Chats
The AI responses every figure in the product is measured from.
A chat is one model’s reply to one tracked prompt, on one day, from one location. Everything else in Recomma is counted off these: visibility is the share of chats naming you, a citation is a link inside one, a gap is a source cited by chats that never name you.
They are kept whole, not reduced to a score, which is the point. A number you cannot trace is not evidence — so when a figure surprises you, this is the page that says why.
What a chat records
| Prompt | The question that was asked |
| Model | Which surface answered — ChatGPT, Perplexity, Gemini |
| Status | Whether it came back, and the error if it did not |
| Fidelity | Interface or API — see models |
| Location | The market it was asked from |
| Brands mentioned | Who was named, and where in the text |
| Sources | Every page the model opened, cited or not |
Sources and citations are not the same
Every citation is a source; most sources are not citations.
- A source is any page the model reached for while answering.
- A citation is a source it went on to reference in the text.
A model can open a page, read it and quote nothing from it. That is a different
fact from never finding the page at all, and the two are recorded separately
all the way down — which is why Sources carries both
Retrieved and Citation rate rather than one number standing in for both.
Read a chat before acting on a number
A visibility figure that moved is a summary of these. Open two or three of the chats behind it before deciding what changed: it is often the question being read differently rather than the brand’s standing moving.
Failed chats
A model that does not reply produces a failed chat. It is counted, excluded from the metrics — there is no text to count — and retried on the next pass. The Overview reports the window’s total beside the chat count and explains only the failures still failing, so a model that has since recovered drops off the explanation without changing the count.