How the AI visibility score works

Two scores, one number

The headline is a blend: 75% visibility, 25% site readiness. They are kept separate underneath because they fail for opposite reasons. Visibility is what the engines say about you and is fixed by getting written about; readiness is whether they can read you and is fixed on your own site in an afternoon. If the site cannot be audited at all, the headline is visibility alone rather than a penalty for our fetch failing.

Where the questions come from

From your website, not from a category typed into a form. We read the site, work out what you sell, to whom and where, and write about fifteen questions a buyer would ask an engine. They fall into four buckets: product questions (35%), niche questions (25%), head-term questions (20%) and comparisons with the names buyers already know (20%). The set is pinned, so a rescan asks the same questions and the score is comparable.

What counts as being named

Each question goes to each engine, and every answer is scored between 0 and 1:

• Not named: 0.
• Named, no position in a list: 0.50.
• Named at position r in a list: 0.50 plus 0.50 times 1/r, so first place is 1.0 and fourth is about 0.63.
• Your own domain cited as a source: another 0.15, capped at 1.0.
• Multiplied by sentiment: positive 1.0, neutral 0.9, negative 0.6.

Visibility is the average over every answer that was actually served, times 100. An engine that errored on a question is left out of the average rather than scored zero; a provider outage is not the same as being invisible.

Why an engine can show n/a

If an engine answered fewer than three of the questions, it is not scored: a number from one or two answers would be noise dressed as a measurement. Google AI Overviews is the usual case, because Google simply does not show an overview for many buying questions. The report says n/a and the other engines carry the score.

The thirteen site checks

Site readiness is the weighted share of passed checks, out of 100. The crawler check is worth six times the canonical check because a site-wide block undoes everything else on the list. Three of the checks have free single-check tools on this site.

What the score does not measure

• Sales or traffic. It measures what engines say, which comes before both.
• Sentiment of your buyers. It measures the sentiment of the engines' words about you, which is a different thing.
• Anything outside the questions asked. Fifteen questions across four engines is a sample, and the report says how big.
• Tomorrow. Engines change; the report is dated and rescans are how you watch the number move.

Reading your report

Page one is the pitch: the score, the per-engine numbers, the thirteen checks, share of voice against the brands the engines named, the exact questions a competitor took, the sources the engines trusted, and five fixes in order. Page two is the diagnostics: which engine names whom, why the top rivals get named, and what each engine says about you. The sample report walks through both pages.

Questions about the score