## TL;DR
Write articles in the fix-first format (symptom, 5-step fix, escalation line), make search work (titles match ticket language), assign every article an owner and a review date, and put the KB inside the agent's ticket workspace. A KB agents must go find will not be used; it must be where they already work.

## The error
```text
(Knowledge management; no error.)
```

## Steps
1. Define the article template: symptom/title in user language, TL;DR fix, numbered steps with expected outputs, escalation criteria. Expected: template published. Every article follows it; consistency is usability.
2. Seed with the top 20 ticket drivers (pull from your ticket data). Expected: seeded. Coverage of the common stuff matters more than total article count.
3. Fix search: titles must contain the words agents type (error text, product names), and the KB must be searchable from the ticket workspace. Expected: searchable. A KB with bad search is a write-only archive.
4. Assign owners and review dates (6 months default). Expected: assigned. Stale articles are worse than no articles; they teach wrong fixes.
5. Add feedback on every article (helpful/not helpful) and review the metrics monthly. Expected: loop running. Retire or fix low-rated articles.

## When to use
- New KB launch
- Fixing an unused KB

## When not to use
- End-user self-help (different audience, different writing)
- One-off documentation

## Compatibility
- ServiceNow Knowledge, Confluence, or any KB with search and ownership

## Variants
### KCS (Knowledge-Centered Service)
The formal methodology: create as you solve, review as a team. Worth adopting at scale.
### AI search over the KB
Good search multiplies KB value; fix the articles first, then the search.

## Why it happens
Agents under time pressure use what is fast and trusted. The KB earns that by matching their language, living in their workflow, and being reliably current.

## Edge cases
- Reward article creation; the people who know the fixes are the busiest.
- Archive aggressively; a KB of 500 stale articles loses to 100 current ones.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_oofSi8puy8WDOaW-rFPNPg