What is recorded per run
For every prompt run, Ryze captures:- Whether your brand was named in the answer text
- Every URL cited as a source, and its domain
- Which competitor brands appeared
- Where in the answer your brand fell
- The full answer text, so results can be re-examined later
Why prompts, not keywords
Conventional rank tracking checks a keyword and returns a position. That model does not transfer, because there is no ranked list in a generated answer — a page is either drawn in or it is not, and the result changes between runs. So the unit of measurement is a prompt: a full question a buyer would actually type. “best waterproof hiking boots for wide feet” rather than “hiking boots”.Mention vs citation
Both are tracked because they have different causes.
A citation without a mention means your page was useful but your brand was not the recommendation. A mention without a citation means ChatGPT knows you but read nothing — content work will not move that number. See why your brand isn’t cited.
Why results move between runs
ChatGPT decides per-query whether to search, and which results to use. Two identical prompts on the same day can produce different sources. This is expected behaviour, not a tracking fault. Consequences for reading the data:- Never draw a conclusion from one run. A single citation is not a trend, and a single miss is not a regression.
- Rate across many prompts over time is the only reliable signal.
- Compare like with like — the same prompt set, run the same way, against the same baseline.
What this does not capture
Tracking measures what the model returns through the interface Ryze queries. It does not observe:- Answers inside individual users’ private conversations
- Personalised results shaped by a user’s chat history or memory
- Traffic attribution — a citation is not a confirmed visit