How to track brand mentions in ChatGPT

ChatGPT is recommending businesses to buyers every day, by name, with reasons. If you are in those answers, you want to know your position and how you are described. If you are not, you want to know who is being named instead. Both are trackable. Here is the manual method, done properly, and an honest account of where it stops scaling.

What “tracking brand mentions” actually means in ChatGPT

Tracking mentions in a chat engine is different from tracking rankings in Google. There is no public index to check and no fixed result page. What you are really measuring is the answer to four questions, sampled over time:

Presence - when a buyer asks a question in your category, does ChatGPT name you at all?
Position - when the answer is a list, where do you sit? Position one in an AI answer takes most of the attention, just as it does in search.
Sentiment - how are you described? “Reliable but premium-priced” quietly disqualifies you from every budget-conscious buyer, and you will never see it happen unless you read the answers.
Substitution - when you are absent, which competitors got the recommendation you wanted?

One casual “ask ChatGPT about us” check measures none of these reliably, because answers vary with phrasing, session history and whether browsing is on. Tracking means asking a fixed question set the same way, on a schedule, and recording results.

The manual method

Build a question set of 10 to 25 decision-stage questions. These are the questions buyers ask before choosing: “best [category] for [use case]”, “[competitor] alternatives”, “[category] in [city]”. Deliberately exclude questions that contain your brand name; asking “tell me about [you]” tests whether ChatGPT knows you exist, not whether it recommends you when it matters. Include the wording variants your customers really use.
Ask them in a clean session, consistently. Use a fresh chat each time so earlier conversation cannot steer the answer, and note whether Search (browsing) is on or off; the two modes answer from different layers and both matter. Keep the mode consistent between runs, or record both.
Record structured results in a spreadsheet. One row per question per run: date, cited yes/no, position if listed, the sentiment of the description (positive, mixed, negative), the exact quote, and every competitor named. The quote matters: three months later, “what changed” arguments are settled by wording, not memory.
Repeat weekly. ChatGPT’s answers move as the web changes, as its retrieval re-ranks, and as models update. Weekly sampling is frequent enough to catch a competitor displacing you or a sentiment shift while it is still fresh; monthly is usually too slow to see what caused a change.
Track your aliases too. If your business is cited under an acronym, a trading name or a common misspelling, a mention can slip past a strict brand-name check. Search your recorded quotes for every name you actually go by.

Done properly, this works. It is also genuinely time-consuming: a 25-question set, asked carefully with results recorded, takes a couple of hours, and that is one engine. Buyers do not only ask ChatGPT: the same questions go to Claude, Gemini, Perplexity, Grok and DeepSeek, and the engines frequently disagree about you. Manual tracking across six engines runs to four or more hours a week, every week, which is exactly the point at which most teams quietly stop doing it, and the data gap begins.

What to do with the data

Tracking is only worth the effort if it drives changes:

Absent everywhere? That is a visibility problem, not a monitoring problem. Run a free SearchScore audit to find what is blocking you (crawler access, entity clarity, citable structure), fix it, and keep tracking to confirm the fixes land.
Cited but described wrongly or negatively? Work the correction levers: your own site’s stated facts, structured data, and the third-party pages the engine is reading. See how to fix wrong AI information.
A competitor consistently ahead of you? Read the citations in their answers: the third-party pages carrying them (roundups, directories, review sites) are your outreach list.
A sudden drop? Check what changed on your side first (site changes, robots.txt, redirects), then look for a model or engine update; the timing in your log usually points at the cause.

When automation earns its keep

The manual method’s weakness is not accuracy, it is sustainability. The SearchScore Tracker automates exactly the process above: it asks the questions your customers ask across six engines (ChatGPT, Claude, Gemini, DeepSeek, Grok and Perplexity) every week, records which engines named you, your position, per-citation sentiment and the full response text, and surfaces which competitors were recommended in your place. Brand-alias matching catches citations under up to four trading names or acronyms, and if the AI hallucinates or a citation does not qualify, you can flag it and your score recomputes to your adjusted view. Everything rolls into a single AI Visibility Score built from decision-stage questions only, and a weekly digest lands the changes in your inbox. Starter is £20/mo for 25 tracked questions with weekly scans; Pro is £79/mo for 100 questions scanned daily.

Whether you automate or not, the sequence is the same: fixed decision-stage question set, consistent method, structured records, weekly cadence. The tool just removes the four hours.

Frequently asked questions

How do I check if ChatGPT mentions my brand?

Ask ChatGPT the decision-stage questions your buyers ask, in a fresh session, and record whether you are named, your position in any list, how you are described, and who is named instead. Avoid questions containing your own brand name: they test recognition, not recommendation. Repeat the same set weekly, because answers change.

Why do ChatGPT’s answers about my brand keep changing?

Answers draw on live web retrieval and the model’s training memory, and both move: your pages and competitors’ pages change, ChatGPT’s retrieval re-ranks, and model updates reshape what is remembered. Phrasing and session history add run-to-run variation on top. This volatility is why one-off checks mislead and scheduled tracking is the reliable measure.

Can Google Analytics show ChatGPT brand mentions?

No. GA4 can show visits referred from chatgpt.com, which happen after a citation has already been won, and shows nothing when ChatGPT recommends a competitor instead. Mentions have to be measured at the engine itself, by asking questions and recording the answers, manually or with a tracking tool.

Is there a tool that tracks brand mentions in ChatGPT automatically?

Yes, several exist. SearchScore Tracker runs your question set weekly across ChatGPT, Claude, Gemini, DeepSeek, Grok and Perplexity, and records citations with position, sentiment and the competitors named in your place, from £20/mo. Whatever tool you choose, check it tracks decision-stage questions and shows you the exact responses, not just a count.

Continue Reading

Part of AI Search — see all guides in this series →