How to Build a Knowledge Base (Help Centre)

How to build a knowledge base: choosing articles, structure, writing rules, search, update cadence and reducing ticket volume.

If a support team answers the same question five times a day, an article is missing. A knowledge base is not marketing content but the support team’s memory.

Which articles to write

Do not start with a blank page. The source material is already in your hands:

  • Support tags: the 15 most frequent topics of the last three months are your first 15 articles.
  • Canned replies: the text your team copies and pastes is already half an article.
  • Onboarding questions: everything a new customer asks in the first week.
  • “How do I” requests: tickets that are usage questions rather than faults.

One simple rule runs the whole thing: a question asked three times becomes an article.

Article structure

  1. Write the title in the user’s language. Not “SMTP configuration” but “Email sending is not working”. People search for their problem, not for your feature names.
  2. Answer in the first paragraph. Long introductions do not get read; put the outcome first.
  3. Number the steps. Each step should contain a single action.
  4. Add screenshots. Remember they need updating when the interface changes.
  5. End with an “if this did not work” section. Never leave the user in a dead end; route them to support.

Structure

Keep the number of categories low. Users will not scan more than about seven. A typical structure: Getting Started, FAQ, Step-by-Step Guides, Troubleshooting, Integrations, Release Notes.

Search matters more than the menu. Most users never click a category — they search. Make the search box prominent and log the searches that return nothing; that list is the most honest source of articles still to write.

Writing rules

  • Use short sentences. A support article is not literature.
  • Avoid jargon or define it the first time it appears.
  • One topic, one article. A long article covering three topics gets found for none of them.
  • Show the date and version. Users need to see whether the article is current.

Keeping it current

An out-of-date knowledge base is worse than none, because it costs trust. A sustainable rhythm:

  • Every product change produces a list of affected articles.
  • The 20 most-read articles are reviewed quarterly.
  • Every article has an owner.
  • “Was this article helpful?” feedback is collected, and the ones voted down are fixed first.

Its relationship with an AI assistant

A knowledge base is also the fuel for an AI assistant. The assistant cannot produce knowledge that does not exist; built on an empty base it starts inventing. So the order is clear: knowledge base first, assistant second.

Measurement

  • Article views: the most-read articles point at areas of the product that need improving.
  • Failed search rate: a direct list of articles to write.
  • Ticket reduction: did tickets on that topic fall after the article was published?
  • Article-to-ticket rate: how many people read the article and still opened a ticket? A high number means the article is incomplete.

Frequently asked questions

Should the knowledge base be public?

General usage articles should be — search traffic reduces support load and brings new customers. Account-specific information belongs behind a login.

How many articles should I start with?

Fifteen good articles beat a hundred mediocre ones. Start with the most frequent questions and grow.

Can videos replace articles?

They complement, not replace. Video is good for showing an interface but it is not searchable; the text must exist.

Next step: to see how the ticket desk and knowledge base work together, request a demo.