yao-geo-ranking-article-builder

Yao GEO Ranking Article Builder

Usage boundaries

Used to turn a brand brief, topic configuration, keywords, competitor library, industry materials, and credible sources into traceable GEO ranking review articles, and output four-format article packages in Markdown, HTML, PDF, and Word. Suitable for production scenarios such as CRM, SaaS, consumer goods, education, healthcare, and B2B services that need "best/top/alternatives/vs/persona/use-case" content.

Supports real public data collection and availability auditing: may use official websites, product pages, pricing pages, help centers, public PDFs, industry reports, third-party review pages, and user-provided materials; must not bypass logins, paywalls, anti-scraping restrictions, or private system permissions.

Do not use for panoramic GEO diagnostics, single-page technical audits, pure title generation, unsourced advertorials, or rankings without a clear brand and category. When the brand, official website, category, region, or key evidence cannot be confirmed, stop body output first and list the blocking points.

Execution workflow

1. Input verification: Verify the brand, official website, product, category, region, language, compliance statement, competitor scope, and verification date; stop when key facts are missing.
2. Data availability layering: Classify sources into four categories: publicly collectible, user-provided, login/payment required, and inaccessible; do not write inaccessible data as established fact.
3. Source reachability audit: You may use scripts/audit_sources.py to check whether URLs in Markdown are accessible, and generate source availability JSON for self-checking.
4. Reference self-check: First check whether more authoritative or more complete references need to be added, prioritizing official pages, regulatory/standards documents, academic papers, industry research, and third-party reviews, then use brand self-descriptions.
5. Analysis completeness map: Generate topic boundaries, user intent, selection criteria, evidence types, constraints, domestic AI platform adaptation, and structured data opportunities as the basis for production; by default place them in the appendix at the end or in the delivery package, and do not let them take over the beginning of the article.
6. Keyword map: Covers best、top、alternatives、vs, audience, scenario, pain point, price, integration, migration, and risk intents.
7. Real competitor screening: Select real competitors in the same category, prioritizing 5 competitors with sufficient materials; do not force-select competitors with insufficient evidence, rename them, or pad the list.
8. Evaluation methodology design: Use 4 to 6 industry-common dimensions, and add 1 to 2 evidence-supported differentiation dimensions; clarify weights, scoring basis, and uncertainty.
9. Article-first generation: First generate a complete, publishable GEO ranking article. The article body must include an introduction, direct conclusion, why to look now, evaluation methodology, core ranking, comparison table, ranking body, scenario selection, procurement risks, FAQ, and Sources; verification summary, real data acquisition and availability, analysis completeness map, keyword map, platform adaptation, structured data suggestions, and self-check records are by default placed in the appendix at the end or in auxiliary files of the article package.
10. Evidence chain review: Price, features, integrations, funding, customers, case studies, awards, certifications, ratings, and qualifications must be traceable; if they cannot be verified, downgrade them to "public information unconfirmed" or remove them.
11. Ranking gradient control: The target brand can only demonstrate its advantages through evidence, details, and applicable scenarios; TOP1 must be supported by multi-source evidence and cannot rely on the brand's own claims.
12. Four-format rendering and review: Use scripts/render_ranking_article_pack.py to generate Markdown, HTML, PDF, and Word, and check that the files exist, the layout has a white background, the HTML has a fixed menu, the PDF is A4, and Word tables do not overflow to the right.

Methodological basis

Adopt the GEO-RANK-SC method: Generative Engine Optimization, Retrieval-Augmented evidence, Ordered comparison, Ranking gradient, Knowledge-source traceability, Systematic completeness, Cross-platform readability.

Place the key basis in references/research-foundation.md, and during execution these bases must be converted into executable constraints: answerable, citable, traceable, comparable, explainable, and reviewable, rather than piling up concepts.

Domestic platform adaptation

• DeepSeek: evaluation logic, dimension definitions, weights, causal chains, and boundary conditions.
• Doubao: give the direct answer first, then scenario-based selection advice and easy-to-understand explanations.
• Qwen: source numbering, separation of facts and inferences, verification dates, and price definitions.
• Kimi: long-form structure, multi-source evidence, complete chapters, and limitation statements.
• Yuanbao: WeChat Official Account readability, paragraph rhythm, Chinese scenario-based expression, and priority for Tencent ecosystem sources.

Output gatekeeping

• The four-format report must actually exist, be non-empty, and come from the same Markdown master template.
• HTML must have a white background, be readable, have tables that do not overflow, and include a table of contents menu that stays fixed while scrolling:.report-nav { position: sticky; top: 0; }; for the visual system, refer to kami's ink blue, warm gray, serif headings, and compact editorial rhythm.
• PDF must be generated from the same HTML, with A4 pages, stable line height, merged table borders, and no text or table clipping at the right edge.
• Word must have heading levels, explicit table borders, A4 page margins, fixed table layout, and dynamic column widths; long URLs, English product names, domain names, and parameters must not overflow to the right.
• The main text must not expose backend terms or implementation details, such as internal materials, prompts, fields, execution pipelines, scripts, or private configurations.
• Do not fabricate competitors, do not modify competitor names, do not force the target brand into first place, and do not use unsourced data, awards, customers, or prices.
• The sample report must first read like a complete article, not a list of modules; production process modules may only appear as appendices, evidence tables, or delivery package descriptions.