Find gaps in your knowledge base
Scan your docs on a schedule, flag what's stale or contradictory, and route fixes to owners.
A knowledge base audit usually happens once, heroically, and never again. This workflow puts it on a schedule: the agent scans your docs regularly, flags what's outdated, contradictory or missing, and routes each finding to the person who can fix it, one small question at a time.
What does this workflow do?
On each run, the agent reads your knowledge base in Notion and Google Docs and looks for four kinds of rot. Outdated pages, where the doc says something newer sources contradict. Contradictions, where two pages disagree and a reader can't know which to trust. Gaps, where questions people actually ask in Slack have no documented answer. And orphans, pages nobody has touched or viewed in months that may deserve archiving.
The gap detection is the part a manual audit can't do: because Agentwork sees the questions your team asks day to day, it knows which answers people needed and didn't find. The report ranks findings by how often the missing or wrong information actually comes up.
How does it work?
- The agent scans on schedule. Your Notion workspace and Google Docs, respecting permissions throughout.
- It cross-checks for contradictions. Pages that disagree with each other, or with what's actually being said and decided in Slack, get flagged with both sources.
- It mines real questions for gaps. Questions asked in Slack that no doc answers become documentation candidates, ranked by frequency.
- It routes fixes to people. Each finding goes to the likely owner as a small, answerable request: "This pricing page says 14 days trial, Stripe says 30. Which is right?" One answer, and the agent can update the page with their approval.
- It reports the trend. A health summary per run: what got fixed, what's newly stale, and whether the knowledge base is getting better or worse.
Why route findings to individuals?
A list of 60 stale pages posted to a channel fixes nothing; everyone assumes someone else owns it. One specific question to one specific person, answerable in a sentence, gets answered. The agent does the tedious part (finding the rot, drafting the correction) and the human does the two-second part: confirming what's true.
Answers compound. Every settled contradiction and filled gap makes the knowledge base more trustworthy, which is what makes every other workflow built on it (support triage, RFP drafting, onboarding) more accurate. This template is maintenance for all the others.
Works with
Notion, Google Docs, Slack, Gmail. Runs weekly or monthly.
Frequently asked questions
How does it decide a page is outdated?
Signals include contradiction by newer documents or decisions, references to things that no longer exist (old pricing, renamed features, departed teammates as owners), and staleness relative to how often the topic changes elsewhere.
Who gets the fix requests?
The likely owner: the page's author or editor, or the person your team's activity suggests owns the topic now. If ownership is unclear, the agent asks the team lead to assign rather than guessing.
Does it edit documents by itself?
It drafts the correction and applies it after the owner approves. Teams that want tighter control keep all edits manual and use the workflow purely for detection and routing.
Can it tell us what's missing as well as what's wrong?
Yes. Because it sees the questions your team asks in Slack, it reports the asked-but-undocumented topics ranked by frequency. That list is usually the highest-value writing your team can do.
How is this different from Notion's own analytics?
Page analytics show views and edits. This workflow reads the content: whether pages are true, consistent with each other, and cover what your team actually needs to know.