How we measure — and when we admit a number is worthless.
Nobody can hand you a perfect number for your AI visibility. So we show you how sure a reading is, and we tell you when not to trust it. This page is the whole method, in plain words.
AI answers are your most important channel, and the least measured one.
The first thing many of your buyers now read about you is an AI’s answer — not your site, not a review. It is also the channel almost nobody measures honestly. Most tools hand you a tidy figure and hope you don’t ask how they got it. We measure it, and we show our work: what a reading is, what it can tell you, and exactly where it can’t. The honesty here is structural, not a slogan — the rest of this page is just us keeping that promise.
One reading is one night. The picture settles over many.
A reading = one nightly pass over your questions, every engine.
Asking an AI once is a snapshot, not a measurement — the same question can come back different tomorrow. So every night we put the same set of your questions to all seven engines and record what each one says. That is one reading. A single night can wobble; the trustworthy picture is the one that holds steady across several.
- ChatGPT
- Perplexity
- AI Overviews
- Gemini
- Claude
- Grok
- DeepSeek
Nightly is deliberate. It is often enough to catch a real move within a day, and slow enough that one strange night never gets mistaken for a trend.
The one number worth watching, computed in the open.
Share of Answer = how often AI answers name you when your buyers ask.
In plain words: across your tracked questions, per engine, we count the share of answers that actually name your brand. That’s it. No weighting you can’t see, no secret formula.
What it is not is just as important. It is not website traffic. It is not a search rank. And it is not a blended score. We never blend metrics into a 0–100 composite — a single invented score hides more than it shows. If you can’t take a number apart and see where it came from, it isn’t a measurement; it’s a mood.
Every number carries how sure we are of it.
Confidence 0–1 = how settled this number is across readings.
Some nights an engine simply answers differently — same question, different mood. If a number only moves inside its usual spread, that’s noise, not news, and we don’t raise an alarm on it. But when the spread is wider than we’d expect, we flag the reading and tell you plainly why, right where the number lives:
When a reading isn’t reliable, we tell you. A reading we don’t trust ourselves gets a mark, not a nicely rounded figure that pretends the doubt away.
Where the answer came from changes what you can do about it.
We separate two things most tools mash into one: what an engine already knows about you from its training, and what it fetches from the live web at answer time. Every reading is badged one way or the other (“grounding”, if you speak jargon).
The difference is practical, not academic. If an answer comes from memory, a fix you ship today can’t show up until the model itself is retrained — patience, not panic. If it comes from the live web, updating your own pages can move the very next reading. Same low number, two completely different next steps — which is why we never hide the badge.
We show timing, not proof of cause.
When a number moves after you shipped something, we show it as exactly that: a correlation with your change — “rose after,” “matches the week you shipped.” We never dress it up as causation, and we never tell you a number moved because of your change — we can’t see inside the models, and neither can anyone else.
“Proven” is something you earn, not something we assert. It only appears after you mark a brief as shipped and at least two later readings move in the direction you expected. Until then it’s timing on a chart — honest, useful, and clearly labelled as timing.
The short list of things you’ll never see from us.
- No invented numbers. Every figure traces back to a reading you can open.
- No predicted “impact” scores. We measure what happened, not a fortune we made up.
- No blended composites. One metric, one name, taken apart on request.
- No countdowns or fake scarcity. Your visibility isn’t a flash sale.
- No success claims without a change you confirmed. See section six.
And a small structural promise that keeps us honest: every number in every email we send deep-links straight to the surface that shows it — so you can always check us against the reading itself.