Track Brand Mentions in ChatGPT for Free: A Repeatable Monthly Routine

A free check tells you one thing well: for the questions you chose, on the day you asked, did ChatGPT name your brand, and did it describe you correctly.

That is useful. It shows whether competitors appear where you don't, and it surfaces wrong prices, old product names and mixed-up facts.

It does not tell you how ChatGPT answers your real buyers. OpenAI's help page says ChatGPT search "typically rewrites your query into one or more targeted queries" before it sends them to search providers. Your prompt is the input you control. The query that actually retrieves pages is one you never see.

So treat the routine below as a consistent instrument. Its value comes from asking the same thing the same way every month. For the deeper version of this method, see our reproducible ChatGPT prompt tracking method.

This takes about an hour the first time and 30 minutes a month after that.

1. Write 10 to 20 fixed prompts. Use the questions a buyer asks before they know your name. Say you run a dental clinic in Austin: "best dentist in Austin for implants", "how much do dental implants cost in Austin", "is [your clinic] good". Mix category questions, comparison questions and one or two questions that name you.
2. Freeze the wording. Store the list in the first tab of your spreadsheet. Never edit a prompt. If you need a new one, add it as a new row with a start date.
3. Use a clean session. OpenAI says that if memory is enabled, ChatGPT may use saved memories when rewriting a search query. Turn memory off in settings, or run the prompts logged out. Start a new chat for each prompt.
4. Run each prompt and log the answer. Use these columns:
   • Date
   • Prompt
   • Brand named (yes or no)
   • Competitors named
   • Citations shown (yes or no)
   • Your URLs cited
   • Facts stated about you
   • Any fact that is wrong, and what the correct fact is
5. Save the full answer text in a linked document or a long-text column. You'll want it when a fact changes.
6. Repeat on the same day each month, with the same account state and the same settings.

Keep the wrong-fact column clean. It becomes the input for an AI error log and correction tracker.

Calculate one number per month: prompts where ChatGPT named you, divided by prompts run. Watch how it moves over three or more months. One month is noise.

Then split the answers by the citations column.

• Citations shown: ChatGPT searched the web. Pages it could fetch shaped the answer. This is where changes to your site can show up.
• No citations: the answer most likely came from the model's training data, which is frozen at a cutoff. As of October 2026, OpenAI's model docs list a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path. Nothing you publish this month reaches these answers.

Next, act on what you see.

• Wrong facts: state the correct fact plainly (the price, the number, the one-line definition) on a page that loads without JavaScript. Vercel found ChatGPT's crawlers fetch JavaScript files but don't execute them.
• Never named in search answers: check access first. OAI-SearchBot controls ChatGPT search, and OpenAI says sites opted out of it "will not be shown in ChatGPT search answers." GPTBot only covers training. Your robots.txt should include something like this:

User-agent: OAI-SearchBot
Allow: /

Then confirm your firewall or CDN doesn't block OpenAI's published searchbot IP ranges. A robots.txt allow means nothing behind a firewall that returns 403.

• Competitors named, you not: look at which sites ChatGPT cites. Being named on those pages is associated with appearing in the answer, with no guaranteed effect.

Be plain with your team about what this routine can't do.

• It measures a proxy. ChatGPT rewrites prompts, and search is used at the model's discretion. Your list approximates real demand. It doesn't capture it.
• It has no rank. There's no position data inside an AI answer to check a rank against. Record named or not named, never a position.
• Answers vary. The same prompt can return a different answer an hour later. A small list exaggerates swings.
• Routing changes the model. Which model serves the answer depends on plan and routing, and each has its own cutoff.
• It covers one engine. ChatGPT is not Perplexity, Gemini or Google AI Overviews. Repeat the routine per engine, as in free AI brand monitoring across several engines, and don't blend them into one number.

None of these make the routine worthless. They make the trend more reliable than any single result. For how prompt sampling compares with numbers the engines publish themselves, read prompt-based tracking vs engine-reported AI data.

If the monthly hour is the problem, automate it. The 99Visibility Free plan costs $0 and covers 1 brand, 5 keywords and 1 audit a month across 2 platforms. The audit checks how those platforms describe you and flags wrong facts, which replaces steps 4 and 5 above. Check which two platforms the plan covers when you start on the Free plan, with no card required.

For a one-off snapshot instead, the free checker on this site runs up to three buyer questions and scores the answers. It is a snapshot, not continuous tracking.

Automation doesn't remove the proxy problem. It just runs the same instrument on schedule. Keep your own prompt list as a cross-check, and read what a free snapshot misses before you rely on any single score.

How many prompts do I need for free ChatGPT brand tracking? Ten to twenty is enough to see a trend without turning the routine into a chore. More prompts reduce noise, but only if you keep running all of them every month. A short list you run consistently beats a long list you abandon. Freeze the wording, add new prompts as new rows with a start date, and never compare months that used different lists.

Should I run my prompts logged in or logged out? Logged out, or logged in with memory turned off. OpenAI says ChatGPT may use saved memories when it rewrites a search query, so a logged-in session with memory on can produce answers your buyers would never see. Whatever you choose, keep it the same every month. Changing the session setup mid-series breaks the comparison.

Does allowing GPTBot help ChatGPT mention my brand? No. OpenAI documents GPTBot as the crawler for training data only, and claims no search benefit from allowing it. ChatGPT search uses OAI-SearchBot, and OpenAI states that each setting is independent of the others. You can block GPTBot and still allow OAI-SearchBot. Check that OAI-SearchBot is allowed and that your firewall lets it through.

Can I see the exact queries ChatGPT runs for my prompts? Not in the ChatGPT app. OpenAI says ChatGPT search rewrites your prompt into one or more targeted queries, and those queries are not shown to you. That's why any prompt list, manual or automated, is a proxy. Treat month-over-month movement on a fixed list as the signal, rather than any single answer.

What should I do when ChatGPT states a wrong fact about my brand? Log the wrong fact, the correct fact and the date. Then make sure the correct fact appears in plain text on a page of your site that loads without JavaScript. If the answer showed no citations, it likely came from training data and won't change until a newer model. If it showed citations, check which pages ChatGPT cited.