Where do we rank in ChatGPT? How to answer your board when AI answers change on every run
The honest answer to "where do we rank in ChatGPT" is that there is no stable rank. A January 2026 study by SparkToro and Gumshoe.ai found that ChatGPT and Google's AI almost never returned the same brand list twice. Report how often you appear across many repeated runs of a fixed question set, whether the description is accurate and whether buyers cite AI in the CRM, and brief the board on trend and range rather than a single position.
Key takeaways
Why "where do we rank" is the wrong question
Directors know search rankings, so they ask the same question of AI assistants. The question assumes a list with a stable order. AI answers are generated each time, and the evidence says the lists do not repeat.
The SparkToro and Gumshoe.ai study, reported by Search Engine Journal on January 30, 2026, ran 2,961 prompts across ChatGPT, Claude and Google's AI features. For 12 brand-recommendation prompts, each run 60 to 100 times per platform, ChatGPT and Google's AI had less than a 1 in 100 chance of returning the same brand list. Returning the same list in the same order was under 0.1%.
The authors conclude that ranking position in AI is an unreliable metric, and that how often a brand appears across many runs is more consistent. They also found more consistency in tight categories such as cloud computing than in broad ones.
Three cautions apply. The study was done with Gumshoe.ai, which sells AI tracking tools. It is not peer reviewed. And the authors left open how many runs are needed for reliable data. We cite it as a reason to be careful, not as a definitive measurement.
What buyers actually do with AI answers
The board question has a second half: does it matter? Two primary sources help.
Gartner's May 20, 2026 release, based on a survey of 645 B2B buyers, says 45% used generative AI during a recent purchase, mostly to research vendors and products. It also says 69% prefer to check AI-generated insights with sales reps, and that 51% think they are more likely to get misleading information from generative AI.
Forrester's January 2026 release says buyers start with genAI searches and then seek validation from trusted human sources. Its analyst is quoted: "Human interactions remain essential, even as genAI reshapes the buying process."
Read together, they say AI is part of research and buyers do not fully trust it. Appearing accurately in the answers matters because it shapes the shortlist, and sales conversations are where those answers are validated.
Get an AI visibility read across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews.
What to measure instead of rank
| Measure | Definition | Source | Honest limitation |
|---|---|---|---|
| Appearance rate | Share of repeated runs of each fixed question in which your company is named | Repeated prompt tests across assistants | Varies by assistant, wording and time; report a range and a trend |
| Accuracy | Share of answers that describe your company, products and category correctly | Same tests, scored by a reviewer against an approved fact sheet | Reviewer judgment; use two reviewers on a sample |
| Competitive presence | Which competitors are named alongside you and how often | Same tests | Not a ranking; a frequency table |
| AI-referred sessions and pipeline | Visits from AI assistants and opportunities where the source or first touch is an assistant | Analytics joined to the CRM | Misses buyers who read an answer and never click |
| Self-reported AI use | Buyers who say an assistant was part of their research | Form field and sales notes | Self-reported; directional only |
Our article on measuring share of answer explains the method in more detail, and our article on AI visibility KPIs for the board covers the quarterly report format.
A repeated-sampling method you can defend
- Fix the question set: 20 to 40 questions that a buyer in your category would ask, across awareness, comparison and selection, written once and approved by product marketing and sales.
- Fix the assistants and settings you will test, and record the date, assistant and version for every run.
- Run each question several times in each assistant in each period. As our working rule, start with 20 runs per question per assistant for a first read, then repeat the batch and compare. If two batches disagree widely, add runs until they agree within a tolerance you set before testing.
- Record whether your company appears, how it is described and who else appears. Score accuracy against a written fact sheet.
- Report the appearance rate as a range across batches, with the trend over time. Do not report single runs.
- Join the results to the CRM view of AI-referred and self-reported pipeline so the board sees both.
The 20-run starting point is our own working rule, not a published standard. The point is to set the rule in advance, apply it consistently and show the range.
The five answers to give your board
We suggest preparing short answers to the five questions that follow. Keep each to two or three sentences.
- Where do we rank? There is no stable rank. Across repeated runs of our fixed question set, we appear in X% to Y% of answers, with the range shown. (Fill in your own measured numbers.)
- Are we described correctly? We scored the answers against our fact sheet and found these specific errors, with owners and fixes.
- Who is named instead of us? These competitors appear most often on these questions, which tells us where to publish and where we need third-party proof.
- Is it producing pipeline? These opportunities cite an AI assistant as a source. We also expect some buyers to use AI without clicking, so we collect self-reported source.
- What are we doing and what do we need? These actions, owners and dates, and this decision from the board.
The first answer should contain your measured range, never an invented benchmark. If you have no baseline yet, say so and give the date you will have one.
For the program structure behind the actions, see our article on who owns AI visibility and our 90-day pilot with a go or no-go decision.
What not to say
- Do not present a vendor tool score as a rank. A score is a model built on samples, so ask the vendor for its method and run counts.
- Do not present a single screenshot as evidence of either success or failure.
- Do not promise a result. Assistants differ, change over time, and are influenced by sources you do not control.
- Do not hide the limits. Boards trust a leader who says what is not yet measurable.
Frequently asked questions
Can you rank in ChatGPT?
Not in the way you rank in search. A January 2026 study by SparkToro and Gumshoe.ai found repeated runs of the same prompt almost never returned the same brand list, so we report how often a brand appears across many runs rather than a position.
How should a CMO report AI visibility to the board?
Report appearance rate as a range with a trend, accuracy of the description, competitors named, and AI-referred and self-reported pipeline from the CRM, with the limitations stated.
How many times should you run each AI prompt?
There is no published standard. Our working rule is to start with 20 runs per question per assistant, repeat the batch and add runs until two batches agree within a tolerance set in advance.
Do B2B buyers use AI assistants to research vendors?
Gartner reported in May 2026 that 45% of 645 surveyed B2B buyers used generative AI during a recent purchase, mostly to research vendors and products, and that 69% prefer to validate AI insights with sales reps.
Sources
- Search Engine Journal: AI recommendations change with nearly every query, reporting the SparkToro and Gumshoe.ai study (published January 30, 2026)
- Gartner press release: Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights (May 20, 2026)
- Forrester press release: The State of Business Buying, 2026 (January 21, 2026)
Related service: Generative engine optimization. See also pricing and client results.
We run enterprise B2B demand generation for teams in Montreal, Toronto, Ottawa and the United States.