What makes an AI knowledge base work for customer support?

An AI knowledge base is the set of documents an AI support agent may answer from, and it works when each source is focused, current and checked by someone who knows the answer. Your help centre is a good place to start.

In Kawl Now, each agent has its own collection. Support can use FAQs, notes and selected GitHub files, and sales uses notes you write. A workspace holds up to 100 sources, each with a title of up to 140 characters and up to 50,000 characters of text.

Start with the questions customers already ask.

Choose a bounded area such as first-time setup or a particular feature. Gather the current answer for each question. Use your own support records where you have permission, and remove personal or account-specific information before turning them into reusable guidance.

Write down the questions the agent should leave to your team, such as billing disputes or account changes, and the contact route it should suggest. Keeping those boundaries explicit makes it easier to judge test conversations.

Write each answer so the agent can find it.

During a call, the agent searches approved sources by keyword and reads up to four matching excerpts, each about 4,500 characters long. Short, focused sources written in your customers’ words are easier to match than long pages that cover many topics.

  • Cover one task or question in each source, and put the answer first.
  • Use the words customers say, such as “invite a teammate” as well as “add a user”.
  • Keep the conditions with the answer: the plan, version or setting it depends on.
  • Describe screens and buttons in words. The agent reads text, not images.

Choose a source for each answer.

Add an FAQ manually or import the GitHub file that holds the current explanation. Check instructions, links and version-specific details before approving it. Public repositories need no token. Ask us before relying on a private one, because access is set up on our side for each workspace.

Prepare sales knowledge as its own collection.

Use approved sales notes for features, pricing and buying questions. The sales agent never sees support sources or GitHub imports, so check that it has the public information it needs.

In the current pilot, website URL import is switched off, so add website information as a sales note and review it before approval.

Test the answers and keep approvals current.

Ask the same question in different ways. Include ambiguous questions and questions outside the sources. When the sources don’t answer, the agent is told to say it doesn’t know, so a “don’t know” in a transcript usually points to a gap you can fill.

Every GitHub re-sync, even of unchanged text, needs approval again before the agent uses it. Repeat the relevant tests when a feature or plan changes. Approval is a publishing control, and answer quality still needs review.

Common questions

Does approving a source guarantee an accurate answer?

No. Approval controls which material the Kawl Now agent can use, but generated answers still need checking. Test questions inside and outside your sources, including ones your team should handle, and read the transcripts before you switch on customer calls.

How do I keep the knowledge base in step with our docs?

Re-sync each GitHub file after it changes, from the dashboard or with Kawl Now’s sync_github tool. Every sync needs approval again before the agent uses it, and nothing syncs automatically on push, so add the re-sync to your docs release routine.

How long should each knowledge base article be?

Keep each source to one topic. Kawl Now accepts up to 50,000 characters per source, but the agent reads up to four excerpts of about 4,500 characters for each search, so a short, focused source is easier to match to a question.