The problem: documentation that nobody trusts
Teams write wiki pages once — then reality diverges. Onboarding docs reference retired services. Runbooks miss the approval step added last quarter. Meeting decisions never publish from notes. Search returns three conflicting pages titled "Auth Overview."
Developers stop searching and ping seniors. AI assistants hallucinate because knowledge is stale or duplicated.
Document quality review is not grammar scoring alone. It is accuracy, findability, linkage to tasks, and freshness — with humans owning what gets published.
Why one-time audits fail
No owner — Pages become orphan after author leaves.
No link to work — Wiki does not reference tasks, releases, or finance decisions that superseded it.
Search without structure — Duplicate titles; no canonical page per topic.
AI rewrite without review — Auto-updated pages silently wrong.
Quality needs continuous review loops tied to project events — releases, incidents, published meetings.
Framework: inventory → gap → draft → approve → publish
1. Inventory — search_docs, wiki index, document list — what exists per topic?
2. Gap analysis — Missing runbooks, stale dates, broken references, decisions in meetings not in wiki.
3. Draft (AI-assisted) — Chatbot or MCP proposes updates from current tasks, get_wiki_page, recent publish_meeting_wiki outputs — citations required.
4. Approve — Subject-matter owner reviews; no silent overwrite of canonical pages.
5. Publish — update_wiki_page or document module; link from affected tasks.
| Quality dimension | Check | WKFGo tool |
|---|---|---|
| Findability | Search returns one canonical page | search_docs, smart_search |
| Freshness | Updated after last release/incident | wiki history, meeting publish |
| Accuracy | Matches board and approvals | list_tasks, get_task cross-check |
| Linkage | Tasks cite wiki; wiki cites decisions | descriptions, publish_meeting_wiki |
Where WKFGo fits: AI assist, human gate
search_docs and smart_search
search_docs and smart_search retrieve wiki and document content for humans and MCP clients. Quality review starts with queries:
- "List wiki pages mentioning deprecated API v1"
- "Find onboarding docs not updated in 180 days"
- "Search for conflicting titles on authentication"
AI summarizes clusters — humans decide merge, archive, or rewrite.
Chatbot wiki vector search
The in-app chatbot uses vector search over wiki — if answers cite wrong pages, that is a quality signal. Track repeated failed Q&A topics for doc backlog.
MCP review workflow
Read-first MCP tools: get_wiki_page, search_docs, get_document, list_documents.
Write with approval: update_wiki_page after owner sign-off — stage 2–3 on adoption ladder.
Pair with publish_meeting_wiki so decisions enter the corpus quality reviewers maintain.
Document module and signatures
Formal specs and signed documents add compliance layer — quality review includes version alignment with wiki summaries for operators.
Rituals that maintain quality
Definition of Done includes doc touch — If behavior changed, one wiki link updated or ticket filed for doc debt.
Release checklist — Search affected runbooks; assign owner before ship.
Quarterly prune — Archive superseded; redirect titles; mark "deprecated" banners.
Doc debt tasks — Track like engineering debt on the board.
Getting started this week
Query search_docs for pages not updated in six months touching production runbooks. Create three doc-debt tasks on the board with owners. For the next release, add wiki search to Definition of Done. Pilot MCP read-only: "Compare onboarding wiki to current task labels on Project Atlas — list gaps with citations." SME approves any update_wiki_page write.
Measure quality by search success — fewer repeated chatbot questions on the same topic means docs improved.
Anti-patterns
- AI auto-publishes wiki without SME review
- Duplicate pages because search is hard
- Beauty rewrite that drops security caveats
- Quality measured by page count not usage
Document quality decays silently. Quarterly, run search_docs for pages not updated in ninety days linked to active releases. AI flags stale specs; humans update_wiki_page after verification—no auto-publish of breaking changes.
Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.
Permission-aware tools mean the same question gets different grounded answers for PM vs contractor—that is a feature, not a bug. Train teams to expect scoped truth.
FAQ
Does WKFGo auto-grade document quality with ML?
No dedicated doc-quality ML score — use search, AI-assisted gap analysis, and human review.
Can AI fix grammar only?
Possible in draft step — still require accuracy review against live project data.
How do bilingual teams maintain quality?
Maintain canonical page per topic per language or clearly marked sections; search both.
Integration with meeting minutes?
Yes — publish_meeting_wiki feeds quality corpus; review merges decision into canonical topic pages.
Better docs through search, review, and links
Use search_docs, chatbot, and MCP to review document quality with AI assist — find gaps, draft updates, approve as a team — so knowledge stays trustworthy.
Try it now
Put these patterns on live project data—not slide decks.