🎯 The GEO score

The GEO score

The GEO score is a deterministic rubric: our checklist of the on-site signals AI engines use, scored the same way every run. It tells you what to fix first. It is not a prediction, and it is not a measurement of AI behaviour.

What this number is, and what it isn't. The GEO score is our own checklist, run against your pages. Same pages, same checks, same score — no model, no sampling, no variance. That makes it excellent at answering "what should I fix first?" and incapable of answering "will AI cite me?". It is a prioritization tool, not a prediction, and not a measurement of AI behaviour. The mention rate is the measurement. These are different instruments and we won't blur them.

How it's composed

Five categories, weighted. The weights are the single tunable constant in the scorer:

Category	Weight — and what it checks
crawler	30% — can AI retrieval bots actually fetch your pages? If they can't, nothing else matters, which is why it's the heaviest.
extract	25% — is the content in a shape an engine can lift into an answer: answer-first openers, specificity, tables, lists.
schema	20% — structured data: Organization and Article JSON-LD, sameAs, @id.
render	15% — does the content survive rendering, or is it only there after JS a crawler may not run?
entity	10% — entity completeness and disambiguation signals.

Within a category, each check scores pass = 1, warn = 0.5, fail = 0, weighted by severity (critical 3× down to low 0.5×). The subscore is that weighted pass-rate, 0–100. The composite is the weighted blend of the subscores.

What happens when we can't measure something

A category we couldn't evaluate is excluded and the remaining weights are renormalized — it is never scored 100 for analysis we never ran. If we measured nothing at all, the score is 0 rather than a flattering number built from nothing.

Each category also reports its coverage: how many checks we scored out of how many we attempted. This is the same argument as n on the rate side — 100% of two checks is not the same claim as 100% of twenty, and the score alone can't tell you which you're looking at.

What a score movement does and doesn't prove

Your score going up means your site now passes checks it used to fail. That is a real, useful fact: the fix is live, and it's on the page. It is worth exactly that and no more.

It does not mean AI cites you more. Re-running our own checklist after changing the very things that checklist checks is close to arithmetic — of course the number went up; we changed the inputs to our own formula. Anyone selling you that arithmetic as a citation lift is selling you a tautology with a chart on it.

So what would show a citation lift? Only the mention rate, re-measured, with its interval — and even then, a rate that moved while the whole category moved is not proof your change caused it. We show you what we observed and state the confounders rather than dressing a correlation up as a result.

On the weights

Honest caveat: the weights above are our informed judgement about what matters, not a coefficient fitted to your outcomes. The product records score snapshots over time so the weights can eventually be tuned toward what actually correlates with visibility gains. Until that calibration exists, treat the score as a well-argued ordering of work — which is what it's for — and not as a fitted model.