Documentation
How measurement works
Which models, how often, and the rules that decide what a number is allowed to say.
AI gives a different answer every time you ask. Measurement that ignores that produces a trend line made of noise, so every rule below exists to stop that.
The models
Every plan includes ChatGPT, Google AI Overview and Google AI Mode, with ChatGPT read on the app buyers open. Perplexity, Gemini and Microsoft Copilot can be added per prompt, on request, and so can the developer API surfaces, labeled as such. Grok is enterprise only.
The app and the developer API are different systems. Asked the same prompts inside the same minute, the ChatGPT API and chatgpt.com shared one domain of seven on the first prompt and, on the second, only Trustpilot and Reddit with no recommended service in common. That is three prompts on one day, published with the limit on it, so we report no rate for it. Developer numbers are shown, labeled, and kept out of your headline. Nobody can measure claude.ai, including us, because it sits behind a login, so any Claude percentage anywhere is an API number.
The cadence
Breadth before repeats: a prompt is asked once per sweep on each model, and the count grows round by round.
- Every tracked prompt is asked once a week on every model in your plan. The answers add up week on week, which is why a rate for one prompt waits for eight answers before it is shown as a percentage.
- A client's next round can be pulled forward once per round with Refresh now, and the round after it follows a week later.
The three rates
Mentioned, cited and recommended are separate truths and never blended into one score. Mentioned means the answer says your name. Cited means the answer cites your site. Recommended means the answer picks you, and it is only graded when the answer contains a verbatim sentence mentioning you.
A failed engine call is a failed run. It shrinks the count of answers read and is never scored as your absence.