yao-geo-content-refiner

Yao GEO Content Refiner

Use cases

Transform existing articles, product pages, WeChat official account drafts, white paper chapters, or SEO pages into content that AI can more easily understand, extract, cite, and recommend.
Need to improve structuring, credibility, FAQ coverage, atomic fact density, semantic information density, and cross-platform citability.
Output Simplified Chinese deliverables for DeepSeek, Doubao, Qwen, Kimi, Tencent Yuanbao, and the WeChat ecosystem.

Full text of the original article, target brand, target question or keyword, target platform, and target readers.
Brand knowledge base, official website materials, product materials, customer cases, reports, qualifications, and authorized citation sources.
Whether to preserve the original style, whether new FAQs are allowed, whether new tables are allowed, and whether new sources are allowed.
Whether online verification is allowed, whether there is access to internal knowledge bases/APIs/tables/CRM/BI, and which data must not be accessed or written into the report.

Writing a new article from scratch, only doing title optimization, site-wide GEO diagnosis, page technical audits, or backend attribution solutions.
There is no original text and no sources, yet specific data, prices, customer cases, qualifications, or performance promises are required to be added.
Unauthorized access to login backends, paid databases, private APIs, customer privacy data, or internal systems.

Define the transformation boundaries, confirm available sources, and prohibit fabricating facts.
First perform an analysis completeness check: input boundaries, source layering, main questions, follow-up questions, scoring, facts, evidence, structure, semantics, platforms, publishing, and self-Review.
Create a real data acquisition and verification plan: distinguish public web pages, user-provided materials, authorized connectors/APIs, internal systems, and inaccessible items; record permissions, time, freshness, and failure handling.
Score the original text across 8 GEO dimensions: semantic density, structural规范性, citability, authority signals, readability, robustness, novelty, and cross-domain contribution.
Extract atomic fact cards: subject, attribute, value, time, source, applicable boundaries, verification status, and citable sentences.
Add structure: direct answers, summaries, H2/H3, tables, FAQs, ordered steps, and source lists.
Add evidence, evidence strength, semantic entities, platform adaptation matrix, and publishing tracking suggestions; remove filler, unverifiable strong assertions, and overly marketing-oriented sentences.
Output reports in four formats, and according to references/artifact-layout.md and references/quality-gates.md self-check.

Word/PDF must not force wide tables into pages. Tables with more than 4 columns must be converted into vertical fact cards in Word/PDF.
Word tables must use fixed page width and explicit column widths, and must not rely on automatic column widths.
The total width of a Word table must be less than the page's usable width; after generation, tblGrid width must be checked.
PDF must be rendered as PNG to check the right-side margin, and must not be flush with the edge or cropped.
HTML may retain wide tables, but must have horizontal scrolling and long text wrapping.
The HTML must have a top sticky menu bar that stays fixed as the page scrolls down, with menu anchors covering the main report modules.