The fifteen checks, explained in one place

The audit is fifteen items in three groups of five. The groups exist because the failures happen in order: a crawler that cannot reach the page never reaches the reading checks, and a page that cannot be read cannot be quoted. Fixing group C while group A fails is fixing the wrong thing.

Every item is machine-verifiable, and every verdict keeps the evidence it was decided from — the response we received, the text we parsed, the timestamp. Where something cannot be measured, the report says so instead of guessing.

Group A — Accessibility & Reachability

Can a machine get to the content at all?

• A1 · crawler access — the AI user agents we test are not blocked by robots.txt, a firewall or a bot rule.
• A2 · content in raw HTML — the text exists in the response without running JavaScript.
• A3 · sitemap exists and is submitted — every listed URL is live and indexable.
• A4 · no dead-link pollution on core pages — links from the pages that matter resolve.
• A5 · stability and HTTPS — response time, certificate validity, and the site answering consistently rather than once.

Group B — Semantic Comprehension

Can a machine tell what the page is?

• B1 · unique, complete title and description — each page states its own subject, and the two agree with the visible heading.
• B2 · structured data (JSON-LD) is present and valid — the markup parses, uses the right type, and is connected to the site entity.
• B3 · heading tree — one <h1>, a hierarchy that does not skip levels, and headings that describe sections rather than decorate them.
• B4 · language and locale signals — lang, and for multilingual sites, an hreflang cluster that is reciprocal and names a default.
• B5 · canonical and duplication — each subject has one canonical URL, and near-duplicates are linked rather than left to compete.

Group C — Extractability & Trust

Can a machine quote it?

• C1 · title and description — hmm, read that as C1: answer-first structure: the conclusion appears before the reasoning.
• C2 · cross-page consistency — the same fact does not differ between pages.
• C3 · extractable structure — lists, tables and short paragraphs that survive being lifted out of context.
• C4 · dated and attributable — the reader can see when the page was last updated and who wrote it.
• C5 · quotable conclusion — at least one sentence states the answer so plainly that quoting it is safe.

How the score works

Each item contributes one point; a report states the total and the evidence per item. The number matters less than the list underneath it: fifteen verdicts with their evidence can be worked through, and a single opaque score cannot. The full definition — including what counts as evidence for a pass — is in the standard, and the methodology page explains how a measurement is taken.