How to check whether ChatGPT mentions your brand
The short answer
Write eight to ten questions a buyer would ask to find a business like yours — never your brand name, which always returns you and measures nothing. Run them in a logged-out or temporary chat so personalisation does not bias the result. Score two things separately: whether your brand is named in the text, and whether your domain appears as a cited source. They are different outcomes with different fixes. Repeat the identical prompt set monthly, because a single run is an anecdote.
- Never test your own brand name — it always returns you.
- Use a logged-out or temporary chat; memory and personalisation will flatter you.
- Named in the answer and cited as a source are different problems.
- Score at least eight answers per engine before trusting the number.
Step 1 — write the prompt set
Eight to ten questions, in the words a buyer would use. The test of a good prompt is that someone who has never heard of you could plausibly type it.
Good prompts look like:
- “Who are the best providers of [what you do] in [your city]?”
- “I need [what you do]. What are my options and what should I watch out for?”
- “Compare the top providers of [what you do] on price and service.”
- “Which provider of [what you do] is best for a small business?”
- “Recommend a reputable [what you do] company and cite your sources.”
Include two or three that do name you — “Is [brand] any good?”, “What are the alternatives to [brand]?” — because how the assistant characterises you, and who it lists as your alternatives, is useful. Just do not count those in your visibility score.
Step 2 — run them clean
Use a temporary chat, a logged-out session or a private window. This matters more than people expect. Personalisation and chat memory bias answers toward things you have discussed, and if you have been discussing your own company all week you will get a flattering, useless result.
Test across ChatGPT, Perplexity, Google AI Mode and Gemini. They disagree with each other far more than most people assume.
Step 3 — score mention and citation separately
Record two independent facts per answer:
| Outcome | What it means | What to fix |
|---|---|---|
| Named and cited | Best case — recommended, and your site is treated as a source | Protect it; citation sets churn |
| Named, not cited | You are recommended on someone else’s authority — fragile | Widen the number of third-party sources that mention you |
| Cited, not named | Useful content, weak brand entity | Entity work, not more blog posts |
| Neither, competitors present | A genuine gap | Find where those competitors are quoted from; that list is the target |
The “named but not cited” case is the one most people misread as success. It means a single third-party source is carrying your visibility. If that source updates its list, you vanish in a week.
Step 4 — repeat identically
Same prompts, same method, monthly. The value is entirely in the comparison, and it only works if nothing about the method changed. Write the prompt set down in a file and do not improve it, however tempting.
Doing it without the spreadsheet
We built a free tool that does the bookkeeping: it generates the prompt set from your category, then scores answers you paste in, tracking mention and citation separately. It runs entirely in your browser — nothing is sent anywhere — which is also why it is free.
It deliberately does not query the APIs for you. Tools that do are measuring the API, not the product your customers use; ChatGPT with browsing, logged in, on the web retrieves differently. Pasting the real answer is slower and more accurate.
Questions people actually ask
Why does asking about my own brand not work?
Because the assistant will find and describe you, which proves only that you have a website. The question that matters commercially is what happens when someone who has never heard of you asks for a recommendation in your category. That is the query your competitors are winning or losing.
Why does being logged out matter so much?
Chat memory and personalisation bias results toward things you have discussed before — which, if you are testing your own company, is your own company. People routinely check from their own logged-in account, see themselves recommended, and conclude they have no problem. Use a temporary chat.
I got different answers running the same prompt twice. Is the test broken?
No, that is the system working as designed. AI answers are non-deterministic and retrieval varies between runs. It is precisely why one run tells you nothing and a fixed prompt set run repeatedly on a schedule tells you something. Treat any single answer as one sample, not a result.
Which engines should I test?
ChatGPT, Perplexity, Google AI Overviews and AI Mode, and Gemini at minimum. They retrieve from different sources and disagree substantially — when four engines were asked to name the best AEO agencies in Australia they produced 24 different firms with almost no overlap. Testing one engine and generalising is a common and expensive mistake.