AI Insight
AI Insights shows you every question readers asked your published documentation, which of them the AI could not answer, and — for each failure — whether you need to write a new topic, fix an existing one, or check that your content is indexed.
Open the report
Select Analytics > AI Insights. The report covers your whole account by default. To narrow it:
- Choose a manual from the document list to see only its questions.
- Choose Last 7 days, Last 30 days, Last 90 days, or Custom to set your own range.
Read the summary
The number that leads the page is the count of questions your documentation could not answer. Select it to jump straight to the list.
Beside it sit three figures for context: how many questions were asked, how many were answered, and an estimate of the support time those answers saved.
Work through the content gaps
The table below the summary is the point of the report. Each row is a question your documentation failed, ranked by how often it failed.
The Why column tells you what kind of fix each row needs:
| Reason | What happened | What to do |
|---|---|---|
| Not covered | The AI found related topics, but none of them answered the question. | Write the missing answer. This is the most common real gap. |
| Nothing found | Search returned no topics at all. | Either nothing covers the subject, or the content exists but is not indexed. Check that the relevant topics are published. |
| No search results | A reader searched and got an empty results page. | Consider coverage, or the words your readers use. They may be searching for a term your documentation never uses. |
| Answer rejected | The AI answered and the reader marked the answer unhelpful. | The topic exists but is wrong, out of date, or unclear. Fix the topic rather than writing a new one. |
The Failures column counts failure signals, not people. One reader asking one question can raise both a search failure and an AI failure, so use the column to rank severity rather than to count readers.
Use the filter on the Why column to work through one kind of fix at a time.
Watch the trend
The chart stacks answered and unanswered questions for each day, so the height of a bar is that day's volume and the red band is the gap.
Reading the bands together matters: a quiet week lowers the red band without your documentation improving at all.
Keep your most-used topics current
Topics the AI answers from lists the topics your answers rely on most, with the number of times each was cited. Select a title to open it in the editor.
These are the topics worth keeping accurate. A stale topic near the top of this list affects every answer built on it.
What the report does not record
- Questions shorter than three words. The viewer treats these as plain searches and does not send them to the AI, so they never appear as AI questions.
- The assistant widget. Questions asked through the embedded widget are not yet recorded.
- Readers. The report counts questions and failures, not people. It holds no reader identities.
Turn a gap into a topic with an AI agent
The same data is available to any agent connected through the Docer MCP server, using the read_content_gaps and read_ai_insights_overview tools. An agent can read the list of failed questions, search your knowledge base for what already exists, draft the missing topic, and leave it as a draft for you to review and publish.