Skip to main content

AI Authoring Model

Summary​

AI may accelerate inventory, comparison, drafting, linking, and impact analysis, but all content remains evidence-bound, visibility-aware, human-owned, and draft-only until approved.

Audience​

  • Documentation authors, AI-tooling maintainers, reviewers, approvers, engineering, architecture, and security teams

Reference Content​

Mandatory rules​

  1. Never invent. Do not create features, APIs, workflows, roles, states, architecture, ownership, or business rules without verified evidence.
  2. Always cite source. Connect meaningful technical claims to repository-relative source, approved specifications, or authoritative documentation.
  3. Prefer executable backend evidence. Use domain, application, persistence, contract, route, and configuration structure before UI labels or prose inventories.
  4. Compare inventory. Search existing documentation and source inventories before creating, moving, or replacing pages.
  5. Preserve metadata. Retain stable IDs, slugs, ownership, visibility, review, source, version, and relationship fields.
  6. Maintain links. Update landing pages, Related Articles, See Also, sidebars, redirects/slugs, and graph edges.
  7. Update the graph. Propose evidence-backed dependencies and remove stale relationships when behavior changes.
  8. Generate drafts only. AI output uses documentation_status: draft and must not represent itself as approved or published.
  9. Never auto-publish. Human review and authorized publication are mandatory.
  10. Protect information. Never expose secrets, credentials, personal/customer data, restricted source, vulnerabilities, or unsafe internal procedures.

Authoring workflow​

AI must label uncertainty with approved maturity language and ask for confirmation when evidence cannot resolve a material decision. It must not use absence of evidence as proof of absence unless the scoped inventory supports that conclusion.

See Also​

Keywords​

  • AI-assisted documentation
  • Evidence-first authoring
  • Human approval

Revision Information​

  • Status: Draft
  • Last reviewed: 2026-07-15
  • Review cycle: Quarterly