You can check whether ChatGPT mentions your brand in an afternoon with 20 prompts, a clean session and a spreadsheet. The part most checks skip is repetition: ChatGPT does not give the same answer twice, so one run of one prompt tells you almost nothing. This method uses 20 unbranded buyer prompts, three runs each, and a mention rate you can track. Sources are dated 28 October 2026.
Start the 14-day trial, no card. Run your five most valuable buyer prompts on six engines.
What are the key takeaways?
- Test with 20 prompts that never name your brand. Mix six angles: problem-first, best-of, context-specific, budget, comparison and decision prompts. Branded prompts only test whether ChatGPT knows you exist.
- Use a clean session. OpenAI's temporary chat can still use your memories and custom instructions unless you choose Unpersonalized before the first message (OpenAI Help Center, read 28 October 2026). Ahrefs advises logging out.
- Run each prompt at least three times. SparkToro and Gumshoe found under a 1-in-100 chance that two runs return the same brand list. We recommend three runs as the minimum for a hand test.
- Record mention rate, not position. The same study found about a 1-in-1,000 chance of the same ordering. Treat position as noise.
- Separate mentions from citations. Ahrefs defines a mention as your brand name appearing in the generated answer. A citation is a link to your page. They need different fixes.
- Manual testing caps out at about 360 answers a pass across six engines. Past that, automate.
What counts as "mentioned" in a ChatGPT answer?
A mention is your brand name appearing in the text of ChatGPT's answer. A citation is a link to one of your pages in its sources. They are separate outcomes and need separate fixes. Score each answer on both, plus whether the description of your brand is accurate.
Ahrefs draws the same line in its guide to monitoring ChatGPT mentions: a missing mention is mostly a footprint problem, meaning your brand is thin in the third-party content the model draws on. A missing citation looks more like classic SEO, meaning your pages are not crawlable, current or clear enough to be sourced. Semrush's Prompt Tracking documentation uses a similar split between Mentions (brand name inside the answer) and Owned Sources (your pages cited).
Use four outcomes per answer:
- Named and cited. Your brand appears in the text and your domain appears in the sources.
- Named, not cited. The model knows you but sources someone else.
- Cited, not named. Your page is a source and the brand is not in the text.
- Absent. Neither.
Add one accuracy field. A mention that quotes last year's price is a problem, not a win.
Which 20 prompts should you test?
Test 20 prompts across six angles, none containing your brand name: four problem-first, four best-of, three context-specific, three budget, three comparison and three decision prompts. Write them the way a buyer types, not the way your marketing team words the category.
The example below uses expense management software for a 50-person company. Replace the category, size and competitor names with yours.
| # | Angle | Example prompt |
|---|---|---|
| 1 | Problem-first | How do I stop employees from losing receipts for company expenses? |
| 2 | Problem-first | What is the easiest way to automate expense reports for a remote team? |
| 3 | Problem-first | Our finance team spends days on month-end expense reconciliation. What software fixes that? |
| 4 | Problem-first | How do small companies manage corporate card spending without a full-time accountant? |
| 5 | Best-of | What is the best expense management software? |
| 6 | Best-of | Best expense management software for a 50-person company |
| 7 | Best-of | Top expense tools for startups in 2026 |
| 8 | Best-of | Which expense management platforms do finance teams recommend? |
| 9 | Context | Expense management software that integrates with QuickBooks and Slack |
| 10 | Context | Expense software for a company with employees in the US and Canada |
| 11 | Context | Which expense tool works best for a nonprofit? |
| 12 | Budget | Cheap expense management software under $10 per user |
| 13 | Budget | Free expense tracking software for a small business |
| 14 | Budget | Is paying for expense software worth it for a 20-person team? |
| 15 | Comparison | [Competitor A] vs [Competitor B] for expense management: which should I pick? |
| 16 | Comparison | Alternatives to [Competitor A] for expense management |
| 17 | Comparison | [Competitor A] vs [Competitor B] vs [Competitor C]: pricing and features |
| 18 | Decision | Is [Competitor A] worth the price? |
| 19 | Decision | What are the downsides of the most popular expense management tools? |
| 20 | Decision | What should I look for when choosing expense management software? |
Ahrefs recommends writing three to five phrasings for each angle, because a single prompt per angle depends too heavily on wording. SparkToro's data shows why. In its test, 142 human-written prompts about headphones had a semantic similarity score of 0.081, which means people asking about the same thing phrase it very differently. Brands still appeared in 55 to 77 percent of those answers. Visibility held up across phrasings even though the wording did not.
Pull real phrasings from your sales calls, support tickets and the "People also ask" box. Add 3 to 5 branded prompts separately ("What is [your brand] and what is it used for?") as an accuracy check, and keep them out of the 20.
How do you run the test without your own history skewing it?
Open a fresh session for each prompt and remove personalization. In ChatGPT, start a temporary chat and choose Unpersonalized before your first message. Ahrefs also advises logging out and using an incognito window. Record the date, whether web search was on, and your location.
