| name | article-optimize |
| description | Use to make a correct article rank in search and get cited by AI answer engines — on-page SEO + GEO + structured data, all first-party, no external packs. SEO/GEO优化/meta描述/schema结构化数据/AI引用/citation/optimize/精选摘要/featured snippet |
| version | 1.0.0 |
| license | Apache-2.0 |
| homepage | https://github.com/weitzu-com/ai-article-factory |
| when_to_use | Use as the article factory's P7 optimization stage, or standalone when content is already correct (passed QA) and needs to be tuned to (1) rank in Google/Bing for the target query and (2) be quoted by ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. Triggers: SEO优化, GEO优化, 让AI引用, optimize this article, add schema, write meta description, win the featured snippet, AI citation, generative engine optimization. Do NOT use to fix factual errors or thin content — that is the QA gate's job. |
| argument-hint | <article folder | draft.md> [target query] [Article|TechArticle] [--howto] |
| metadata | {"author":"weitzu-com","tags":["seo","geo","structured-data","json-ld","schema","ai-citations","featured-snippet","meta-tags","internal-linking","on-page","optimize","SEO优化","GEO优化","结构化数据","AI引用"]} |
Article Optimize (SEO + GEO + Structured Data)
The factory's P7 stage, made first-party. You take an article that is already correct (P8 QA territory, not yours) and tune it so one piece does two jobs at once: gets indexed and ranked by search engines, and gets quoted by AI answer engines.
First principle
A single well-structured passage that fully answers one question — placed first, sourced, and entity-consistent — is the shared optimum for both a long-tail ranking and an AI citation. Optimize the passage, not the page chrome.
Everything below is pure on-page logic. No crawler, no keyword tool, no external skill pack is required — if those are connected the work gets sharper, but the stage runs standalone on the draft text alone.
Guardrails (read first)
- Right before optimized. If the content is wrong, thin, or unsourced, stop and hand back to QA. Do not paper over weak content with markup.
- Markup mirrors visible content. Never put a claim, rating, FAQ answer, or step in JSON-LD that a reader cannot see on the page. Mismatched schema gets the page demoted, not promoted.
- No invented facts. The "exclusive information" and any statistic must already exist in the draft (sourced via the claim ledger
[C#]). You restructure and surface; you never fabricate.
- One edit, one reason. Each change you make maps to a checklist item below so the human can review it.
Workflow (8 steps)
Run in order. After each step, note what changed and why.
1. Lock the target
Identify the primary query (the one phrase the article must own) and 2–4 secondary/long-tail queries. Identify the canonical entity (brand / product / person / concept) and its one canonical name. If the entity name varies across the draft, normalize to one spelling everywhere — AI disambiguation depends on it.
2. SEO — on-page signals
Produce these artifacts (see the SEO checklist for pass bars):
- Title tag: contains the primary query + an information-gain hook (a number, year, outcome, or angle the SERP doesn't already have). ≤ 60 Latin chars / ≤ 30 CJK chars. Do not duplicate the H1 verbatim if a sharper title wins more clicks.
- H1: one per page, contains the primary query, reads naturally.
- Meta description: ≤ 155 chars (≤ 80 for CJK), front-loads the answer, ends with a reason to click. Never keyword-stuff.
- : short, hyphenated, lowercase, primary keyword, no stop-word filler, no dates that will rot.
