The step
Every article is graded against four groups of checks, weighted by how much each actually moves citation:
Meta — 15% Title length, description length, canonical URL, Open Graph tags. The cheapest points on the page and the most commonly left on the table.
Structure — 30% A single H1. H2 and H3 depth. Lists. Alt text coverage. Content depth against real thresholds rather than a vague "long enough".
Schema — 20% Article or BlogPosting. FAQPage. Organization with sameAs, graded on how many. BreadcrumbList. Author or Person markup, which is the E-E-A-T signal most sites skip entirely.
Content — 35%, and this is where pages fail Read by Claude, not pattern-matched: is there a clear one-sentence definition of the topic? Cited statistics? Direct-answer formatting? Expert quotes? Is the piece semantically dense, or padded?
Four sub-scores, blended into one out of 100. Articles only — not collection pages, not product pages. Different job, different rules.
The pattern we see most
Well-written articles that make no extractable claim. Fluent, on-brand, correctly targeted, and offering an engine nothing it can quote with confidence.
Content carries the heaviest weight for exactly that reason. A page can pass every technical check and still be unquotable.
The fix is usually small. A specific number where there was a generalisation. A named source. A definition in the first paragraph instead of the fourth. Rarely a rewrite.
Where we are
This is a step we run on client content today and a checker we keep building out as a proprietary internal tool. We are an AI-native agency; the instruments are ours and they keep improving.
Powers
The Signal Engine — and it is the quality gate on everything we publish.
Your free AI Perception Report shows where your best pages are losing citations.
Book the call →