OpenAI's Help Center says a temporary chat stays out of your history, but it "can use existing memories, custom instructions, and plugins to personalize responses" unless you pick Unpersonalized before the conversation begins. You cannot change that setting mid-chat. A plain temporary chat on an account with Memory on is therefore not a clean test.
Checklist for every run:
- New chat each time. Never ask prompt 6 in the same thread as prompt 5.
- Unpersonalized temporary chat, or logged out.
- Same device, browser and location for the whole pass.
- Web search on for one pass and off for another. Ahrefs notes that with search off, ChatGPT draws on training data, and with search on, live sources shape the answer. A gap with search off is a brand-recognition problem. A gap with search on is a content or sourcing problem.
- Write down the date. Answers drift as models and sources update.
Brand Radar, Ahrefs' tracker, collects its chatbot answers from the web version of each assistant with no stored user data, no prior context and no personalization, which is the condition you are recreating by hand.
How many times should you run each prompt?
Run each prompt at least three times; we recommend that as the minimum for a hand test. The research bar is far higher: in SparkToro's study, each prompt was run 60 to 100 times per tool. Three runs separates "never" from "usually" but cannot prove a small difference, so pool all 20 prompts for the headline rate.
SparkToro and Gumshoe published the study on 28 January 2026 and updated it on 28 June 2026. About 600 volunteers ran 12 prompts 2,961 times across ChatGPT, Claude and Google's AI surfaces in November and December 2025. Findings:
- Under a 1-in-100 chance that two runs of the same prompt returned the same list of brands.
- About a 1-in-1,000 chance of the same list in the same order.
- Rand Fishkin concluded that visibility percentage "across dozens to hundreds of prompts run multiple times" is a reasonable metric, and that any "ranking position in AI" is unreliable.
Two caveats from the study's own disclosures. It is not peer reviewed. And co-researcher Patrick O'Donnell works at Gumshoe.ai, a vendor in this category.
What three runs does and does not do. With three runs, a prompt's mention rate can only be 0, 33, 67 or 100 percent. That is enough to separate "never" from "usually". It is not enough to call a 33 versus a 67 a real difference. Use the 20-prompt average, which pools 60 runs, for the headline number and treat any single prompt as a lead. If you can afford five runs, the average gets steadier.
For one outlier data point, SparkToro found the agency Smartsites in 85 of 95 Google AI responses for its prompt set.
What should you record in the sheet?
Record one row per answer: date, engine, search on or off, prompt number, run number, whether you were named, whether you were cited, the cited domains, competitors named, and whether the description was accurate. Compute mention rate and citation rate from those rows. Do not record position.
| Column | Values | Why |
|---|---|---|
| Date | YYYY-MM-DD | Answers drift; you need a baseline |
| Engine | ChatGPT, Perplexity, Gemini, Claude, Google AI | Engines disagree |
| Search | On / Off | Separates recognition from sourcing |
| Prompt # | 1 to 20 | Ties back to the angle |
| Run # | 1 to 3 | Gives you a rate |
| Named | Y / N | Mention rate |
| Cited | Y / N | Citation rate |
| Domains cited | List | Shows who the model trusts |
| Competitors named | List | Share of voice |
| Accurate | Y / N / n/a | Catches stale facts |
Four numbers to compute after each pass:
- Mention rate = answers naming you ÷ total answers.
- Citation rate = answers citing your domain ÷ total answers.
- Competitor frequency = how often each competitor is named, ranked.
- Top cited domains = the sources that appear most across all answers.
The last one is the most useful for action. If three review sites account for most citations on your category prompts, that is where your brand needs to be present.
Can a free checker do this for you?
A free checker gives you a one-time snapshot, not a repeatable test. Ahrefs, Semrush and HubSpot all offer one. Each runs its own prompts, so you cannot control the wording, the number of runs or the session conditions. Use them to sanity-check your manual result.
| Tool | What it covers | Limits (28 October 2026) |
|---|---|---|
| Ahrefs AI Visibility Checker | ChatGPT, Gemini, Perplexity, Copilot, AI Overviews, AI Mode | No signup; top-5 preview |
| Semrush AI Search Visibility Checker | ChatGPT, Gemini, AI Overviews, AI Mode | No account; 3 uses a day |
| HubSpot AI Search Grader | ChatGPT (GPT-5.4 mini), Perplexity, Gemini | Free and one-time; short form for the full report; score out of 100 across five weighted dimensions |
None of them lets you run your own 20 prompts three times each. They answer "roughly where do we stand", which is a fine first question. For a repeatable baseline, use the sheet above.
When does the manual method stop working?
The manual method breaks at 360 answers a pass: 20 prompts, 3 runs and 6 engines, each one read and logged by hand, before you repeat it weekly or add a web-search-off pass. Past that point, automate.
The arithmetic scales fast:
- 20 prompts x 3 runs x 6 engines = 360 answers per pass.
- Add the web-search-off variant on ChatGPT and Perplexity and it is 480.
- Repeat weekly for a month and it is more than 1,400 answers.
- At SparkToro's 60 to 100 runs per prompt, 20 prompts on 6 engines is 7,200 to 12,000 answers per pass.
