How to Check If ChatGPT Recommends Your Brand
In short: write six to eight buyer questions without your brand name, ask each one in a fresh chat on ChatGPT and a few other assistants, repeat several times, and count how often you are named. It takes about an hour and needs no tools.
The goal is simple: find out whether an AI assistant names your brand when a customer asks a question before they know who you are. Because answers vary from run to run, the method below repeats each question and counts, instead of trusting one answer.
Step 1: Write the questions your customers ask
Write six to eight questions a real customer would ask before buying or hiring in your category. Leave your brand name out, because a shopper who does not know you cannot ask for you by name. Good questions are specific: a use case, a type of customer, a feature or a price point. Mix a few broad ones with several narrow ones. For the wording, borrow phrases from your customer emails, reviews and the search terms in your analytics.
- "Best lightweight tents for a first backpacking trip"
- "Best project management software for a small agency"
- "Who can build a mobile app for a small fintech company?"
Step 2: Use a fresh chat for every question
Open a new chat for each question so earlier answers do not influence later ones. If the assistant offers a temporary or private chat, use it, so memory and past conversations do not shape the answer. Keep the wording exactly the same every time you repeat the check. Note the model name shown, whether web search was on and whether you were logged in, because these can change answers.
Step 3: Ask each question at least three times
Answers are not stable. In a SparkToro and Gumshoe study of nearly 3,000 runs, the odds of getting the same list of brands twice were under 1 in 100, and the authors advise tracking how often a brand appears across runs rather than its exact position. So ask every question at least three times, each in its own fresh chat, and repeat the whole set on another day.
Step 4: Record what the assistant says
A simple spreadsheet is enough. One row per answer:
| Date | Assistant | Question | Run | You named? | Other brands named | Sources cited | Described correctly? | Settings |
|---|---|---|---|---|---|---|---|---|
| Oct 15 | ChatGPT | Best tent for windy conditions | 1 | No | Brand A, Brand B, Brand C | reviewsite.example | n/a | model shown; search on; logged out |
The row above is a made-up illustration of the format, not a result.
Step 5: Work out your mention rate
Count the answers where you were named and divide by the total. For example, with seven questions asked three times each you have 21 answers. If you were named in 4 of them, your mention rate is 4 of 21, about 19%. With 21 answers, one more mention moves the rate by about 5 points (5 of 21 is 24%), so treat differences of a few points as noise. The numbers here are only a worked example, not a benchmark. What matters is the comparison later: same questions, same wording, same assistants.
Step 6: Ask about your brand by name
In a new chat, ask "What is [your brand] and what do they make?" and "Is [your brand] a good choice for [use case]?". Check the answer for wrong facts: products, prices, location, claims. Keep these answers apart from your category questions. A brand usually looks better when you ask about it by name than when you ask a buyer question, so this tests accuracy, not visibility.
Step 7: Repeat on other assistants
Run the same questions on Perplexity, Gemini, Claude and Google's AI Overviews or AI Mode. They use different sources and show them differently, so a brand can appear in one and be missing in another. Perplexity shows its sources next to the answer, which makes it a good place to see where Perplexity's recommendations come from; other assistants may use different sources. For Google's AI features, Search Console has a Generative AI performance report, which Google points to for measurement. How to Measure AI Search Traffic and Visibility explains that report and what your analytics can show.
Step 8: Repeat a week later
Run the same questions again after a week and compare. Look for patterns across runs, not for a single answer.
How to read your results
- Known by name, but missing from category questions. The assistant knows who you are but does not name you to people still choosing. You need clear pages and independent proof for those specific questions.
- Named, but described wrongly. Fix the facts at the source: your own pages first, then listings and directories.
- The same websites keep appearing as sources. Those pages are where recommendations are decided. See whether you are on them and how you are described.
- The same competitors keep appearing. Look at why: what do their pages answer that yours do not?
When you know what is missing, this guide covers the usual causes and this one gives the checklist.
What a self-check cannot tell you
A handful of questions is a single reading, not a ranking. Answers vary between runs, assistants and days, and your own history can influence what you see. Repeating the same questions weekly across more assistants is separate ongoing work (how we test).
Questions people ask
How many questions should I test?
Six to eight is enough for a first self-check. For ongoing tracking we use up to 50 questions, because a larger set shows patterns that a handful cannot.
Why does ChatGPT give me a different answer each time?
Assistants generate answers fresh each time, and the sources they pull in change. A published study by SparkToro and Gumshoe, covering nearly 3,000 runs, found identical brand lists were very rare. That is why you repeat each question and count how often you appear.
Do I need a paid tool to check this?
No. A spreadsheet and an hour are enough for a first look. Tools help when you want to run many questions every week across several assistants.
Sources
- SparkToro and Gumshoe study on repeat AI recommendation lists, Search Engine Land, Jan 28, 2026. Run with Gumshoe.ai, an AI-tracking startup that sells such tools, and described by Search Engine Journal as not peer-reviewed research. Volunteers ran the prompts in their own AI settings. Tested ChatGPT, Claude and Google's AI. Sample: 600 volunteers, 12 prompts, nearly 3,000 runs.
- Google, optimizing for generative AI features in Search, updated Jul 10, 2026: Generative AI performance report in Search Console.
- Top Seed AI, how we test: the method behind our own checks.
Related guides
We show you which pages AI reads in your field, then write the comparison page you're missing, ready to publish.
Get your free comparison page