Document insights
Document insights surface the articles your AI cites most and the ones it ignores, so you can find knowledge gaps and prune dead weight.
Document insights show how your knowledge base is actually being used. Two views matter: the documents the AI reaches for constantly, and the ones it never touches. Together they tell you which articles to keep, which to update, and which to retire. Go to Insights → Documents to open the page.
The two views
Switch between Most referenced and Least useful with the toggle at the top of the page.
Most referenced
This view ranks documents by how often the AI cited them in a customer answer over the selected date range. Each row shows the document name, its parent project, the citation count, and the average customer rating for the conversations that used it.
These are your most valuable articles. Keep them current.
Least useful
This view shows documents the AI rarely cites, or documents that show up in low-rated conversations. Each row shows the document, the last time it was cited, and the rating signal.
A document can be "least useful" for a few reasons. It might be obsolete, it might be poorly written so retrieval can't match it to a question, or it might be assigned to the wrong widget.
A document with zero citations isn't always dead weight — it might cover a question customers haven't asked yet. Cross-reference against the topic cloud on Conversation insights before you delete anything.
Filters
Both views respect the same filters in the top bar.
Filter by widget
Narrow the view to a single widget to see what its assigned documents are doing. Useful when you want to audit the marketing widget without the support widget's documents polluting the view.
Pick a date range
The date picker in the top right controls the time window for citation counts. A 7-day window is good for spotting recent shifts; a 90-day window is better for retiring documents you're truly sure about.
Finding knowledge gaps
The fastest way to grow your knowledge base is to follow the signal here.
Open Least useful
Sort by lowest citation count first.
Look for low-rated chats
A document that shows up only in low-rated conversations is one customers tried to use and didn't get a clean answer from. That's a rewrite candidate.
Cross-reference with feedback
Open Insights → Feedback for the same date range. Verbatim comments often name the topic the AI missed.
Update or write the document
Make the edit in Documents. New chats on the topic pick it up immediately.
Pruning dead weight
A messy knowledge base hurts retrieval. The bigger the index, the more chances the AI has to grab the wrong chunk.
Sort Least useful by oldest last-cited date
Documents not cited in months are the strongest deletion candidates.
Open each one and read it
Make sure it isn't an evergreen article that just hasn't come up yet.
Archive or delete
Use the archive action in Documents if you want to keep the record but stop the AI from retrieving it.