Integration
How to connect support insights help to a chat widget
How to connect support insights help to a chat widget — answered from your own docs. How Chatref – AI-Powered Help Desk Software teams use Chatref (website widg
Support insights analyze what your users ask through Chatref's widget, surface knowledge gaps and trending topics, and deliver digest emails with actionable recommendations. When you update your knowledge base based on those insights, the widget's answers improve for every future visitor - closing a tight feedback loop between what users need and what the agent delivers.
What connects to what
The Chatref insights engine sits between your widget and your knowledge base, turning raw conversation data into operational direction. Every question a visitor asks through the widget gets analyzed, tagged by topic, and fed into a synthesis layer that identifies patterns, gaps, and emerging issues across all your chats.
Three pieces work together:
- The widget captures every question visitors ask on your site, along with the agent's response and whether the visitor accepted it or asked for a human.
- The insights engine processes that stream - grouping similar questions, flagging topics where the agent struggled, and detecting spikes in specific issue types.
- Your knowledge base receives the output: you learn which docs to update, which topics need new content, and what your users actually care about rather than what you assumed they would ask.
The loop is not passive reporting. The digest emails you receive name specific topics - for example, "3 users stuck on API keys this week" or "Import questions up 40% since the new release" - and point you directly at the content gaps causing the friction. When you fill those gaps, the widget answers those questions correctly next time without a human stepping in.
This is distinct from a search analytics dashboard. The insights engine does not just count query volume - it evaluates whether the agent resolved the question successfully, tags conversations by outcome, and prioritizes the gaps that are actually causing repeated handoffs or dead ends.
How to set it up
The connection between insights and the widget is automatic once both are active, but the quality of the insights depends on how you prepare the knowledge base upfront.
Start with your content. Upload your help docs, setup guides, FAQs, and any pages from your site that cover common customer questions. Chatref learns from these sources and uses them as the only reference material when answering through the widget. The broader and more accurate your starting knowledge base, the more meaningful the gap analysis becomes - the insights engine can only flag what is missing relative to what is already there.
Add the widget snippet to your site. One line of code placed in your site's <head> or footer makes the widget available on every page. Once live, every visitor question flows into the conversation inbox and the insights pipeline simultaneously. No additional wiring or configuration is required to connect widget conversations to insights.
Let the widget run for a cycle. The insights engine needs a representative sample of conversations to surface patterns worth acting on. For a site with steady traffic, a few days to a week is usually enough to see which topics dominate. For lower-volume sites, give it two weeks before drawing strong conclusions.
Review the digest emails when they arrive. Each digest surfaces the top topics users asked about, highlights conversations where the agent could not resolve the issue, and flags questions that triggered a human handoff. Use these signals to decide what to update in your knowledge base next.
Update your knowledge base based on what you learn. If the digest shows users repeatedly asking about a feature your docs do not cover, add that content. If the agent answered correctly but users kept rephrasing the same question, the existing doc might need clearer language or a different structure. After updating, Chatref re-ingests the content automatically, and the widget begins serving the improved answers within minutes.
Chatref - AI-Powered Help Desk Software handles the technical plumbing for you. You do not configure integrations, set up webhooks, or write automation rules to connect insights to the widget. The data flow is built into the platform.
What users see
The end-user experience does not change when insights are active - there is no badge, no "powered by insights" label, no visual shift. The widget behaves the same way it always does: a visitor types a question, and the agent answers from your knowledge base.
What changes over time is the quality and precision of those answers. After you act on the insights and fill knowledge gaps, visitors who ask previously uncovered questions get a direct, sourced answer instead of a fallback or a prompt to contact support. A user asking about a feature you added to the knowledge base last week gets the right answer on the first attempt.
Visitors who asked questions the agent previously handled poorly will not see any improvement unless you update the underlying content. The insights engine tells you what to fix - it does not retrain or reprogram the widget automatically. The operator is the link between what the data shows and what the widget delivers.
For your team, the shift shows up in the shared inbox. As the knowledge base improves, the volume of conversations that require a human takeover drops. The conversations that do land in the inbox tend to be more complex - the genuinely unusual cases rather than the repeat questions the agent now handles correctly. This makes the inbox more manageable and the work more interesting for support staff.
Troubleshooting
Insights emails are not arriving. Confirm the widget is live and receiving visitor questions. The insights engine needs conversation data to generate digest emails. If the widget is newly installed and has not yet seen enough traffic, wait for chat volume to build. Also check your email spam folder and ensure the account email address is correct in your Chatref settings.
The digest flags the wrong topics. Insights are only as accurate as the conversation volume. In the first week or two with low traffic, topic clustering may be imprecise. Give the engine more data. If miscategorization persists, review whether your knowledge base content clearly distinguishes between related topics - fuzzy boundaries in your docs can cause the agent to conflate similar questions, which skews the insights tagging.
I updated my knowledge base but the widget still gives old answers. Chatref re-ingests content after updates, but allow a few minutes for the changes to propagate. If the issue persists longer, verify the new content was saved correctly and that the document you updated is actually included in the agent's source set. A document that was accidentally excluded from training will not affect widget answers even after edits.
Users report the widget can not answer a question that my docs cover. This may be a phrasing mismatch rather than a content gap. The user might be asking with terminology your docs do not use. Review the exact question text in the conversation inbox, then check whether your knowledge base addresses the topic in language your users actually use. If not, add a section that bridges the vocabulary gap - the insights digest often flags these phrasing mismatches as "unresolved" even when the topic technically exists in the docs.
Handoff rates are not dropping after updates. Not all questions should be deflected. Some issues genuinely need a human - account-specific billing questions, edge-case bugs, or sensitive topics like data deletion requests. If handoff rates remain high, filter the digest to see which topics are driving the handoffs. If they are genuinely human-required topics, the rate is appropriate. If they are questions the agent should handle, the knowledge base may need more precise, step-by-step guidance for those specific scenarios.
FAQ
What causes support insights problems for Chatref - AI-Powered Help Desk Software?
Most support insights issues trace back to insufficient conversation volume or a knowledge base that does not accurately reflect what users ask. If the widget is new or your site has low traffic, the insights engine has too little data to surface reliable patterns. If your knowledge base covers only high-level concepts but misses the specific error messages, workflows, and edge cases users encounter, the insight quality will reflect that gap. A third common cause is content drift - your product changes but the knowledge base stays the same, so the agent and the insights both operate on outdated information.
How do I improve support insights for Chatref - AI-Powered Help Desk Software?
Treat your knowledge base as a living asset that evolves with every digest email. When a digest flags a trending topic, add or update the relevant content within the same week while the pattern is still fresh. Prioritize the topics that consistently trigger human handoffs over topics that the agent already handles well. Keep your knowledge base structure flat and topic-focused - long, catch-all documents make it harder for the agent to locate the right answer and for the insights engine to isolate specific gaps. Finally, review the actual conversation transcripts in the inbox periodically. The digest summarizes patterns, but reading real user language reveals phrasing mismatches and nuance that the summary can miss.
Related guides
Put this into practice
Chatref answers your customers from your own content, day and night. Add it to your site and go live in minutes – free to start.