How we measure share of AI voice
Our studies start with a simple question: when buyers ask, how often does each brand actually get named? Here is the method.
Published 17 Jun 2026
Most "AI visibility" numbers fall apart under scrutiny because they never define what they counted. Our studies start by fixing that.
For a given market, we fan out the questions buyers actually ask into a set of prompts, then run them against the live interfaces of ChatGPT, Google AI Mode, Gemini and Perplexity. We capture the real answers, not API responses, because the interface and the API can return different results for the same question.
The metric and the denominator
Share of AI voice is the share of relevant answers in which a brand is named, measured against a fixed denominator: the total set of prompts run. That denominator is the part most reports skip, and the part that makes a number defensible.
We then validate mentions to strip out false positives, normalise so brands with different prompt volumes can be compared, and only report a ranking once it clears a minimum base. Every study keeps an audit log so the result can be reproduced.
It is deliberately unglamorous. But it is the difference between a headline a journalist can trust and one they will quietly bin.