How do you track brand mentions in ChatGPT?
ChatGPT has no dashboard that tells a brand how often it is named. You track it the way a survey works: ask the same 20 buyer questions on a schedule, ask each one at least 2 times in a clean session, save every answer and source, and count how often your brand is mentioned, recommended and cited compared with your competitors. Done by hand this takes about 4 hours a month for 3 engines. A tracking tool is worth paying for when you need daily data or more questions than that.
There is no position 1 in ChatGPT, and no two users see the same answer.
Answers vary between runs
The same question can name different vendors on a second try. One answer is a sample, not a ranking.
Context changes the search
OpenAI says ChatGPT may use saved memories and an approximate location from the IP address when it turns a question into search queries. Your signed-in answer is not your buyer's answer.
Named is not chosen
A brand can be mentioned in passing while a competitor is the one recommended. Track mentions and recommendations separately.
Track brand mentions in ChatGPT in seven steps
- 01
Write 20 buyer questions.
Use the words your buyers type, not your positioning. A useful mix: 10 category questions ("best [category] for [segment]"), 5 comparison or alternatives questions ("[competitor] alternatives"), 3 use-case questions and 2 branded questions ("what is [your brand]"). Mark each one branded or unbranded.
- 02
Use a clean session and write down the conditions.
Sign out, or use a Temporary Chat, so saved memories do not shape the answer. Record the date, the model shown, your country and whether web search ran. OpenAI says ChatGPT may use saved memories and your approximate location when it rewrites a question into search queries.
- 03
Ask every question at least twice.
Answers change between runs. Two runs per question is the minimum that tells a pattern from a one-off; three is better for the questions you care about most.
- 04
Save the whole answer and every source.
Copy the answer text and the URLs under Sources into the worksheet. A screenshot is hard to count later, and a score without the answer cannot be checked.
- 05
Classify each answer the same way.
Mentioned: your brand is named. Recommended: it is presented as a good choice for the need in the question. Cited: a page on your domain is a source. Also note which competitors were recommended and in what order.
- 06
Count over answers, not questions.
Divide by the answers you collected. Mention rate is answers that name you ÷ all answers. Recommendation rate is answers that recommend you ÷ all answers. Do the same for each competitor.
- 07
Rerun the same set every month.
Keep the questions, engines and competitors fixed so this month's number can be compared with last month's. Add new questions as a separate group instead of mixing them in.
The free AI visibility audit template is a CSV with a column for each field above: question, engine, date, run, mentioned, recommended, position, citations and competitors. Use the same steps for Gemini, Perplexity and Google AI Overviews.
20 questions × 2 runs = 40 answers to count.
| What you count | Answers | Rate |
|---|---|---|
| Your brand mentioned | 14 of 40 | 35% |
| Your brand recommended | 6 of 40 | 15% |
| Competitor A recommended | 22 of 40 | 55% |
The brand is named in 35% of answers but recommended in only 15%, while Competitor A is recommended in 55%. That gap, not the mention rate, is the finding. The next question is why; the guide on why ChatGPT recommends your competitor lists the usual reasons and how to check each. These numbers are made up to show the arithmetic; they are not a RecoProof customer result.
Five ways manual tracking gives the wrong answer.
Asking in your own signed-in account.
Your memories and past chats can change what you see. Your buyers see something else.
Counting a mention as a win.
Being named in a list of ten is not the same as being recommended for the buyer's need.
Changing the questions every month.
The trend becomes meaningless. Freeze the set and add new questions separately.
Asking only branded questions.
"What is [brand]?" always names you. Buyers who do not know you yet ask unbranded questions.
Judging from one run.
One answer can be a different draw. Repeat before you act on it.
Pay for tracking when the spreadsheet stops scaling.
- You need daily data, not a monthly snapshot.
- You track more than about 50 questions or more than 3 engines.
- Several people need the same dashboard and alerts.
- You report AI visibility to leadership every week.
Tracking 50 prompts on 3 engines every day starts at about $90 a month with the tools we priced. The AI visibility tool pricing guide compares seven of them per prompt.
Tracking tells you that you lose. It does not tell you why.
If you already know a competitor is recommended instead of you, a daily count adds little. RecoProof's Recommendation Audit asks 30 approved buyer questions on 3 engines, keeps every answer and source, compares three competitors and turns the gaps into a 90-day plan. It costs $499 once at the founding price. RecoProof sells this audit, so weigh this paragraph accordingly. To see one real buyer question first, run the free AI visibility check.
What the ChatGPT claims on this page rest on.
Checked 2026-10-29. OpenAI changes ChatGPT often; the method above does not depend on any one feature.
- OpenAI Help Center: Searching the web with ChatGPT (query rewriting, memories, approximate location, citations)
- OpenAI Help Center: Ads in ChatGPT (Temporary Chats do not create or update memories)
- RecoProof methodology (how mentions, recommendations and citations are classified)