- Engines
- We measure ChatGPT through OpenAI's developer API with web search enabled; these answers may not match exactly what people see in the ChatGPT app. Google AI Mode and AI Overviews are captures of Google's results page, collected with the city and language we send. Gemini measurement is currently paused. The table below lists each system's method and status.
- Questions
- Real question patterns about you: brand, branch, service and “near me” questions. The question set is fixed; when it changes we record the version and compare before/after on the same set.
- Repetition
- Each question is asked several times on each engine; AI answers can change every time.
- Measurement cycle
- Google AI Mode and AI Overviews are measured every week. ChatGPT questions rotate: part of the set each week, every question at least every four weeks (Gemini will follow the same cycle once enabled). Week-over-week drop alerts are calculated only on a fixed set of the same questions each week (Google AI Mode). A question that wasn't measured is not counted as “not mentioned”; rates use measured answers only.
- Ranges
- We report rates with a 95% confidence range, not a single number. With few answers the range is wide, and we say so.
- Confirmed and Likely
- A claim that clearly contradicts an approved fact is “Confirmed”; one that is incomplete or unclear is “Likely”. We don't present what we're unsure of as a finding.
- Approval gate
- We extract your information from your website; nothing is measured until you approve it.
- Seasonal and on-demand services
- Services offered only in certain periods or on demand are not marked “not offered”; if AI mentions them without their period, we flag that separately.
- Fix cycle
- When you mark an issue as fixed, we ask the same questions again and report the time to fix and whether it reappears.
- What we can't measure
- Answers personalised to someone's account; how often real users ask a question; what AI will say next. Results are a summary of the answers we measured, not a forecast.