How scoring works

Deduction from perfect

Each category starts at 100 and loses points for failing checks. It is not a sum of earned points, and that choice matters more than it looks.

With a deduction model, adding a new check that your site passes leaves your score unchanged. With an earned-points model, every new check dilutes the ones already there, so a score from March means something different to a score from June. Deduction keeps historical trends comparable as the engine grows.

Severity deductions

Each finding carries a severity. Unless a specific check overrides it, that severity sets the deduction:

Individual checks override this where severity alone understates the consequence. A noindex directive deducts 40 rather than 30, because it removes the page from search results entirely — no other single line of markup does more damage.

Why a warning costs 40% of a failure

A warning means the check found something imperfect but functional. A title of 72 characters gets truncated in search results; it does not stop the page ranking. Charging it the same as a missing title would flatten a distinction that matters when you are deciding what to fix first.

Checks that do not apply

A check marked not applicable is excluded from every count — it neither rewards nor penalises. If it counted as a pass, a page with no forms would outscore a page with a well-built form on the conversion category, which would be rewarding the absence of a feature.

Combining categories

Category scores combine into the overall score using fixed weights that sum to 100. The weights reflect commercial consequence, not an even split:

If a category is missing — because an analyzer failed and was skipped — the remaining weights are renormalised, so a partial audit still produces a score on the same 0–100 scale rather than an artificially low one.

Grades

Estimated impact of a fix

Each recommendation shows an expected score gain. It is calculated, not guessed: we sum the deductions the linked findings caused, clamp that to the category's remaining headroom, and re-weight it into the overall score.

The clamp is what keeps the number honest. Without it, a page failing one check badly would show an impossible “+40 overall” — a category cannot exceed 100, so recovering more points than it lost yields nothing extra.

When an AI model writes the summary, these numbers still come from this calculation rather than from the model. They sit next to commercial claims, so they have to be computed.

Common questions