Automation changes the economics, not the method. Mentionova runs the same prompts on a schedule across ChatGPT, Perplexity, Claude, Gemini, Google and Reddit, stores every answer, and logs mentions, citations, position and sentiment per run. It refreshes weekly, with options for every two weeks or monthly. That is not the 60-to-100-run cadence SparkToro describes, and it is not daily. It does remove the six hours of copy-and-paste and gives you a trend line. The AI visibility tracking product page lists what each run records. Setup for your prompt set is in the adding prompts guide.
The 14-day trial gives you 5 prompts, 1 workspace and all six engines, no card. Use your five weakest prompts from the manual pass.
Take your five weakest prompts into a 14-day trial, no card, and see all six engines side by side.
For vendor options and what each costs, see AI visibility tool pricing in 2026 and best AI visibility tools.
What do you do with the gaps?
Sort your results into four buckets and fix each differently. Absent on search-off runs means a brand-recognition gap: build third-party mentions. Absent on search-on runs means a sourcing gap: get onto the pages the model already cites. Named but wrong means update stale pages. Cited, not named means sharpen your own copy.
Ahrefs' guidance on each case:
- Zero mentions. Build content clusters around the questions buyers ask, pursue placements in the publications ChatGPT cites, generate reviews and state plainly what you do and who you are for.
- Inaccurate mentions. Publish updated pages that state the correct fact, such as the current price, and update third-party profiles. Some errors fade as models refresh.
- Competitors ahead. List the prompts where they appear and you do not, then study the sources behind those answers.
Why bother? Pew Research's analysis of 68,879 searches by 900 US adults in March 2025, published 22 July 2025, found users clicked a traditional result on 8 percent of visits when an AI summary appeared, against 15 percent without one, and clicked a link inside the summary on 1 percent of visits. The study is observational. If fewer people click through, being named in the answer matters more.
More on how engines pick sources: how AI engines cite, how to rank on ChatGPT and the six-engine divergence index. For the statistics behind these claims, see LLM citation statistics and AI recommendation statistics.
Where does each number come from?
All sources read on 28 October 2026.
- SparkToro and Gumshoe, "AI brand recommendations are inconsistent" study, published 28 January 2026, modified 28 June 2026 (sparktoro.com).
- Pew Research Center short read on AI summaries and clicks, 22 July 2025 (pewresearch.org).
- OpenAI Help Center, "Temporary chat in ChatGPT" (help.openai.com/en/articles/8914046).
- Ahrefs, "How to Monitor Brand Mentions in ChatGPT" (ahrefs.com/blog/monitor-brand-mentions-chatgpt) and the Brand Radar help center article, updated 3 September 2026.
- Semrush Knowledge Base, Prompt Tracking (semrush.com/kb/1503-prompt-tracking).
- Free checker pages: ahrefs.com/ai-visibility-checker, semrush.com/free-tools/ai-search-visibility-checker, hubspot.com/ai-search-grader.
- Mentionova plan limits: mentionova.com/docs/billing/plans-overview.
What do people ask about checking ChatGPT for brand mentions?
How do I check if ChatGPT mentions my brand?
Write 20 buyer prompts that do not name your brand, run each three times in a fresh, unpersonalized ChatGPT session, and log whether you were named and cited. Divide mentions by total answers to get a mention rate. Repeat monthly, or automate with a tracker.
Can I just ask ChatGPT whether it knows my brand?
You can, but it only tests recognition, not recommendation. Buyers rarely type your brand name first. Use that as one accuracy check on a few branded prompts, and base your visibility score on unbranded category prompts where you have to be chosen.
Why does ChatGPT give different answers to the same question?
Its responses are not deterministic. SparkToro's study found under a 1-in-100 chance that two runs returned the same list of brands. Ahrefs makes the same point: wording, mentions and citations are in near-constant flux. That is why you measure a rate across many runs.
How many prompts should I test?
Start with 20 across six angles. That is enough to see patterns without making manual logging unmanageable. Mentionova's trial uses 5 prompts, and plan limits run from 25 to 250 prompts for brands. SparkToro's own bar is dozens to hundreds of prompts run multiple times.
Does my own ChatGPT history affect the result?
Yes, if personalization is on. A temporary chat can still use memories and custom instructions unless you choose Unpersonalized before the first message. Logging out or using incognito also helps. Test in the same conditions every time.
Is there a free tool to check brand visibility in ChatGPT?
Ahrefs, Semrush and HubSpot each offer a free checker. Ahrefs needs no signup, Semrush limits you to 3 uses a day, and HubSpot's is one-time with a form for the full report. They give a snapshot, not a repeatable baseline.
How often should I re-run the check?
Monthly is a sensible minimum for a manual pass, since answers drift. Weekly makes sense after you ship content aimed at a gap. Ahrefs suggests looking for changes 4 to 8 weeks after you publish or earn press.
What is the difference between a mention and a citation?
A mention is your brand name in the answer text. A citation is a link to your page in the sources. You can have either without the other, and each is fixed differently: mentions through wider third-party presence, citations through crawlable, current, clearly structured pages.
Run the same prompts across six engines for 14 days. Start the trial, no card.