Your buyers ask ChatGPT or Doubao before they open anyone’s website, and you have no idea whether those answers contain your name. This guide goes from an empty workspace to the first report. At the end you hold one number and one list: your mention rate, and the names that got mentioned in your place.
A workspace in two sentences
Paste your website, a paper link or just the name into the chat. The AI CMO researches it and hands back a profile for you to check. While checking, pick an identity. It decides four readings: what counts as your own entity, what counts as an alternative, what counts as a question, and what counts as an audience segment:
| Identity | Your entity | Extra in the report |
|---|---|---|
| Company | company, product | none |
| Researcher | paper, method | none |
| Practitioner (clinic, law firm) | the practice itself | official contacts, direct rate |
A brand rule is one sentence, say “never mention price”, and it sticks. The profile lives on the Brand tab and you can edit it later.
Where keywords and phrasings come from
The keyword researcher writes them. Each keyword carries a phrasing (the question a person would type), an audience, a usage scenario and a stage: awareness, comparison or purchase. On the Brand tab under “Topics & Questions” a keyword is one row with its phrasings folded underneath.
Two rules decide what gets asked:
- A keyword counts as confirmed only with search-volume evidence. Where none can be found it is let through and marked “volume unverified”; it does not block collection.
- Phrasings that contain your brand name are never asked. Nobody types “is Sourso good” before they have heard of Sourso, and being named in that answer proves nothing. Those rows are marked “brand name, not collected”.
Who the collectors ask
International platforms are ChatGPT, Gemini and Claude, through their APIs. China platforms are DeepSeek and Doubao. Each keyword goes out with at most three phrasings, so twenty keywords means at most sixty questions per platform per round. Ask the AI CMO in chat to run a collection; after that it runs on its own, Monday 9 am by default.
Reading the first brand analyst report
When the collection finishes the AI CMO walks you through the whole chain in chat. The report itself sits on the Operate tab. On a first read, look at three things:
- Mention rate: the share of answers that named you, split by platform.
- Named instead: in the answers that did not name you, who did appear. This list is more honest than the competitor list you would have written yourself.
- Cited sources: which domains and pages the AIs leaned on. If your website is not among them, that is the first thing to fix.
Below that come strengths, weaknesses and suggestions. The suggestions turn into cards at the end of the chat turn, and tapping a card is a dispatch.
The position line, in one paragraph
At the top of the report there is a “position” block: which tier you are on, who sits beside you, and the nearest cell to move into. Before the first collection it reads “not calibrated”; one collection fills it. How the line is computed and why a tier change needs two collections in a row is in Climb from the tier you are on.
Practitioners get one more number: direct rate
A clinic or a law firm has an extra “official contacts” section on the Brand tab for its phone number and address. The report then adds a direct rate: when an AI gives the user a number to call, is it yours or a listing platform’s. A high mention rate is worth little if the phone number underneath belongs to someone else.
Before you start
- Make sure your website loads; the profile is built from it. For a paper, paste the paper link.
- Pick the right identity when you check the profile. The reports differ by identity, and changing it later recomputes them.
- Do not read too much into the first report. One round a week; look for a trend after two.
Further reading: Climb from the tier you are on · Write for AI answers