A fixed position in ChatGPT does not exist, because AI builds every answer from scratch and the list of names keeps changing. What does mean something is the percentage of many repeated answers in which you are mentioned.
A fixed position in ChatGPT does not exist. Ask the same question again and you often get a different list, in a different order. Anyone who promises you a "number 1 spot in ChatGPT" is selling something that cannot be measured.
That is not just our opinion. It follows from repeated research by SparkToro, the company of Rand Fishkin. Below you can read what that research showed, what you can measure instead and how to recognise an empty promise.
Why a position in ChatGPT does not exist
An AI assistant does not look things up in a fixed index the way a classic search engine does. The model builds every answer from scratch, with randomness as a built-in part. As a result, the list of names shifts with every question.
Google largely shows the same results for the same search, so a position means something there. ChatGPT lacks that stability. A position there is at best a snapshot of one single run.
Who asks the question also makes a difference. People phrase the same need in very different ways, as SparkToro showed in February 2026. So there is not even one single question on which you could hold a spot.
What the SparkToro research showed
In January 2026 Rand Fishkin, together with Patrick O'Donnell of Gumshoe.ai, published research into this inconsistency. Volunteers asked ChatGPT, Claude and Google AI 12 prompts over and over, each 60 to 100 times. According to the write-up by Search Engine Land of this January 2026 research, it came to 2,961 runs in total.
- If you ask ChatGPT or Google AI the same question 100 times, the chance of getting the same list of brands twice is less than 1 in 100 (SparkToro, January 2026).
- The chance that the order is also the same is less than 1 in 1,000 (SparkToro, January 2026).
- The AI sometimes named influencers, TikTok accounts and estate agencies that no longer existed.
AIs do not give consistent lists of brand or product recommendations. If you don't like an answer, or your brand doesn't show up where you want it to, just ask a few more times.
Put simply: AI does not give fixed lists of recommendations. If your brand is not in it, just ask a few more times. The reverse is also true: if you are in it, that may be different next time.
That is why Fishkin is blunt about rank positions. In January 2026 he called tracking them "foolhardy", in other words reckless.
In the Office Hours of March 2026 it was even shorter: "Rankings? Probably nonsense." In his view, rank positions are most likely nonsense.
Our founder Gijs Bodenstaff ran prompts for this research as a volunteer. That gives him experience with how it was set up. It does not mean that Rand Fishkin or SparkToro endorses this site.
What does count: visibility as a percentage
The research also showed something else: some names keep coming back. For the question about digital marketing consultants with e-commerce expertise, Smartsites appeared in 85 of the 95 Google AI answers. The hospital City of Hope appeared in 69 of the 71 ChatGPT answers, or 97 percent (SparkToro, January 2026).
That is the measure that does mean something: in what percentage of the answers you are mentioned. Fishkin himself changed his mind on this. His conclusion in January 2026: "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric".
In plain terms: the percentage of visibility across dozens to hundreds of prompts, each run several times, is a reasonable yardstick. In the Office Hours of March 2026 he called that percentage "more useful than rank position". So measuring is not pointless, measuring positions is; how to do it properly is explained in measuring your own AI visibility with the ten-times rule.
Visibility does say something about being mentioned, but nothing about quality. In the GEO benchmark by Gold Lemon (September 2026) the most mentioned provider appeared in 15.2% of the answers. That same provider scored 0.0 on transparency of its offering, the lowest of the top 20.
How large the pool of possible names is differs per topic. In a small market, such as a handful of car dealers in one town, the answers overlap more (SparkToro, January 2026). For a local specialist that is good news, see AI visibility for local businesses.
If someone promises you number 1 in ChatGPT
An offer for a "number 1 spot in ChatGPT" or a "top position in AI" is a red flag. It promises an outcome that does not exist and that nobody can reliably show. A screenshot of one favourable answer proves nothing.
So why is such a promise still being sold? A position is easy to explain, because everyone knows the idea of being at the top of Google. Fishkin warns about exactly this in the Office Hours of March 2026:
I'm worried that they're giving you rankings instead of visibility.
In other words, he fears that tools and providers give you positions instead of visibility. A position feels concrete, but in AI it mainly measures chance.
So always ask a provider these questions:
- How often do you run each prompt, and why that often?
- Where do the prompts come from: real customer questions or made up?
- Do you report a visibility percentage or a position?
- Do you show the raw answers and the calculation?
The last point ties in with Fishkin's closing advice from January 2026:
Before you spend a dime tracking AI visibility, make sure your provider answers the questions we've surfaced here and shows their math.
So do not spend a penny until the provider answers these questions and shows the sums. You will find more questions in the questions to ask every agency up front.
In our own assessment, a position guarantee in AI counts as a penalty point. How we weigh that is set out in the LSAI method for assessing specialists.
What you can do now
- With every AI offer, ask whether a position or a percentage is reported, and reject positions.
- Ask ChatGPT your most important customer question at least ten times, as SparkToro advised in March 2026, and note each time whether you are mentioned.
- Keep track of your results in a tally sheet for your AI mentions.
- Follow the independent ranking of GEO agencies; the first edition is in preparation.
Frequently asked questions
Is there a fixed position in ChatGPT?
No. ChatGPT builds every answer from scratch, so the list and the order keep changing. A position is at best a snapshot of one answer. See also: LSAI method.
Why does ChatGPT keep giving different answers?
The model contains an element of chance, and people ask the same question in very different ways. As a result, names and order shift with every question. See also: Getting mentioned in ChatGPT: the evidence lies outside your website.
So what should I measure instead?
The percentage of answers in which you are mentioned, across many repeats of the same questions. Rand Fishkin calls that a reasonable yardstick, unlike a position.
Can an agency guarantee me a top spot in ChatGPT?
No. Nobody controls the order of AI answers, and that order keeps changing. A guarantee of a position is therefore a red flag. See also: Measuring AI mentions: branded searches and direct traffic as evidence.
Is a screenshot of a good AI answer proof?
No. It is one answer out of many possible ones. Ask for the raw answers from many repeats and for the calculation behind them. See also: What is GEO? Mostly just SEO with a new label.
Is measuring AI visibility worthwhile at all, then?
Yes, if you measure a percentage across many runs with realistic questions. Only measuring positions is pointless.
Does this also apply to Claude and Google AI?
Yes. The SparkToro research looked at ChatGPT, Claude and Google AI. There too, the lists and the order varied widely.
Why does AI sometimes name businesses that no longer exist?
AI builds answers from what it has learned and found, and that can be out of date. In the SparkToro research even defunct accounts and offices came up.
Am I mentioned more often in a small local market?
The overlap between answers is greater when there are few possible names. A local specialist therefore sees a stable picture sooner than a national category does.
How do I recognise a provider that measures honestly?
They report percentages, run prompts many times, take questions from real customer data and show their calculation. They do not promise positions.
Sources
- NEW Research: AIs are highly inconsistent when recommending brands or products (Rand Fishkin, SparkToro) (January 2026)
- Write-up of the SparkToro research (Search Engine Land) (January 2026)
- Office Hours: Can You Actually Track AI Visibility? (SparkToro) (March 2026)
- Why Everyone Prompts AI Differently (Amanda Natividad, SparkToro) (February 2026)
- GEO-benchmark oktober 2026 (Gold Lemon) (September 2026)