How we measure
The methodology is public: what we observe, how we calculate and what we do not claim. Every number in the report traces back to the answer it came from.
Three separate parts
A · the website audit (only what we verified). B · observing AI answers. C · recommendations derived from evidence. Observations and hypotheses are never mixed.
Same questions, same market
Every AI receives exactly the same questions, in the same language and market, in independent sessions. We never put your name or site content into discovery questions, and never ask the model to recommend you.
Mention ≠ citation
A mention means the brand name appears in the answer. A citation means your domain is among the answer’s sources. We report them separately.
Observed position, not a ranking
Position exists only when the answer contains a numbered list. It is the position in that answer, not a universal ranking. Otherwise we show N/A.
Numerator, denominator, interval
Every rate shows n (e.g. 12/26) and a 95% Wilson interval. Under 5 observations we say “sample too small” instead of a percentage.
Errors are not “absence”
A missing answer, timeout or provider error is never counted as “brand absent”. The channel is flagged and credits for the undelivered part are returned.
No silent substitutions
If an AI is unavailable we never quietly swap in another. The exact setup (model, search, date) appears in the report.
API, not the chat app
We measure through official APIs. The apps people use may have a different model, history or personalisation, so results can differ.
What we do not promise
We do not guarantee appearing in AI. Schema markup, FAQs or llms.txt are not guarantees of inclusion. Recommendations carry an evidence level and a verification method.