The AI Citation Scorecard

This is the exact 100-point rubric we grade every page against before it goes live. Steal it. Use it on your own site.

Everyone in AI search has a scoreboard nobody can read, because none of them published the rules of the game. Here are ours. Three criteria, 100 points, specific penalties. If your page scores under 70, an AI engine is unlikely to cite it — and below is exactly why, and exactly what to fix.

Criterion 1 — Answer Fragment Density (35 points)

Can an AI lift a clean, standalone answer straight off your page? This measures whether your content contains extractable, self-contained answers — direct-answer openers, question-format H3 headings, and responses kept under roughly 80 words so they can be quoted whole.

You earn points for: leading each section with the answer, not the wind-up; framing headings as the actual questions people ask; keeping key answers short enough to extract intact.

You lose points for: burying the answer three paragraphs deep; atmosphere-only sections that describe a mood but state no fact; forcing the AI to synthesize an answer you never actually wrote down.

Criterion 2 — Claim Proof Density (35 points)

Is every claim backed by verifiable proof within two sentences of making it? This is where most pages collapse. AI engines discount unverifiable superlatives and reward claims paired with named entities, measurements, linked sources, and quantified outcomes.

You earn points for: naming names; attaching a number to every claim; linking to a source; quantifying the outcome (“grew from 8.5% to 19.3%” beats “grew significantly”).

You lose points for: unverifiable superlatives (“the best,” “industry-leading”); standalone assertions with no proof attached; citing a generic third-party study in place of your own first-party evidence.

Criterion 3 — Semantic Structure (30 points)

Can a machine parse the shape of your page without guessing? This measures a clean H2/H3 hierarchy, question-format headings, aria-labels, linked entities, and a single consistent voice throughout.

You earn points for: a logical heading tree; headings phrased as questions; entities linked to authoritative references; one voice from top to bottom.

You lose points for: emoji in headings; nested FAQ lists; raw shortcodes bleeding onto the page; spacer characters used for layout; voice shifting mid-page as if stitched from three writers.

How to Read Your Score

Score	What It Means
90–100	Citation-ready. Clean answers, proof on every claim, machine-parseable.
70–89	Competitive, with clear gaps. Usually a proof-density problem.
50–69	Invisible to AI. Reads fine to humans, gives an engine nothing to lift.
Under 50	Atmosphere with no substance. All feel, no fact, no proof.

Why We Give This Away

Because publishing the standard is how you become the standard. Anyone can read this rubric. Almost no one can apply it consistently across a whole site — that’s the work, and that’s what we do. The scorecard is free. Making your pages score 90+ is the job.

Frequently Asked Questions

How do you measure whether a page will get cited by AI?

We score it against a 100-point rubric across three criteria: Answer Fragment Density (35 points, extractable standalone answers), Claim Proof Density (35 points, verifiable proof within two sentences of each claim), and Semantic Structure (30 points, machine-parseable heading hierarchy and linked entities). Pages scoring 90 or above are citation-ready.

What is the single biggest reason a page fails to get cited?

Claim Proof Density. Most pages state claims without attaching verifiable proof — named entities, measurements, or linked sources — within two sentences. AI engines discount unverifiable superlatives and reward quantified, sourced claims.

What score does a page need to be cited by ChatGPT or Perplexity?

In our experience, pages scoring 90 or above on this rubric are consistently citation-ready. Pages between 70 and 89 are competitive but have clear gaps, usually in proof density. Below 70, an AI engine has little clean material to extract.

Can I use this scorecard on my own website?

Yes. It is published to be used. Score your own pages against the three criteria, find your lowest-scoring one, and fix its proof density first — that is almost always the highest-value single change.