Use this when
content quality, E-E-A-T, content analysis, readability check, thin content, content audit.
Process
Before scoring anything, run Google's own three-question heuristic from its helpful-content guide: Who created it (a visible byline and credentials, non-negotiable for YMYL topics)? How was it created (process disclosure where a reader would reasonably ask, especially for AI-assisted content, and genuine first-hand evidence where claimed)? Why does it exist (to help people, not to attract search clicks — watch for content written to a word-count target or churned purely for a freshness signal)? Weak answers on all three put the page at real risk under Google's core ranking system, not just a cosmetic issue.
Score E-E-A-T across four pillars, weighted the way Google has actually stated it (trust matters most, not an equal split): Trustworthiness 30 (contact info, HTTPS, transparent corrections, date stamps), Expertise 25 (author credentials, technical depth, accurate well-sourced claims), Authoritativeness 25 (external citations, brand mentions, being cited by other experts), Experience 20 (original research, first-hand photos, case studies, proprietary data).
Content metrics to check: word count against topical-coverage floors by page type (these are floors for adequate coverage, not ranking targets — a 500-word page that fully answers the query beats a padded 2,000-word page); readability as a quality indicator only, never an optimization target (Google doesn't use Flesch scores for ranking); natural keyword presence in title/H1/first 100 words without stuffing; clean heading hierarchy; 3-5 relevant internal links per 1,000 words with descriptive anchor text.
On AI-generated content: Google's raters assess low-quality, scaled, or copied patterns, not AI authorship as a standalone flag. Acceptable AI content demonstrates genuine E-E-A-T, has human oversight, and adds original insight; low-quality markers are generic phrasing, no original insight, repetitive structure, and no author attribution — score the actual content quality, never infer or claim AI authorship as a scoring input.
If asked to clean up the user's own draft (not someone else's published content — decline that), two things can help: stripping invisible Unicode characters (zero-width codepoints, directional overrides, hidden tag-character text smuggling) and swapping a short list of conservative AI-typical phrases ("delve into" → "explore") one-for-one, never paraphrasing or adding content. Be honest about scope: statistical watermarking schemes live in word-choice statistics, not codepoints, and nothing reliably detects or removes those — don't claim otherwise.
Report: Content Quality Score with the four-pillar E-E-A-T breakdown and an AI Citation Readiness score, issues found, and specific recommendations.