How to track brand mentions in ChatGPT: a manual method and an automated one
A step-by-step method for tracking when ChatGPT mentions your brand: build a prompt set, run clean checks, log what matters, then automate it with daily tracking, citations, and sentiment.
Why this is harder than it looks
Tracking brand mentions on the web is a solved problem: a monitoring tool crawls pages and alerts you when your name appears. ChatGPT doesn't work like that. There is no page to crawl. Each answer is generated on demand, it can change between two identical questions asked a minute apart, and it can depend on whether the model searched the web, what it remembers about the user, and which country the question came from.
So "tracking mentions in ChatGPT" really means sampling. You ask a defined set of questions, repeatedly, under controlled conditions, and record what comes back. With ChatGPT at 820M+ weekly active users, the sample is worth taking seriously. Below is how to do it by hand first, which teaches you what to measure, and then how to automate it.
Step 1: build a prompt set that matches how buyers ask
Your brand name is the least useful prompt. People who already know you will ask about you; the mentions that matter happen when someone asks a category question and ChatGPT decides who to name.
Build the list in four groups:
- Category prompts. "What's the best payroll software for a 20-person startup?" These are where new buyers find you.
- Comparison prompts. "Gusto vs Rippling for a small team" (with your own competitors). ChatGPT often names a third option in these answers.
- Problem prompts. "How do I run payroll for contractors in three countries?" No brand in the question, but a recommendation in the answer.
- Brand prompts. "Is [your brand] good for startups?" Useful for checking accuracy and sentiment, less so for visibility.
Write each prompt the way a person would type it, not the way a keyword tool phrases it. Twenty to thirty prompts is enough to start by hand.
Step 2: run clean checks
Personalization skews results. If you ask ChatGPT about your own category from your work account, its memory of past conversations may nudge the answer toward you.
Use a temporary chat or a logged-out session for each check. If your market spans countries, note that answers can differ by location, and check from the location that matters.
Then run each prompt more than once. Three runs per prompt is a reasonable minimum for a manual check; it shows you quickly which mentions are stable and which come and go.
Step 3: log the right fields
A spreadsheet works for a manual baseline. For each prompt and run, record:
| Field | What to note | Why it matters |
|---|---|---|
| Mentioned? | Yes or no | The base visibility number |
| Position | First, second, later, or "also consider" | Being listed fifth is not the same as being recommended |
| Framing | How ChatGPT describes you, in a few words | Catches outdated or wrong claims |
| Sentiment | Positive, neutral, or negative | A mention with a warning attached can hurt |
| Competitors named | Who else appeared | Your share of voice against them |
| Sources cited | The links shown, if the answer searched the web | Tells you where ChatGPT got its information |
The last column is the most useful one. When ChatGPT searches the web, the sources it cites tell you which pages shape the answer. If a competitor's comparison page or a Reddit thread keeps appearing, that's where you need to show up.
Step 4: calculate a simple visibility rate
For each prompt, divide the runs where you were mentioned by total runs. Average that across the prompt set and you have a visibility rate. Do the same for two or three competitors. Now you can say "we're mentioned in about a third of category answers, our main rival in about half," which is a sentence your leadership can act on.
Repeat the exercise monthly. A single baseline is a snapshot. The trend is what tells you whether your work is moving anything.
Step 5: know the limits of the manual method
Do this for a month and three problems show up.
It doesn't scale. Thirty prompts, three runs each, is ninety checks. Add a second market or a weekly cadence and someone's whole week disappears.
It can't explain movement. If you drop out of an answer, your spreadsheet shows the drop. It can't show whether ChatGPT's crawler stopped reading your page, or whether a new source replaced yours.
And it only covers ChatGPT. Buyers also ask Gemini, Perplexity, Claude, Copilot, and Google's AI Overviews, and answers differ between them.
Free one-shot checkers sit in between. HubSpot AI Search Grader gives an instant score across ChatGPT, Perplexity, and Gemini without an account, but our listing notes it has no history, no custom prompts, and scores that vary between runs. Good for a first look, not for tracking.
Step 6: automate the sampling
This is where a tracker earns its fee. The workflow above maps directly onto Promptwatch, which is the tool we'd use for it.
Start free. Explore tracks 10 prompts on ChatGPT, which is enough to load your best category prompts and see the daily pattern. Moving to Essential ($95/mo, 7-day trial) gives you 50 prompts across ChatGPT and every other engine Promptwatch monitors, with real UI monitoring rather than API calls alone, so you see what users see.
Each field from step 3 has a counterpart:
- Mentioned, position, competitors. Prompt tracking with share-of-voice and competitive benchmarking, plus prompt trends that show what changed between checks.
- Sentiment and framing. Sentiment analysis across answers.
- Sources cited. Citation analytics at page and domain level, including Reddit threads, YouTube videos, and offsite mentions, with citation trends over time.
- Which prompts to track. Search volumes, difficulty scores, and query fan-outs per prompt, so you can see which questions get asked and the sub-searches ChatGPT runs to answer them.
- Country differences. Country-level tracking on every paid plan; state and city tracking from Business.
Step 7: connect mentions to the rest of the funnel
A mention count is where most trackers stop. Two more layers turn it into something you can act on.
Agent Analytics, from the Professional plan ($245/mo, 25M crawler logs), reads your CDN logs and shows which pages ChatGPTBot fetched, how often, and with what errors, plus the path from crawl to citation. When a mention disappears and the crawler stopped visiting the page that used to be cited, the fix is technical, not editorial.
Visitor analytics, from Essential, adds a lightweight script or GTM template that records AI-referred sessions and conversions. That answers the question that follows every visibility report: did those mentions send anyone?
If you sell products, ChatGPT Shopping tracking shows how you appear in shopping results, and Ads Radar (Business and every agency plan) captures sponsored ads that appear inside ChatGPT Search answers to your tracked prompts.
A realistic weekly routine
Once tracking runs on its own, the job becomes a short weekly review. Check which prompts moved. For any lost mention, look at the citations that replaced you and the crawler logs for the page that used to be cited. Decide whether the fix is a page update, a technical fix, or showing up in a source ChatGPT trusts. Then check visitor analytics a few weeks later to see whether it paid off.
For more tools that track ChatGPT, see the full ranking.
FAQ
Can I track ChatGPT mentions for free?
By hand, yes, using the method above. Promptwatch's free Explore tier automates it for 10 prompts on ChatGPT.
Why does ChatGPT give different answers to the same question?
Answers are generated each time and can vary with web search results, memory, and location. That's why tracking means repeated sampling, not a single check.
Does ChatGPT mention brands it hasn't crawled?
It can mention brands from training data or from other sites that discuss them. But if you want your own pages cited, its crawler needs to be able to read them, which crawler logs confirm.