Improve knowledge from recurring gaps

What you’ll accomplish

Turn repeated conversation evidence into safer knowledge—not automatic product facts.

Finish with

A reviewed source change you have retested

1. Review the actual conversationA detected gap is a review signal, not a verdict.

   Filter for a recurring topic, then read the original question and final state. A labeled gap can coexist with a resolved conversation.

2. Find the responsible content problemDistinguish missing facts from stale, unclear, or hard-to-retrieve facts.

   Check whether the answer is already present, then identify the smallest source improvement. Prioritize repeated gaps with material visitor impact.

3. Get the answer approvedUse an owner-confirmed source, never visitor wording as product truth.

   Draft an audience-appropriate answer. Have the policy owner verify it before publishing.

4. Re-index and prove the improvementWait for Ready, then repeat the original question and close variations.

   Update the authoritative website, file, or text source. Track future source usage and gap evidence, but do not treat a low gap rate as proof of complete coverage.

Interpret evidence carefully

Knowledge-gap analytics depend on completed conversation analysis. Pending or failed analyses are not equivalent to a negative result. Gap rate is the share of analyzed conversations with a detected gap, not a direct measure of all unanswered questions. A low rate does not prove complete knowledge coverage, and a single gap does not prove a widespread content problem.

Prioritize patterns with repeated, material user impact. Use a direct source change for factual corrections, a clearer structured FAQ for recurring ambiguity, and an explicit handoff path for questions the agent should not answer.

Last updated on