You publish a document, share the link, and questions start arriving. Individually they are chat logs. In aggregate they are the most honest description you will get of what your audience wants and where it gets stuck. Document Intelligence turns that pile into statistics, keyword counts, clustered topics, AI-written reports, and an analyst you can question in plain language.
Intelligence is for owner admins; readers never see it. For the raw exchanges underneath the numbers, see conversations and moderation.
Getting in, and what unlocks it
There are four entry points. Open a document for editing and choose the Intelligence tab in the Learning group, or use the Intelligence action (lightbulb icon) in its row menu. A collection has its own tab, and the dashboard sidebar has one in the System group covering every document in your owner group.
Two conditions gate it. Plan: Intelligence is part of the analytics features included on Pro, Business, and Unlimited, and on Free the tab shows an upgrade prompt. It must also be enabled on the platform; if not, the tab says Intelligence is not enabled on this server (plans and limits). Data: until a document has received at least 5 questions its tab shows a locked screen reading Intelligence unlocks at 5 questions, with a progress bar.
Before reading any number, notice the Exclude owner & admin traffic switch at the top of every Intelligence view. It is off by default, so figures include your own team's testing — which, before a document is widely shared, is most of the data. Switch it on and questions from your owner admins and Docutrain staff drop out, the panel is tinted, and a Learner traffic only badge appears, so a screenshot always says what it omits. It applies to everything on the tab, reports included.
The panels
Overview is pure measurement, computed from the conversations with no AI interpretation. Its cards cover Questions, Conversations, Q / conversation, Askers (distinct people estimated by network address), Satisfaction (the share of rated answers with a thumbs-up), and Last 7 days with its week-on-week change. Below sit Questions over time, bucketed by day, week, or month; Geography, switchable By country, By region, and By city; and a response-time line.
Keyword density counts the exact words and phrases people type: counts, not interpretation. Side-by-side Phrases and Words charts list top terms with how many questions contain them and what share that is. A toggle switches between bar view and treemap, and Show all expands to 50 terms. Filler words are removed and variants grouped (Stemmed · stopwords removed), and under 20 questions the footer warns of Thin data. It works from the very first question, unlike the rest of the tab.
Docutrain Assistant introduces Doc, an analyst you question in plain language. Doc is not the reader-facing assistant; its subject is the questions your audience asked — recurring themes, points of confusion, gaps between what people want and what the document covers. It grounds answers in the question list itself, quotes examples, and separates what the data shows from what it is inferring. Open chat starts a session, with three suggested prompts: What are people most confused about?, Where are the content gaps?, What changed recently? Answers stream live and New chat clears the thread.
Reports
Generate new report produces a durable AI-written snapshot of everything asked so far. Only one generates at a time. The new row appears at the top of Reports history as Generating, analysis usually takes one to two minutes, and you can close the editor — it continues in the background, with an animated Intelligence indicator on the document's library row. Cancel stops it. Past 25,000 questions a report analyzes the most recent 25,000 and says so; a failed run shows Failed with the error inside.
Reports are written about your learners, not as a critique of the document. Each has a narrative in five sections: what learners are trying to learn, where they are getting stuck, apparent prerequisite knowledge gaps, how they are using the material, and an overall audience summary. Three charts follow: Themes by sentiment, a treemap sized by question volume and colored by dominant feeling (Curious, Confused, Skeptical, Enthusiastic, Neutral); Learner intent, a donut over Learn, Verify, Apply, Other; and Sentiment mix. A themes table lists each theme with its Frequency, Sentiment, Learner gap, and verbatim Example questions.
Every report sits in Reports history with its date, questions analyzed, and status chip. View / Hide expands it inline, Copy puts the formatted text on your clipboard, PDF downloads it with charts intact. Deleting is restricted to platform administrators — owner admins do not see the trash icon — and a generating report must be cancelled first.
Sharing Intelligence outside your organization
Report share links gives stakeholders outside your organization a private read-only window without a Docutrain account. Enter an email, select Send invite, and the recipient gets a link unique to them. Links work without signing in, expire after 30 days, and can be revoked. Each invite shows the email, a status dot (Active, Expired, Revoked), the expiry, and how often it has been opened. Resend emails a fresh link and invalidates the previous one, which is also how you renew an expired one; Revoke and Reactivate switch a link off and on, and the trash icon removes the invite entirely.
The recipient gets a standalone page carrying your organization's logo and the document's title, a Read-only badge, and a sidebar for Overview, Keyword density, Reports, and Assistant — Doc included. Export PDF downloads the most recent report and Share copies the page link; they cannot generate, delete, or change anything. A link opened after it expired or was revoked shows a plain explanation such as This link has expired rather than any data.
Organization-wide, Topics, and collections
The dashboard's Intelligence tab aggregates Overview, Keyword density, and Doc across your whole library, plus a breakdown panel whose Documents and Collections tabs list Questions, Satisfaction, and Last activity per row; selecting a row opens that scope's own tab. A footnote explains that conversations spanning several documents count toward each one, so that column can exceed the overall question count — the Collections tab ends with an italic Multi-document (no collection) row, and those totals do reconcile. Reports are generated per document or per collection, never at organization level.
Topics is where clustering happens. Where the keyword panels count literal words, Topics groups the questions into named themes and ranks them. It appears at all three scopes with one difference of framing: topics are always clustered across the owner's whole question corpus, and at document or collection scope the volumes and gaps shown are that scope's share. Until your organization has roughly 200 questions the panel says so instead.
Once active there are two searchable lists. Content gaps ranks by gap score, where higher means answers more often went wrong — no source found, re-asked, or thumbed down — weighted by how often people asked; a gap of 0 means no problems were detected. All topics lists everything, with a Volume / Conversations toggle setting the ranking. Each row carries a label, a metric line such as "14 questions · 9 conversations", a trend against the previous 30 days (New, Rising, Steady, Falling, Stopped, Dormant), an answer-quality signal (Answered well, Mixed, Poorly answered, No signal), and a volume chart.
A collection gets the same toolset once saved, covering every conversation held inside it, including chats spanning several of its documents.
Intelligence gets sharper as questions accumulate. Generating a report at milestones — before and after a content revision, or at the end of a training cohort — gives you comparable snapshots, and the raw conversation logs show the exchanges behind whatever it flags.