How do you track brand mentions in ChatGPT, Claude and Grok answers?
Published . Updated .
To track brand mentions in ChatGPT, Claude and Grok, ask each assistant the same set of buyer questions on a schedule and record, for every answer, whether it named your brand, cited your site, read your pages or searched for you by name. Keep those four counts apart. When we put one question to Claude and Grok a day apart, two centres that Grok recommended and cited turned up in Claude only on a page Claude read, and four of the centres Claude named were named to rule them out.
This is the manual check we use, and it is one step in a wider GEO workflow built on fan-out data. The example is real. On 10 and 11 September 2026 we asked Claude (Opus 5) and Grok (Grok 4) to search the web and compare secondary maths tuition centres in Jurong East, Singapore, recorded both turns with Only Fan Out, and then scored the 15 centres that turned up in either turn with the extension's own watchlist matcher. It shows what one account saw on two days.
What counts as a brand mention in an AI answer?
Your brand can turn up in four places in one turn, and each tool counts a different one. The answer text is what a person reads. The other three sit behind it, in the searches the assistant ran and the pages it read and cited.
| Where your brand shows up | What it tells you | Where you can see it |
|---|---|---|
| Named in the answer | The assistant put you in front of the reader, as a pick, a listing or a warning | The reply itself |
| Cited | A sentence in the answer rests on one of your pages | Inline links and the Sources list in ChatGPT, citations in Claude and Grok |
| Searched by name | The assistant knew to look for you before it answered | Only in the search data, such as the Only Fan Out side panel |
| Fetched, not cited | The assistant read your page and no sentence leant on it | Only in the search data |
Ahrefs counts the first kind. Its list of the 100 most mentioned brands in ChatGPT, published on 11 November 2026, counts an answer once if it names a brand at all, across 5,516,381 US answers, and keeps cited pages in a separate report. A US beta of a ChatGPT page for businesses reportedly shows them their mentions and clicks and groups the conversations that cited their website into topics, according to Search Engine Roundtable on 10 November 2026. We could not read OpenAI's own page on it, so treat those details as reported. Neither shows the searches behind an answer, or the pages the assistant read and left out.
What did the same question show in Claude and Grok?
Each assistant recommended a different set of centres, and Zenith Education Studio was the only one both recommended. Claude named 10 of the 15 centres and cited one centre's site. Grok named 7 and cited the sites of 6.
| Centre | Claude, 10 Sep 2026 | Grok, 11 Sep 2026 |
|---|---|---|
| Zenith Education Studio | Recommended, not cited | Recommended, cited, searched by name |
| Tim Gan Math | Named to rule out (a JC specialist) | Recommended first, two pages cited |
| Matrix Math | Named to rule out (mostly primary) | Recommended, cited, searched by name |
| AGrader | On a page Claude read, not named | Recommended, cited, searched by name |
| Indigo | On a page Claude read, not named | Recommended, cited |
| Math Academy | Absent | Named as a nearby alternative, cited |
| EduFirst | On a page Claude read, not named | Listed in its comparison table, not cited |
| The Math Lab | Absent | Searched by name, two pages read, not named |
| UNBOXED Math | Recommended, not cited | On a page Grok read, not named |
| Master Maths | Recommended, its price page cited | Absent |
| Raymond's Math & Science Studio | Recommended first, not cited | Absent |
| Tuition with Arthur Lee | Recommended, not cited | Absent |
| Kiasu Maths | Named as a budget backup, not cited | Absent |
| Achevas | Named to rule out (a JC specialist) | Absent |
| Trinity Learning | Named to rule out (mostly primary) | Absent |
A tracker that only asked "was the brand named?" would score Tim Gan Math as a hit in both answers. Claude named it to tell the reader to look elsewhere.
Why is a mention not always good news?
Because an assistant names brands to warn people off as well as to recommend them. Four of the 10 centres Claude named, Tim Gan Math, Achevas, Matrix Math and Trinity Learning, were named in its first paragraph as poor fits for a secondary student. A count of names scores all four as wins, so log the tone of every mention: recommended, listed or warned against.
Cited and recommended are separate lists too. Claude cited three pages, two fee guides on tutorly.sg and Master Maths' page on tuition costs, for the price ranges near the end of its answer. The other five centres it recommended had no citation at all. They were the pins on its map, and Claude does not cite the places it plots, which our help page on seeing the places Claude used covers. Grok went the other way, and all seven of its citations were the centres' own pages.
