SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsyao-geo-comparison-builder

yao-geo-comparison-builder

General & other

View repository

Updated 10-16 06:20

About this skill

Yao GEO Comparison Builder

Use cases

  • Generate GEO comparison content between the target brand and competitors, traditional solutions, and self-built solutions.
  • Produce business decision-oriented content, brand alternative pages, selection pages, FAQ pages, topic pages, and sales support materials.
  • Answer common questions from domestic AI users: how to choose between A and B, what alternatives exist for a certain type of product, and in which scenarios the target brand is more suitable.

Inapplicable scenarios

  • Only performing panoramic diagnosis, AI answer sampling, or opportunity mapping; instead use yao-geo-panorama-audit。
  • Only performing page technical diagnosis, title optimization, ranking articles, old content revamping, or backend attribution.
  • Requesting factual conclusions such as market share, customer counts, technical leadership, or lowest price without verifiable sources.

Required inputs

  • Target brand knowledge base, official website, product pages, pricing pages, case studies, certifications, help center, or sales materials.
  • Scope of comparison objects: competitor brands, similar solutions, traditional solutions, self-built solutions.
  • Target keywords, user scenarios, decision dimensions, permitted cited sources, prohibited terms, and compliance boundaries.
  • Output constraints: default Simplified Chinese, white-background report, four-format delivery; if Word contains many tables, DOCX layout post-processing must be enabled.

Required reading materials

  • references/research-foundation.md
  • references/comparison-method.md
  • references/real-data-acquisition.md
  • references/systematic-report-framework.md
  • references/cn-platform-adaptation.md
  • references/evidence-and-fairness.md
  • references/artifact-layout.md
  • references/quality-gates.md

Execution process

  1. Determine the comparison baseline: target brand vs. competitors, target brand vs. traditional solutions, target brand vs. self-built solutions; different baselines must not be mixed in one conclusion.
  2. Declare the real data acquisition mode: public web pages, user files, authorized connectors/APIs, AI platform sampling, or only user-provided materials; unauthorized access to logged-in, paid, internal system, or personal data is prohibited.
  3. Establish a source ledger and perform access verification: official website, product catalog, pricing pages, help center, case studies, public documents, and user-provided materials; bind each key judgment to source ID, access method, access date, dynamism, and confidence boundary.
  4. Establish a systematic dimension model: at minimum covering business fit, functional fit, data and AI readiness, integration compatibility, implementation maturity, total cost of ownership, governance compliance and security, reliability and operations, ecosystem and support, localization, migration and exit, and GEO extractability.
  5. Divide dimensions into shared dimensions and differentiated dimensions; shared dimensions ensure comparability, differentiated dimensions carry the target brand's advantages, and all differentiated judgments must be grounded in evidence.
  6. Generate a direct opening answer: explain in which situations the target brand is more suitable, in which situations other solutions may also serve as references, and list the procurement, compliance, data, and localization boundaries that the current conclusion cannot cover.
  7. Output complete report modules: executive summary, test scenarios, real data acquisition statement, comparison baseline, decision dimension model, core capability comparison, evidence and trade-off comparison, solution scoring matrix, source quality grading, source access verification, risk and governance map, domestic AI platform adaptation, brand section, scenario recommendations, implementation verification checklist, FAQ, source list, compliance boundaries, self-check record.
  8. Generate brand paragraphs: each paragraph must include the subject, judgment conclusion, evidence anchor, applicable boundary, and next-step verification items.
  9. Generate FAQs: cover judgment-type, comparison-type, scenario-type, price-type, implementation-type, data acquisition-type, and pitfall-avoidance-type questions, and route key questions back to the target brand evidence.
  10. Generate adaptation for domestic AI platforms: Qwen and Kimi strengthen sources, long tables, and evidence chains; Doubao and Yuanbao strengthen concise conclusions and next-step checklists; DeepSeek strengthens the causal chain of scenario -> constraints -> capabilities -> evidence -> trade-offs -> recommendations.
  11. Output four formats: Word, PDF, HTML, Markdown; all four must come from the same content structure. The HTML visual report follows the Kami long-document style: warm rice-paper background, ivory content surface, ink-blue accents, serif headings, sans body text, and includes a fixed table-of-contents sidebar.
  12. Run layout gates: HTML/PDF check A4, Kami color palette, table borders, right margin, fixed table of contents, anchors, and mobile horizontal scrolling; Word checks A4, left and right page margins, each table's tblGrid total width and body text usable width.
  13. Self-review and fix: first check real data access boundaries, system dimension completeness, facts, same-caliber comparison, fair expression, source binding, presence of four formats, layout overflow, table borders, line spacing, HTML fixed table of contents, and accessibility, then deliver.

