Playbook9 min read
How to check your brand in ChatGPT answers, step by step
You check your brand in ChatGPT by asking the questions your buyers ask, in a logged-out session, many times, and counting the share of answers that name you. A single answer is not evidence in either direction, because the same prompt returns a different list of brands almost every time.
How do you check if ChatGPT mentions your brand?
You write a fixed list of buyer questions, run each one in a logged-out ChatGPT session more than once, and record whether your brand appears. Then you divide the runs that name you by all runs. That share is your visibility figure for the period.
The method is slow and free. Ahrefs describes manual checking as a way to set an initial benchmark, because the work does not scale to ongoing trackingmarket. A benchmark is what a company with no AI visibility data needs first, and a spreadsheet is enough to produce it.
Why does one ChatGPT answer prove nothing?
Because the answer is assembled again every time you ask. SparkToro and Gumshoe.ai ran 2,961 prompts, each 60 to 100 times per platform, and found under a one in a hundred chance that ChatGPT or Google's AI returned the same list of brands twicemarket. AirOps measured the same instability from another angle: only 30% of brands stay visible from one answer to the nextmarket.
Two consequences follow. A single answer that includes you does not mean you are visible, and a single answer that omits you does not mean you are invisible. The SparkToro data also shows what is stable: top brands in a category appeared in 55% to 77% of headphone responsesmarket. The share of answers is steadier than any position in a list.
How do you set up a fair test?
Remove everything that personalizes the answer, then fix the variables you want to compare. Four settings matter.
- Log out or open a private window, so past chats and saved memory do not shape the resultmarket.
- Test with web search on and off, because the model's stored knowledge and live retrieval can give different answersmarket.
- Ask unbranded questions. A prompt that names your brand will almost always return your brand, so it measures nothingmarket.
- Run every prompt more than once. LLM Pulse suggests twice on different days, and Frase suggests repeating the whole batch a week latermarket.
Search-on and search-off results follow different timelines. Waikay puts the lag for grounded search at days to weeks and for training data at months to over a yearmarket. If you appear only with search on, your content is being retrieved. If you appear with search off, the model has learned something about you.
How many prompts do you need?
Between 20 and 50 for a first audit. meetergo suggests a bank of 20 to 30 questions that customers actually use, and LLM Pulse recommends 30 to 50 for a first full auditmarket. Frase sets the lower bound for a quick start at five to ten real buyer questionsmarket.
| Audit size | Prompts | Runs per prompt | Best for |
|---|---|---|---|
| Quick look | 5 to 10 buyer questionsmarket | Repeat the batch one week later | Finding out whether you appear at all |
| Small company benchmark | 20 to 30 promptsmarket | At least 2market | A baseline you can compare next month |
| Full first audit | 30 to 50 promptsmarket | At least 2, on different daysmarket | A company that sells in many categories |
| Weekly program | 20 to 50 promptsmarket | Same set each weekmarket | Teams that invest in AI visibility |
| Tracker-based monitoring | Hundreds of promptsestimate | Automated | Brands with many products |
Build the list from the questions your sales calls produce: price comparisons, "best for" questions, "how do I choose" questions and problem descriptions. Group them by intent so you can see which kind of question names you and which does not.
What do you record for each answer?
Record three things: whether your brand is named, whether your site is cited with a link, and how the answer describes you. AirOps defines these as mention rate, citation share and a sentiment scoremarket, and LLM Pulse adds position-weighted visibility and accuracy of the factual claimsmarket.
A mention without a link still shapes what a buyer believes about your company, while a citation can send measurable traffic. Keep them in separate columns. A one-row-per-run sheet with the prompt, the date, the search setting, the mention, the citation and a line of the exact wording is enough.
Accuracy deserves its own column. If ChatGPT names you but gets your service or your prices wrong, the mention has cost you something, and a mention count alone would hide that.
Where do mentions come from, and what does that change?
Mostly from third-party pages. AirOps found that 85% of AI brand mentions in its data came from third-party pages rather than the brand's own domainmarket. A study by Shero Commerce points the same way: only 2.8% of 1,851 cited sources were brand-owned pages, while third-party review sites made up 59%market.
So when your count is low, check the third-party pages first: comparison articles, listicles and review sites that name your competitors and skip you. Your own blog is one input among several, and the audit sheet shows you which questions have no outside page naming you.
Can robots.txt keep you out of ChatGPT answers?
Yes. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links, and that changes to robots.txt can take about 24 hours to applymarket. Search Engine World describes OAI-SearchBot as the crawler that surfaces websites inside AI search resultsmarket.
Read your robots.txt before you read any answer. A site that blocks the search crawler by an old rule, a security plugin or a copied template has a setup problem, and no amount of content will fix it. Blocking GPTBot is a separate decision, because that crawler gathers content that may be used in training.
How often should you repeat the check, and what then?
Weekly if you are actively working on visibility, monthly as the usual rhythm and quarterly at minimum. LLM Pulse gives exactly that laddermarket, and Frase and AirOps both describe a weekly cadence for a consistent prompt setmarket.
Keep the prompt list frozen between rounds. Changing questions changes the measurement, and you will not know whether a gain came from your work. After a content or outside-mention push, give it time before judging, since training data takes months to update while retrieval can move in days to weeksmarket.
Frequently asked questions
How do I check if ChatGPT mentions my brand for free?
Use a logged-out session, ask a fixed list of unbranded buyer questions and record every answer in a spreadsheetmarket. Run each question more than once and count the runs that name your brand.
How many prompts should I run?
A first audit uses 20 to 50 promptsmarket. Five to ten real buyer questions work as a quick lookmarket, but they only show whether you appear at all.
Why does ChatGPT give a different answer each time?
Because the answer is generated afresh for each request. SparkToro found under a one in a hundred chance of the same brand list across repeated runsmarket, which is why a share of many runs replaces any single result.
Should I test with web search on or off?
Both. Off shows what the model has stored about you, on shows what it finds right nowmarket. The two results can differ a lot, and the difference tells you which problem you have.
Can I block ChatGPT search from showing my site?
Yes, by disallowing OAI-SearchBot in robots.txt, and the change can take about 24 hours to applymarket. A blocked site can still appear as a navigational link, but not in answers.
Sources
- SparkToro and Gumshoe.ai research, reported by Search Engine Journal (2026-01-30): 2,961 prompts, 60 to 100 runs each, under one in a hundred chance of the same brand list, top headphone brands in 55% to 77% of responses.
- AirOps, "AI visibility metrics" (2026): 30% of brands stay visible from one answer to the next, 85% of mentions from third-party pages, mention rate, citation share, sentiment, weekly cadence of 20 to 50 prompts.
- Ahrefs, "How to monitor brand mentions in ChatGPT" (2026): logged-out and private session, web search on and off, manual checking as an initial benchmark.
- LLM Pulse, "AI visibility audit guide" (2026): 30 to 50 prompts, each run twice on different days, private session, quarterly, monthly or weekly re-audit.
- meetergo, "How to know if your brand is showing up in AI answers" (2026): logged-out sessions, unbranded buyer questions, a bank of 20 to 30 prompts.
- Waikay, "AI visibility channels" (2026): browsing on versus off, grounded search in days to weeks, training data in months to over a year, unbranded category prompts.
- Frase, "How to check if ChatGPT cites your brand" (2026): five to ten buyer questions, repeat the batch a week later, weekly checks.
- OpenAI, crawler documentation (2026), and Search Engine World (2026): OAI-SearchBot, robots.txt, about 24 hours.
- Search Engine Journal on the Shero Commerce study (2026-08-26): 2.8% of 1,851 cited sources brand-owned, 59% third-party review sites.