Evaluate page content for usefulness, E-E-A-T, readability, thinness, and AI citation readiness, plus last-mile draft cleanup (AI-typical phrasing and invisible
复制下面这句话，粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具，它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-seo-content-404c3719bcc6 ，按照其中的说明把「seo-content」安装到你（当前 AI 工具）中。执行前先告诉我将运行的命令和写入的位置，等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件，由原作者维护。在 GitHub 查看
Before scoring E-E-A-T sub-factors, every page audit should pass Google's own three-question heuristic from the helpful-content guide:
| Question | What to look for |
|---|---|
| Who created it? | Visible byline, author bio page, professional credentials. Required where readers expect it; non-negotiable for YMYL. |
| How was it created? | Process disclosure where readers would reasonably ask, especially for AI-assisted content. Original research / first-hand evidence / lived experience. |
| Why does it exist? | "To help people" rather than "to attract search clicks." Watch for niche entry without expertise, content churn for freshness signals, content written to a word-count target. |
Primary source: https://developers.google.com/search/docs/fundamentals/creating-helpful-content
When all three answers are weak, the page is at risk under the core ranking system's helpfulness signals (formerly the standalone Helpful Content System, merged into core during the March 2024 update).
Read ${CLAUDE_PLUGIN_ROOT}/skills/seo/references/eeat-framework.md for full criteria.
Compare against page type minimums:
| Page Type | Minimum |
|---|---|
| Homepage | 500 |
| Service page | 800 |
| Blog post | 1,500 |
| Product page | 300+ (400+ for complex products) |
| Location page | 500-600 |
Important: These are topical coverage floors, not targets. Google has confirmed word count is NOT a direct ranking factor. The goal is comprehensive topical coverage; a 500-word page that thoroughly answers the query will outrank a 2,000-word page that doesn't. Use these as guidelines for adequate coverage depth, not rigid requirements.
Note: Flesch Reading Ease is a useful proxy for content accessibility but is NOT a direct Google ranking factor. John Mueller has confirmed Google does not use basic readability scores for ranking. Yoast deprioritized Flesch scores in v19.3. Use readability analysis as a content quality indicator, not as an SEO metric to optimize directly.
Google's raters assess low-quality, scaled, copied, or AI-generated main content patterns rather than AI authorship as a standalone issue.
Helpful Content System (March 2024): The Helpful Content System was merged into Google's core ranking algorithm during the March 2024 core update. It no longer operates as a standalone classifier. Helpfulness signals are now weighted within every core update. The same principles apply (people-first content, demonstrating E-E-A-T, satisfying user intent), but enforcement is continuous rather than through separate HCU updates. Google now also documents continuous, smaller unannounced core updates between major ones (changelog 2025-12-09).
Gen-AI optimization is SEO (Google docs, 2026-06-29): the official "optimizing for generative AI features" guide states you do not need new AI files, markup, Markdown, content chunking, or AI-specific rewrites; chasing inauthentic "mentions" is unhelpful. AEO/GEO is rebranded SEO rooted in core ranking/quality.
Honest scoping (Google docs, 2026-06-05): per "Using third-party SEO tools, services, and advice," no tool guarantees rankings and third-party tools have no access to Google's internal ranking data. claude-seo's scores are heuristics, not Google-internal signals, so say so in reports, and validate GEO/AEO findings against Google's official guidance (Search Console is the first-party source).
For "humanize this", "remove watermarks", or "clean up this draft", run the bundled cleanup script on the user's own content:
"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run content_humanize.py draft.md -o cleaned.md
cat draft.md | "${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run content_humanize.py --json
Two deterministic passes, both logged in the JSON output:
invisible_removed): strips zero-width
codepoints, directional marks/overrides, Unicode tag characters (hidden
text smuggling), and normalizes exotic spaces. Emoji sequences (ZWJ,
variation selectors next to emoji) are preserved.changes): conservative 1:1 swaps from the
replacement table ("delve into" → "explore", etc.). Nothing is
paraphrased or added.Scope honesty: statistical watermarks (SynthID-style token-probability schemes) live in word choice, not codepoints. No tool reliably detects or removes them; do not claim otherwise in reports. This cleanup is for editing the user's own drafts, not for laundering third-party content — decline requests to strip provenance from content the user doesn't own.
Optimize for AI search engines (ChatGPT, Perplexity, Google AI Overviews):
Google AI Mode is Google's conversational AI search surface. Google upgrades the AI Mode model often (Gemini 3.5 Flash became the default on 2026-05-19, and newer Flash models have shipped since, per blog.google); never tie advice to a model version. Treat third-party AI Mode usage, citation, and link-share figures as methodology-dependent unless primary-sourced, and optimize for both AI Mode and AI Overviews (see the seo-geo skill).
Key optimization strategies for AI citation:
Generative Engine Optimization (GEO):
Per Google's AI optimization guide, "optimizing for generative AI search is optimizing for the search experience, and thus still SEO": AI Overviews and AI Mode are grounded in the same ranking and quality systems as classic Search. The optimization signals that matter (quotability, attribution, heading hierarchy, freshness) are SEO fundamentals applied to AI-search surfaces, not a separate discipline. Cross-reference the seo-geo skill for detailed workflows; both surfaces share the primary-source synthesis in ${CLAUDE_PLUGIN_ROOT}/skills/seo-geo/references/google-ai-optimization-guide.md.
Before attributing a traffic or ranking change to anything, list the confirmed Google updates in that window from the primary-source ledger:
"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run seo_updates.py --since <yyyy-mm> --json
Every entry cites a Google-owned URL. If freshness.stale is true, say the
ledger may miss recent updates and check status.search.google.com before
drawing conclusions. A date overlap is a hypothesis, never proof of cause.
| Factor | Score | Key Signals |
|---|---|---|
| Experience | XX/20 | ... |
| Expertise | XX/25 | ... |
| Authoritativeness | XX/25 | ... |
| Trustworthiness | XX/30 | ... |
Weights are this skill's own scoring model, ordered to reflect Google's stated hierarchy: Trust is most important (30), then Expertise/ Authoritativeness (25 each), then Experience (20); maxima sum to 100. Google publishes no numeric E-E-A-T weights (only that trust is most important), so treat the split as our internal model. Do not use an equal 25/25/25/25 split (it contradicts Google's "trust is most important").
If DataForSEO MCP tools are available, use kw_data_google_ads_search_volume for real keyword volume data, dataforseo_labs_bulk_keyword_difficulty for difficulty scores, dataforseo_labs_search_intent for intent classification, and content_analysis_summary for content quality analysis.
| Scenario | Action |
|---|---|
| URL unreachable (DNS failure, connection refused) | Report the error clearly. Do not guess page content. Suggest the user verify the URL and try again. |
| Content behind paywall (402/403, login wall) | Report that the content is not publicly accessible. Analyze only the visible portion (meta tags, headers) and note the limitation. |
| Thin content (fewer than 100 words retrievable) | Report the findings as-is rather than guessing. Flag the page as potentially JavaScript-rendered or gated, and suggest the user provide the full text directly. |
For prompt-guided content optimization, use /seo flow optimize <url> and /seo flow win <url>: FLOW's optimize and win prompts provide structured E-E-A-T improvement and BOFU conversion workflows.