How we measure
We will never give you a single AI score. A score hides which assistant, which question and how many answers it came from. We show you all three.
The questions your buyers ask
Every question set has at least 15 questions, and 15–31 for a typical business. We write them the way your buyers type them, and you see every one next to its answers.
- “Who's a good family dentist in Lakewood, Ohio that's taking new patients?”
- “Dentists in Lakewood OH with evening hours”
- “Gentle dentist for anxious patients in Lakewood”
- “Which Lakewood dentists take Delta Dental?”
- “Lakewood dentist that does implants”
Example set for a family dental practice. 5 of 15 shown.
Four assistants, asked separately
ChatGPT, Claude, Gemini and Perplexity each get every question in a fresh session. Every transcript records the assistant, the model name and the run date, so a rate can always be traced to the answers behind it.
The model each assistant is asked with today. The model that actually answered is recorded on every transcript.
Two assistants, or it isn't reported
Nothing is reported unless two assistants independently confirm it. One assistant naming you once could be noise. Two separately is a finding.
The rule covers what we report as a finding: where you stand against a competitor, an accuracy problem in what assistants say about you, and a change from one run to the next. Each needs two or more assistants. A single assistant's answer is still shown, in its transcript and in that assistant's own count.
Every rate shows its n
A rate without its denominator can't be checked. We always write both numbers, and say where they came from.
Example figures.
What we don't claim
We say what we measured and stop there.
- We don't claim a fix caused a change. We measure again and show what moved.
- We don't claim to know how an assistant ranks businesses internally.
- We don't promise you'll “rank #1 in AI”. Nobody can.
- We don't blend assistants into one number.