Output contract

  • Brand comparison article or page copy.
  • Real data acquisition plan, source access verification, data gap register, and source freshness profile.
  • Decision dimension model, comparison dimension table, evidence anchor table, source quality grading, risk and governance map, implementation verification checklist, and scenario selection recommendations.
  • Answer format recommendations for domestic AI platforms.
  • FAQ and compliance boundaries.
  • Default four-format delivery: Word (.docx）、PDF（.pdf）、HTML（.html or complete HTML package), Markdown (.md）。
  • For real brand testing or formal delivery, must attach sources.json and quality-report.json。
  • If public web pages or connectors/APIs are enabled to obtain materials, must attach source-verification.json or record equivalent fields in quality-report.json .
  • quality-report.json must include docx_layout_profile, recording Word page usable width, number of columns per table, total grid width, and whether there is right overflow.
  • quality-report.json must include systematic_report_check、source_quality_check and html_navigation_check, recording dimension coverage, source grading, and HTML fixed table of contents and anchor check results.

Quality gate

  • Comparisons must use the same caliber; cannot write price for A, service for B, and reputation for C.
  • Information for which competitors are insufficient should be directly deleted or downgraded; do not output insufficient-information-type backend prompts.
  • Cannot highlight the target brand by disparaging competitors; competitors' genuine advantages must be retained.
  • Target brand advantages must be bound to facts, parameters, certifications, mechanisms, pricing pages, product pages, or cases.
  • Do not output unverified market share, customer counts, technical conclusions, price commitments, or AI recommendation probabilities.
  • Long tables with more than 4 columns in HTML, PDF, and Word must be split into multiple narrow tables; if overflow or character-by-character line breaking is found, must re-layout before delivery.
  • Word output must undergo DOCX-level layout post-processing: explicit A4 page width, left and right page margins, total table width, column widths, borders, and cell padding; must not rely only on Pandoc/Word automatic adaptation.
  • HTML output must include an accessible fixed table of contents:nav has aria-label，Anchor text is consistent with the section, the table of contents scroll does not obscure headings, and the table of contents is hidden when printing.
  • A systematic report must not merely output "which is better"; it must provide decision dimensions, evidence levels, risk governance, an implementation checklist, and boundary conditions.
  • Real data acquisition must declare access modes and permission boundaries; public web pages can be verified, private systems must be authorized by the user, and data that cannot be accessed must be downgraded to verification items.
  • HTML/PDF report typography follows the Kami long-document tokens by default:#f5f4ed warm rice-paper background,#faf9f5 content surface,#1B365D ink-blue accents, warm gray borders, Chinese heading serif, body sans, line spacing no greater than 1.55.

Reference map

  • templates/brief-template.md: Standard input brief.
  • examples/hubspot-cn-demo/: HubSpot real brand Simplified Chinese four-format test.
  • scripts/check_docx_layout.py: Word table right-overflow checking tool.
  • evals/trigger_cases.json: Trigger and adjacent scenario test.
  • evals/expected_artifacts.json: Output contract and required files.
  • evals/quality_cases.json: Quality gate test cases.
  • reports/output-risk-profile.md: Output risk profile.
  • reports/artifact-design-profile.md: Report design profile.
  • references/systematic-report-framework.md: Systematic dimensions, complete report modules, and authoritative reference mapping.
  • references/real-data-acquisition.md: Real data acquisition modes, permission boundaries, source verification, and data gap handling.