Skip to main content
ClientX monitors every conversation and flags questions where the AI couldn’t find a good answer in your knowledge base. These appear as gaps — a continuously updated list of questions that reveal exactly where your knowledge base needs work. Instead of guessing what to add, you get a clear signal from real visitor behavior.

​
How gap detection works

When a visitor asks something your knowledge base doesn’t cover well, the AI agent logs the question and marks it as a gap. ClientX records:
  • The question the visitor asked
  • The page they were on when they asked it
  • What the bot said in response (so you can see if it deflected or guessed)
  • An AI-suggested answer draft, generated from your existing knowledge base content
Gaps are surfaced in the Gaps tab within 24 hours of the conversation. They reflect the last 30 days of traffic, so the list stays focused on current visitor needs rather than historical outliers.

​
Reviewing gaps

1

Open the Gaps tab

Go to Knowledge Base > AI Gaps in the left sidebar. You’ll see a count badge showing how many unanswered questions are waiting for review.
2

Review each flagged question

Each gap card shows the visitor’s question, the date it occurred, and a link to the page where it was asked. Click What the bot said to expand the agent’s original response.
3

Draft an answer with AI

Click Draft with AI to generate a suggested answer from your knowledge base. ClientX uses semantic search to find the most relevant existing content, then drafts an answer the agent can use. Review the draft — edit it directly in the text area before saving.
4

Save the answer as a Q&A card

Click Save as Q&A to add the approved answer to your knowledge base immediately. The gap is removed from the list and the agent can now answer this question accurately in future conversations.
5

Dismiss irrelevant gaps

Click the ✕ button to dismiss gaps that aren’t relevant to your product or that you don’t want to address. Dismissed gaps are removed from the list permanently.
AI-drafted answers are marked “ClientX drafted · review before saving.” Always read the draft before saving — the AI uses your existing knowledge base as context, but you should verify accuracy, especially for pricing, compliance, or product-specific claims.

​
AI-suggested content

The Suggestions tab surfaces content proposals the AI generated proactively — not from a single unanswered question, but from patterns across many conversations. These are topics visitors ask about repeatedly that your knowledge base doesn’t cover yet. Suggestions are submitted by team members testing the chat widget via the shared preview link. Each suggestion includes:
  • The question or topic that was flagged
  • A suggested answer (if one was provided by the tester)
  • The date and submitter who flagged it
To act on a suggestion, review the suggested answer, edit it if needed, and click Add to KB. The content is added as a Q&A card immediately. Click Dismiss to remove suggestions that aren’t relevant.

​
ICP Profile

The Target ICP section (found under Targeting in the sidebar) helps the AI understand who your best customers are. Filling this out improves two things: lead scoring (the agent correctly identifies high-fit visitors) and conversation quality (the agent prioritizes the right pain points and use cases). Your ICP profile includes:
Industries and verticals, company size, funding stage, revenue range, and geography. These fields tell the agent what a “good fit” company looks like so it can recognize them in conversation.
The job titles and roles that typically purchase your product — VP of Sales, Head of Marketing, RevOps Manager, Founder. The agent uses this to detect buyer intent signals in conversation.
The problems your ICP experiences (one per line) and the signals that indicate they’re ready to buy. Buying signals help the agent prioritize and escalate high-intent visitors in your inbox.
Who is NOT a good fit — B2C businesses, companies under a certain size, industries you don’t serve. The agent uses this to avoid spending time on visitors who won’t convert.
Typical deal size and how long your sales cycle runs. Used for lead scoring and to set realistic expectations in conversations.

​
Generating your ICP with AI

If you haven’t set an ICP yet, ClientX offers a short wizard. Answer five questions about your product, your buyers, company size, problems you solve, and industries you serve — then click Generate ICP. ClientX creates a full structured profile you can review and edit before saving.
Review your knowledge gaps weekly. A well-maintained knowledge base — one where gaps are addressed quickly and the ICP is kept current — dramatically improves AI response quality and lead conversion rates. Most teams see measurable improvement after their first gap-review session.

​
Coverage score

Your overall knowledge base health is reflected in the coverage score shown in the sidebar (0–100). Regularly filling gaps, approving suggestions, and keeping your ICP up to date moves this score upward. Aim for 70+ for consistent, confident agent responses.

Add sources

Add URLs, files, Q&A pairs, and competitor cards to expand your knowledge base.

Knowledge base overview

Learn how ClientX retrieves content and what the coverage score means.