A search by name is no promise of a place in the answer either. Grok ran a search for The Math Lab Jurong East tuition, read two pages that mention the centre, and left it out. Without the search data you would score The Math Lab as absent and never learn that Grok had looked for it. Our post on fetched versus cited sources covers why assistants read more pages than they cite.
How do you track your brand in ChatGPT, Claude and Grok by hand?
Run a fixed set of questions in each assistant on a schedule, record every turn, and log the four places separately.
- Write 20 to 30 questions a buyer would ask before choosing a business like yours, in their own words: best-of, comparison, price and problem questions. Keep the wording fixed from one round to the next.
- In Only Fan Out's Settings, add your brand to the Watchlist with its other spellings as aliases, and add your domain as a second entry. The help page on setting up the watchlist has the steps.
- Start a fresh chat for each question, with saved memory off where the assistant has it, so earlier chats do not steer the answer. In Claude, web search is a switch under the + button in the chat box, as Anthropic's web search help page describes. ChatGPT searches by itself when a question might benefit from the web, according to OpenAI's ChatGPT search help page.
- Send each question with the side panel open. ChatGPT has only sent its search text while the answer streams since 14 September 2026, so a panel opened afterwards can miss it.
- Read the Watchlist section at the top of Live for each entry's status and counts, then read the answer yourself and note the tone of each mention.
- Export the conversation as CSV. Each row carries the watchlist terms it hit, so the file drops straight into a tracking sheet. Our guide to exporting ChatGPT search queries and sources to CSV has the sheet layout and formulas.
- Ask each question twice per round, and repeat the round every week or fortnight. Answers to the same question change between runs, so judge the trend over several rounds.
Why add your domain as well as your brand name?
A brand name entry can miss a citation of your own page. Grok cited Zenith Education Studio's Jurong East page, but that page's title never says Zenith and its address runs the words together as zenitheducationstudio.com, so the brand entry marked Zenith as only mentioned. A domain entry for zenitheducationstudio.com marked it as cited. Claude's citation of Master Maths went the same way, missed by the brand entry and caught by a domain entry for mastermaths.com.sg.
What should a brand tracking sheet hold?
One row per answer, with a column for each place your brand can show up and one for tone.
| Column | What goes in it |
|---|---|
| Date, assistant and model | For example 11 Sep 2026, Grok, grok-4 |
| Question | Exactly as sent |
| Named | Yes or no, and whether you came first, in a table or further down |
| Tone | Recommended, listed or warned against |
| Cited | Which of your pages, if any |
| Searched by name | The search line that named you |
| Fetched, not cited | Your pages the assistant read and left out |
| Others named | The other brands in the answer, with their tone |
To turn a few rounds of this into a review of your whole site, our step-by-step GEO audit scores each topic by reach, cite rate and mentions with no page behind them.
What can manual tracking not tell you?
It tells you what one account saw on the days you looked. OpenAI says ChatGPT uses general location from your IP address to improve search results, so someone in another city may see other names, and nothing in a recorded turn shows how OpenAI, Anthropic or xAI choose sources inside the model.
Only Fan Out scores the answers you capture yourself, one conversation at a time, and keeps them in your browser. It has no rank tracking, alerts or dashboards, it does not run prompts for you over time, and it does not change rankings or citations.
Related questions
Is a brand mention in ChatGPT the same as a citation?
No. A mention is your brand named in the answer, and a citation is a link to one of your pages that a sentence in the answer rests on. In our Claude test, five of the six recommended centres were named with no citation.
How often should you check brand mentions in AI answers?
Every week or fortnight, with the same questions each time and each question asked twice. Answers to one question change between runs, so a single check proves little.
Does Only Fan Out track brand mentions automatically?
No. It scores the ChatGPT, Claude and Grok answers you capture yourself against your watchlist, in your browser. It has no alerts, dashboards or scheduled runs.
See the fan-out for your own prompts
Only Fan Out is a free Chrome side panel that lists the queries, fetched pages and cited pages behind each ChatGPT, Claude or Grok answer. Nothing leaves your browser.
Add to Chrome, it's free