A useful knowledge base helps a defined audience complete a task or make a decision with current, trustworthy instructions. Publishing more articles is not the goal. Coverage, findability, comprehension, successful action and maintenance are.
Begin with repeated questions and failed searches. Give every article an owner, evidence source, audience and review trigger. Connect self-service to a clear route for human help.
Choose the audience and task
Separate customer, employee, partner and agent knowledge where access, language or context differs. For each priority topic, state what the reader is trying to accomplish and the condition that sends them to support.
A broad article called “Using the system” is difficult to find and maintain. Prefer bounded tasks with clear outcomes.
Use demand evidence
Analyse support requests, search queries, zero-result searches, repeated contacts, onboarding questions and product changes. Interview users in their own words.
Prioritise by frequency, consequence and suitability for self-service. A rare safety-critical question may outrank a frequent low-impact one.
Create a content brief
| Field | Purpose |
|---|---|
| User question | matches real language and intent |
| Audience/prerequisite | prevents unsafe or irrelevant use |
| Outcome | defines successful completion |
| Evidence source | supports accuracy and change review |
| Escalation | routes exceptions to help |
| Owner/review trigger | keeps the article current |
Write for action
Use a direct title, short answer, prerequisites, ordered steps, expected result and exception guidance. Put warnings before the risky action. Use screenshots only where visual identification reduces uncertainty.
Describe interface labels accurately but do not let the article become unusable after a minor visual change. Explain the decision and result around the click.
Design findability
Use user language in title, summary, headings and synonyms. Link from the product or workflow moment where the question occurs. Curate related articles by next task, not keyword similarity alone.
Test with misspellings, old terminology and problem descriptions. Search that returns many plausible results may still fail the user.
Connect knowledge to service
Agents should find, use and improve the same controlled knowledge. Capture which article was sent and whether the request reopened. Provide a way to contact support with the attempted article and context.
Do not force self-service for complex, sensitive or urgent cases. A knowledge base should shorten the path to resolution, not hide support.
Collect useful feedback
A helpful/not-helpful control needs a reason or linked behavior. Pair feedback with search term, article version, task outcome and subsequent contact where privacy permits.
Review comments quickly but validate before changing instructions. One frustrated response may reveal a real exception; aggregate ratings alone rarely explain it.
Operate ownership and review
Assign content owner, subject expert and publishing authority. Set time-based review only where necessary; prefer triggers such as policy, product, price, role or process change.
Show last reviewed date internally, retain versions and expire content that has lost an owner. Redirect or replace obsolete public paths where readers may still arrive.
Test articles with users
Give a draft to someone in the target audience and observe the task. Note terms they do not recognise, decisions they cannot make and steps where the expected result differs. Test an exception and the help route.
Do not coach during the first attempt; assistance can hide a content gap. Update the article and repeat with a new participant where the task is material.
Control screenshots and media
Use images when they identify an interface state, compare options or clarify a concept. Crop sensitive data, add descriptive alt text and retain source and update ownership. A decorative screenshot creates maintenance without helping the task.
When interface details change frequently, use annotated concepts or short text around the stable decision. Review media with the same trigger as the instructions.
Use assisted search responsibly
Autocomplete, suggested answers and generative retrieval can shorten discovery, but they need a controlled source set, citations or traceability, access enforcement and fallback. Test unsupported questions, outdated content and restricted material.
Measure whether the answer completes the task and whether users can reach the underlying article. A fluent summary must not conceal uncertainty or invent a procedure.
Support language and regional variation
Identify content that differs by language, jurisdiction, product edition or customer entitlement. Keep a shared authority where possible and make variation explicit. Translation needs subject review, not only linguistic conversion.
Track source and translated versions so a policy change cannot leave one audience with obsolete instructions.
Measure successful use
Track successful search, task completion, repeated contact, escalation quality, stale content, review timeliness and agent reuse. Treat “deflection” carefully: no ticket may mean success or abandonment.
Sample user journeys and support outcomes. A small set of clear, maintained articles can outperform a large ungoverned library.
Run the maintenance loop
Each week review failed searches and high-impact feedback; each month review recurring support demand and content gaps; after every material change review affected articles before release.
Retire duplicates and consolidate authority. The knowledge base becomes trustworthy when users learn that one findable answer reflects the current service and tells them what to do when it does not fit.