How we measure what AI recommends.

Each category has a shared library of 30 to 40 questions buyers ask: what is, X vs Y, and best for. Recommendation questions never name a brand. You pick the topics you compete in, so you're only scored on what you sell.

New categories are generated on first grade, checked, and reused. Paid plans add your own questions.

Apps answer differently from their APIs, so consumer surfaces are captured from the app. Coding assistants call APIs, so that's how the coding lane is asked.

Brand names and aliases are matched in the answer text, not in links. Order gives position; cited sources are kept. A light model pass catches brands we didn't know about.

Questions that name you are reported apart from the headline, because they inflate it. Identical questions share one answer for a few days, so nobody pays twice.

A rate from 40 answers is an estimate, so every rate carries a 95% range (Wilson). A change is called real only when the ranges stop overlapping, on questions and engines measured both times. Everything else is labelled noise.

Answers vary between asks and shift as models update. We don't promise rankings or traffic. Recommendations are things we observed or can verify, not guarantees.