you are, the language you use, and rising interest in a topic as things that move the list (Source: Google, 2020).
Use that Google list anyway, because it is the best free signal anybody has. Just write the limit down beside whatever the AI chat apps gave back. The questions came from Google searches, which is the closest free stand-in for what people ask an AI chat app, and it is not the same thing. Anyone who tells you their question list is what people really type into ChatGPT is selling a guess with a confident face on it.
Does the list change every month?
No, and that is the point of having one. Freeze the questions once. Then put the same ones to the four AI chat apps ChatGPT, Perplexity, Gemini, and Claude, month after month, and write the date beside each round. Count how often your business gets named in the replies. Change the questions and that count means nothing, because two months are no longer being compared on the same thing.
So what does a business owner do about it?
Three things. Write down the ten questions customers ask you most, in their words. Then type the start of each one into Google and add the suggestions that fit what you sell. Then put the whole list to ChatGPT, Perplexity, Gemini, and Claude, and note whether your business comes up.
Keep the real wording. Do not tidy "blocked drain" into "drainage services", because the tidy version is not what anybody typed. Put each question several times rather than once, because the reply moves on its own: about 600 volunteers ran the same prompts through the real chat apps, 2,961 runs in all, and the same prompt gave back the same list of businesses less than 1 time in 100 (Source: SparkToro, 2026). Write the date beside the result, leave the list alone, and run it again next month with the very same questions. If your business never comes up at all, that is the finding, and it is the thing worth working on rather than a reason to change the questions. A short list asked properly beats a long list asked once.
Where does that leave a business owner?
Nobody can show you the questions people really type into AI. The honest way to do this is to build the list out of real searches, say plainly that those searches are a stand-in, and keep the wording frozen. That is what AI Says Who, the service behind this page, does.
We research the questions your customers actually use, put them to ChatGPT, Perplexity, Gemini, and Claude several times each, and count how often your business comes up. The question list and the date sit beside the count, so next month means something. Gemini is asked with no live web search, so it reports what it already knows about your business and never shows a link. The first check is free, and the button below starts it.
Free. No card, no call.