Skip to content
AI Says Who

Articles

Is there a number one spot inside an AI answer?

Last updated

Answer: No. There is no top spot inside an AI answer, because a chat app writes a new answer each time somebody asks. 600 volunteers ran the same questions through ChatGPT and other chat apps 2,961 times. The businesses came back in the same order under 1 time in 1,000 (Source: SparkToro, 2026). Count how often those apps name your business.

A page of search results has slots. One thing sits at the top, another sits under it, and the next person who looks sees the same arrangement. A chat app does not work like that. It writes a reply in sentences, choosing one word after another, for every person who asks. So the order of names inside that reply is not a position anybody holds. It is how one sentence came out that time. A tool that reports a business sitting in third place inside an AI answer is reporting a number that was never there to read. What can be counted honestly instead is further down this page, with the tests that show why.

Why is there no number one spot to win?

Nothing is being kept in order, so there is no slot to hold. A chat app writes a new answer every time somebody asks, picking one word, then the next. Two people asking the same thing get two pieces of writing, not two copies of one list.

Search engines made everyone used to the other arrangement, where a result has a position and that position is the same for anybody who looks. Nothing inside a written AI reply works that way. The names in it were assembled on the spot, in the order the sentence needed them, and then thrown away.

What happens to the order when the same question is put twice?

It moves nearly every time. In January 2026 SparkToro, the research firm run by Rand Fishkin, published a test with 600 volunteers. They put 12 questions to the most popular AI tools in the United States, among them ChatGPT and Claude, 2,961 runs in all. The same list of businesses came back twice less than 1 time in 100. The same list in the same order came back less than 1 time in 1,000 (Source: SparkToro, 2026). So the order is the part that moves most.

Read that with its limits. The volunteers used their own accounts and copied the replies out by hand, so their own history and where they were sitting were part of what came back. Twelve questions is a small set, and the test covered a handful of apps rather than every one. The direction is the useful part, and it is a strong one: the list of names moves, and the order of those names moves ten times harder.

Can the order be locked down by turning the randomness off?

No, and that has been measured as well. An AI model has a setting that controls how much it varies its wording. In September 2025 an AI research company called Thinking Machines Lab turned that setting down to nothing, so the model would repeat itself as closely as it can. Then it sent the identical request 1,000 times to one large AI model, the same kind of software that answers questions inside a chat app. Back came 80 different answers (Source: Thinking Machines Lab, 2025).

The reason is not a loose setting. Answers get worked out in batches alongside whatever else is being handled at that moment, and the size of the batch changes the arithmetic very slightly, which is enough to change a word. That test ran on one large model that anybody can download, not on ChatGPT, Perplexity, Gemini, or Claude, so treat it as evidence about where the wobble comes from rather than a figure about those four. It comes from the way answers are produced, which is not a dial a business can reach.

So what does a tool mean when it shows a position?

It means somebody turned one reply into a ranked list. Fishkin, whose firm ran the volunteer test, put it bluntly: any tool that gives a "ranking position in AI" is "full of baloney" (Source: SparkToro, 2026).

A position needs a stable order underneath it before it can mean anything, and the order is the least stable part of the whole thing. So a report that moves a business from fourth place to second place has probably measured the weather. Next week it moves back, nothing about the business having changed.

What can honestly be counted instead?

Count how often an AI app names the business at all. Write a list of real customer questions and then never change it. Put each question to ChatGPT, Perplexity, Gemini, and Claude at least three times, and five is better. Count how many replies name the business, and write the date beside the count.

Fishkin called position tracking baloney, and in the same breath he said this other count holds up: "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric" (Source: SparkToro, 2026).

A question on that list reads like the words a customer would really use, such as how much a new boiler costs in Leeds, or who can clear a blocked drain tonight. There is also no good score to aim for. The only honest comparison is the same count next month, on the same questions, so the number next to the date is the whole point.

Two rules keep that count honest. Count each app on its own, because ChatGPT, Perplexity, Gemini, and Claude do not behave alike, and one blended figure across all four hides the app a business is missing from. And keep the questions and the depth the same every month, because a count over a different list of questions is not comparable to the one before it.

Why does being named at all matter more than where the name lands?

Because the person asking gets one answer and acts on it. There is no second page of results to click. There is no scrolling down to position nine. Being named in the reply, or missing from it, is the whole difference.

The asking is also ordinary now. 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).

That is also why a missing name is worth more attention than a low one. A business that never comes up has a real problem to work on. A business that came up second one week and fourth the next has no problem at all, and chasing that movement spends money on noise.

What should a business owner do about it?

Stop reading the position, and start reading the count. Run the check in the section above, then look at one thing only: how often the business gets named at all. A count of zero is the useful kind of bad news, because then the job is getting named in the first place rather than climbing. Getting named is a separate job, and the related answers listed with this page are the ones that cover it.

And put a question to any report that hands you a position. If it says a business ranks third inside an AI answer, ask what the reading was last week, and ask how many times it put each question. A tool that puts each question once a day has one coin flip per question, and the order inside that single answer is the part the volunteer test found moves hardest.

Keep the record boring and it starts to mean something. Same questions, same depth, the date written beside each round, counted for each app on its own. If the business is never named, that is the finding worth acting on, and it is a far better use of a morning than arguing with a position that was never real.

Where does that leave a business owner?

Nobody can sell a number one spot inside an AI answer, because there is no spot to sell. The honest version is a count. How often does a business get named, over a list of questions that never changes? Count it for each app on its own, and write the date beside it. That is what AI Says Who, the service behind this page, does.

We research the questions customers actually use, put them to ChatGPT, Perplexity, Gemini, and Claude several times each, and count how often the business comes up. No position, no single blended score, and the question list and the date sit beside every number. Gemini is asked with no live web search, so it reports what it already knows about a business and never shows a link. The first check is free, and the button below starts it.

Get my free Snapshot

Free. No card, no call.