One ChatGPT check tells you almost nothing about your business

Ask again next week and the answer moves

Say you typed your business into ChatGPT a month ago and saw your name come back. You closed the tab and you've been going on that answer ever since. Type the same question today and there's a fair chance your name is missing. Nothing broke at your end. The answer you saw was built on the spot for that one question, and the next person to ask gets a freshly built one of their own. So one check is a sample of one.

What you want is a rate. How often does your name come back across several runs of the same question? That number costs about ten minutes of your week (spread over a few days), and it holds up over time, because the swing in any single reading averages out across several.

Two research teams got to the same finding in the past year, from opposite ends and with different methods.

What the research actually counted

AirOps published The 2026 State of AI Search on 2 December 2025, built with SEO researcher Kevin Indig off millions of AI-search datapoints across ChatGPT, Perplexity, Gemini and Google AI Mode. The report also measured how much the answers move. Only 30% of brands stay visible from one answer to the next, and just 20% are still there across five runs in a row (AirOps, December 2025).

Read that as a dental practice owner, a medspa manager or an HVAC contractor. Ask ChatGPT about your business today, see your name, and there's roughly a 2 in 3 chance the name is gone tomorrow when you ask the exact same thing. Nothing at your business has to change for that to happen, and no agency did anything wrong the night before.

One more number from the same report takes some of the sting out of that. More than half of the brands that drop out of an answer come back within two runs (AirOps, December 2025). So a miss on any single day is weak evidence, and a thin reason to change anything.

A caution on that study before you lean on it. It looks at big brands, so the numbers mainly describe big brands. What it documents is how the answers get built, and that holds at any size. Take the small-business read as an extension of the pattern, and keep the source attached when you repeat the number.

My own read comes from the smaller sites I audit by hand, and the swing AirOps describes is what I see there too. A business named on Monday and gone by Thursday, with nothing at the business to explain it.

The coin-flip study

The second piece of research came out of a much scrappier setup. Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai had 600 volunteers type 12 prompts into ChatGPT, Claude and Google's AI a combined 2,961 times (over November and December 2025). They turned every answer into a ranked list of brands. Then they compared the lists.

There's a less than 1 in 100 chance that ChatGPT or Google's AI, if asked 100 times, will give you the same list of brands in any two responses.

That's SparkToro's own write-up, published 27 January 2026. Checking once with ChatGPT and treating the answer as final is like flipping a coin once and calling it a law of physics.

Their clearest case is a search for the best cancer hospital on the West Coast. City of Hope in Los Angeles showed up in 69 of 71 ChatGPT answers (a 97% appearance rate). It was the top mention in only 25 of those responses. So a business can get named almost every time and still bounce around in the order from one answer to the next. Getting named and landing at the top are two separate wins, and an owner who only ever checks once has no way of telling which of them they've got.

Run the same question five to ten times, then count

The check costs about ten minutes spread over a week, and nobody has to touch your website.

1. Write down the question a customer would type. Who does teeth whitening near me. Which company installs a mini-split. Which lawyer handles a slip-and-fall case. Their words, off the top of their head, the way they'd say it to a friend.
2. Open a fresh chat every time. A new chat means the assistant has no memory of the last answer, so each run is a clean read. Signing out of a paid account helps too (a free account is what most customers are using).
3. Ask five to ten times across a few days. Tick a box each time your business gets named. Keep it on paper (or in one note on your phone).
4. Do the same on Claude, Perplexity and Google's AI (same question, same wording). Some will name you and some never will, and that gap is worth knowing before you spend money anywhere.

I have sat with owners while they worked through this. It usually lands on run three or four, when the same question comes back with a different set of names and they watch it move for themselves.

The count is the number to act on. Do not hold it up against the 30% above, because the two count different things. AirOps looked at a large group of brands and asked how many kept their spot from one answer to the next. You are counting one business, yours, across your own runs. City of Hope shows the gap. It got named in 97% of its own answers, in a report where only 30% of brands held their place.

So here is a rule of thumb for reading your own tally. It comes from the audits I run by hand, and neither study backs it. Coming back in most of your runs means the assistants reach for you often and you're holding a spot in the group they choose from. Coming back once or not at all is a real gap, and the next section is where to start on it.

Write the date next to the count. Do it again in a month, using the same wording on the same question, because a reworded question is a different question and the counts won't line up. Two counts a month apart tell you which way you're moving, which is more than any single answer will ever give you.

Give the assistant a page it can point at

The AirOps report splits what an assistant does with your business into two things. A mention is the assistant saying your name in the answer. A citation is the assistant pointing at one specific page of yours as where it got the information.

Brands that earn both are 40% more likely to come back in the next answer than brands with a citation alone (AirOps, December 2025). The brands that hold their place across runs tend to have both. That is an association, measured across a large set of brands, and it is as far as the evidence goes. Adding a citable page to a business that only ever gets named is a move nobody has tested.

Give the assistant a page worth pointing at. A services page that names the service the way a customer says it, with the price or the price range, how long the job takes, and who it's for. Written in plain text on the page, because a price that only exists inside a photo or behind a contact form stays invisible to the assistant.

Most homepages carry the business name and not much a customer could act on, so the assistant has no particular page of yours to cite when it builds the answer. That's one reason so many owners get named one week and skipped the next. The services page above answers the customer's question in full, in the words they would use, so the assistant has a reason to name you again on the following run.

The answers will keep moving, and anybody selling you a guaranteed ranking inside ChatGPT is guessing. So track the rate the same way each month, and keep a page worth citing.