Assess observable content characteristics for AI-mediated search — with evidence, never a probability.
15 active of 25 published rules · each active rule is versioned and backed by evidence. Planned rules are on the roadmap.
The main content is short (under 300 words), which limits its usefulness as source material.
What it checks
Counts main-content words; flags pages under 300 words.
Confidence
Medium — derived from evidence quality, not asserted.
A long page with no H2–H6 headings is hard to scan and to quote.
What it checks
On pages over 800 words, checks for at least one H2–H6 heading.
Confidence
High — derived from evidence quality, not asserted.
Heading levels jump (e.g. an H2 straight to an H4), breaking the semantic document outline used to extract structure.
What it checks
Reads headings in document order and flags any jump of more than one level.
Confidence
High — derived from evidence quality, not asserted.
The title promises enumerable content (e.g. “Top 10…”, “types of…”) but the page has no list or table markup to carry it.
What it checks
Detects list-style titles, then checks for any <ul>/<ol>/<table>.
Confidence
Medium — derived from evidence quality, not asserted.
The average sentence runs long (over 30 words), which lowers readability and the quality of extracted answers.
What it checks
Computes mean words-per-sentence across the main content.
Confidence
High — derived from evidence quality, not asserted.
The page has two or more question-style headings but never marks the content up as FAQPage or HowTo schema.
What it checks
Counts interrogative headings and checks the JSON-LD for FAQPage/HowTo.
Confidence
High — derived from evidence quality, not asserted.
A long page (over 1,200 words) that makes numeric claims cites zero external sources.
What it checks
On 1,200+-word pages with a numeric claim, checks for at least one outbound external link.
Confidence
High — derived from evidence quality, not asserted.
The page exposes no publication or modification date, so AI systems cannot judge its freshness. Same underlying signal as EEAT-030, evaluated here for AI-search freshness.
What it checks
On article-type pages, checks for any publication or modification date.
Confidence
High — derived from evidence quality, not asserted.
The title promises a comparison (contains “vs”), but the body has no table to lay the options side by side.
What it checks
Detects “vs / versus” titles, then checks for a <table>.
Confidence
Medium — derived from evidence quality, not asserted.
Long-form content (over 1,500 words) offers no summary or TL;DR in its first 100 words for skimmers and AI extractors.
What it checks
On 1,500+-word pages, scans the first 100 words for a summary/TL;DR cue.
Confidence
High — derived from evidence quality, not asserted.
A long page (over 1,000 words) has zero internal links pointing to related content on the same site.
What it checks
On 1,000+-word pages, counts internal links to same-site pages.
Confidence
High — derived from evidence quality, not asserted.
A how-to style title exists, but the steps are not presented as an ordered (numbered) list.
What it checks
Detects “how to” titles, then checks for an <ol> element.
Confidence
Medium — derived from evidence quality, not asserted.
The intro matches known filler phrasing (e.g. “In today’s world…”) that adds no informational value. Informational-only flag.
What it checks
Matches the first 100 words against a filler-phrase pattern list.
Confidence
Medium — derived from evidence quality, not asserted.
The reading-ease score is below 60 (fairly difficult), which is too complex for general informational content.
What it checks
Computes the Flesch reading-ease score from sentence and syllable counts.
Confidence
Medium — derived from evidence quality, not asserted.
The most-repeated content word exceeds 3% of total words — a keyword-stuffing signal that reads as low-quality to AI systems.
What it checks
Computes the frequency of the single most-repeated non-stopword against total word count.
Confidence
Medium — derived from evidence quality, not asserted.
The title is a question, but nothing in the first 300 words reads as a direct answer.
What it checks
Needs answer-pattern detection over the first 300 words (not just prose presence). Deferred to avoid false positives.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
The title is a plain noun phrase (e.g. “Content Marketing”) but no defining sentence appears early.
What it checks
Needs reliable definitional-sentence detection tied to the title term. Deferred to avoid false positives.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
The paragraph right after a main subheading does not directly answer what the heading implies.
What it checks
Needs heading→paragraph pairing plus answer-pattern detection on that paragraph.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
An identified direct-answer paragraph is too long to extract cleanly as a snippet.
What it checks
Depends on first identifying a “direct-answer” paragraph (see AISR-016).
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
The direct-answer paragraph contains no number, date or named statistic.
What it checks
Depends on first identifying a “direct-answer” paragraph (see AISR-016).
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
A heading phrased as a question is not answered within the first 50 words beneath it.
What it checks
Needs heading→following-text pairing plus answer-pattern detection.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
Too many acronyms are used without ever being spelled out.
What it checks
Needs acronym extraction paired with definition/expansion detection to avoid false positives.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
Term/description pairs are not marked up with semantic <dl> HTML.
What it checks
Needs to recognise conceptual term/description pairs in prose — not reliably detectable yet.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
Data that should be a table is rendered as plain text or <div>s.
What it checks
Needs to recognise implicit tabular structure in non-table markup.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.
An internal link’s anchor text does not match the destination page’s main heading.
What it checks
Needs fetching every internal destination and comparing anchor text to its heading — deferred for cost.
Status
Published in the catalogue; not yet evaluated by the scanner, so it never produces a finding.