Paste a page URL and get per-signal verdicts on the structure AI engines lift answers from: self-contained chunks, a direct-answer lead, Q&A and step structure, citable tables, heading discipline. Every negative verdict quotes the passage that earned it. There is no 0–100 score, and that is the point.
One fetch of the raw HTML, no JavaScript executed — this is the page a non-rendering AI crawler sees. Six of the seven citability signals are machine-judged; the seventh (the quotable sentence) is human judgment, and the tool says so instead of faking it.
Page read
Your content is not in the raw HTML
The seven signals
How to read the verdicts
✅ The measured structure is present.
⚠️ What the machine can check is fine or borderline; the substance needs your own read (Signal 2 can never do better than this).
❌ A failure shape was found — the evidence is quoted next to it.
N/A Not applicable: the page form does not call for this signal. That is not a failure — inventing the missing form is the fake-fix anti-pattern the playbook warns against.
1 · Self-contained chunk
—
Lifted alone, does a paragraph resolve without its neighbors?
How this is measured
Measured: the share of substantial paragraphs that open on a pronoun or back-reference (This, It, They…), plus explicit broken-chain phrases like "see above" or "the table below". Offenders are quoted. definition · how to fix it
2 · Direct-answer / TL;DR block
—
Is the answer in the first one or two sentences?
How this is measured
Measured: a lead-sized opening paragraph exists before the first H2, and sections do not open with a deeper heading. Whether the lead truly answers is semantic judgment a machine cannot make — this signal is capped at ⚠️ by design. definition · how to fix it
3 · Q&A / FAQ structure
—
Do question-shaped headings have real answers under them?
How this is measured
Measured: question-shaped H2–H4s and whether content follows each one. Whether they match real user queries is not machine-checkable. A page with no question headings gets N/A, not ❌ — do not invent an FAQ. definition · how to fix it
4 · Step / HowTo structure
—
If the page has a procedure, is it a numbered list?
How this is measured
Measured: ordered lists and procedure-shaped headings ("How to…") without one, plus step length. Imperative phrasing is not checked. A page with no procedure gets N/A, not ❌. definition · how to fix it
5 · Citable table / list
—
Does each row read alone, with caption and column labels?
How this is measured
Measured: a caption (or a heading directly above the table) and header cells. Rows that mean nothing when lifted are quoted. A page with no data tables gets N/A. definition · how to fix it
6 · Heading-hierarchy discipline
—
One H1, clean nesting, no skipped levels?
How this is measured
Measured: H1 count, skipped levels (H2 → H4), empty decorative headings — the one fully mechanical signal. Skips are quoted with both headings. definition · how to fix it
7 · Liftable quotable sentence
Human judgment
Is there a crisp standalone claim per section that survives extraction?
Not machine-judged, ever. Deciding whether a sentence is quotable is exactly the semantic judgment a structural check cannot make, so this row stays human. The playbook walks it with worked examples. definition
Why there is no score
Every citability score on the market ships without a published formula. A bare number whose method is opaque is a rumour, not a measurement — two tools compute it differently and neither can be checked. The citability audit playbook documents the pattern, names names, and rejects the composite score on principle.
What ships instead is what is actually reproducible: a per-signal verdict with the evidence quoted — the paragraph that opens on a pronoun, the H2 that jumps to an H4, the table with no header row. You can verify every verdict against your own page in seconds.
How the check works
Fetch the page once, raw HTML only, no JavaScript executed — deliberately: it is what a non-rendering AI crawler ingests. If the HTML carries almost no readable text, that becomes the headline finding and the signals are not judged at all.
Cut the document to its content region (main or article, else body with navigation, header, footer and sidebars stripped) and collect headings, paragraphs, ordered lists and tables in document order.
Judge six signals against the structural checks each card above spells out, in ladder order. Detection of pronouns, question shapes and procedure cues supports English and Chinese pages.
Report four-state verdicts — ✅ / ⚠️ / ❌ / N/A — quoting the exact passage behind every negative one. Conditional signals (3, 4, 5) report N/A when the page form does not call for them.
What this check can and cannot judge
It judges structure, not truth or authority. A page can carry all six structural signals and still not be cited — retrieval, freshness and E-E-A-T are separate gates, and citability is only the grounding-stage lever.
Signal 2 is verified structurally, not semantically: the tool confirms a lead-sized paragraph sits in the answer slot, not that it answers. That is why its best verdict is ⚠️.
Signal 7 (the quotable sentence) is absent from the machine verdicts on purpose. Any tool claiming to score it with a regex is scoring sentence length, not quotability.
The heuristics are calibrated for English and Chinese. Other languages get the structural signals only, and the tool says which signal was skipped rather than guessing.
The fetch runs without JavaScript. If your content only exists after client-side rendering, the tool reports exactly that — which is itself the most important citability finding on such a page (see server-side rendering).
Frequently asked questions
Why does this tool refuse to give a 0–100 citability score?
Because a score needs a published formula to be a measurement, and none of the scored citability tools publish theirs. The citability audit playbook §8 documents this — commercial graders ship composite numbers with undisclosed weights, which makes the numbers unverifiable by construction. Per-signal verdicts with quoted evidence are reproducible; you can check every one against your own page. A composite would add nothing but false authority.
What does N/A mean? Is a missing FAQ hurting my citability?
N/A means the page form does not call for that signal, and that is fine. Signals 3, 4 and 5 are conditional: a definition page needs no step list, a narrative needs no FAQ. The playbook explicitly warns against inventing question headings or numbered steps where neither fits — that is the fake-fix anti-pattern, and it trips AI content filters instead of helping. Fix ❌ verdicts; leave N/A alone.
Why does Signal 2 never show a green check?
Because the machine can only verify that a lead-sized paragraph occupies the answer slot — not that it actually answers the question. Claiming ✅ would overstate what was measured, which is the exact failure this tool exists to avoid. Read Signal 2's ⚠️ as "the structure is in place; now read the lead yourself and ask whether the first sentence answers".
Why fetch without executing JavaScript?
Because most AI crawlers do not execute it either. The check deliberately ingests what GPTBot or ClaudeBot sees: the raw HTML response. If your content only appears after client-side rendering, the tool reports the page as effectively empty — the single most consequential citability finding for such a site, and one a browser-based checker would miss entirely.
All six signals pass — will AI engines cite me now?
Not necessarily. Citability is structural readiness at the grounding stage: it decides whether an engine that has already retrieved your page can lift a passage from it. Being retrieved at all depends on indexing, crawler access and relevance, and being trusted depends on E-E-A-T signals. This check tells you the passage-level structure is not the bottleneck — the citability entry maps where the other gates sit.