Publish a document, share the link, and questions start arriving. One at a time they are chat logs. In aggregate they are the most honest account you will ever get of what your audience wanted and where your material let them down.

Nobody fills in your feedback form. Everybody asks the question they actually have.

Document Intelligence is the tab that reads that pile: measurements, clustered topics, AI-written reports, and an analyst you can interrogate in plain language. It belongs to owner admins, readers never see it, and it comes with the analytics on Pro and above.

The switch to find before you read a single number

At the top of every Intelligence view sits Exclude owner & admin traffic, and it is off by default.

Which means that on a new document, most of what you are looking at is your own team testing it. Your questions, your colleagues' questions, the six variations you tried while checking the tone. Read those numbers as audience behavior and you will draw exactly the wrong conclusion.

Switch it on and questions from your admins drop out, the panel takes on a tint, and a Learner traffic only badge appears. That badge is the thoughtful part: a screenshot pasted into a deck always says what it left out, so nobody downstream has to ask whether the number was filtered.

The tab unlocks at five questions, with a progress bar until then. Keyword counting works from the first one.

Counting, then interpreting

The panels are deliberately layered, and the order matters.

Overview is pure measurement with no AI anywhere near it: questions, conversations, questions per conversation, distinct askers, satisfaction from thumbs ratings, last seven days with the week-on-week change. Then questions over time, geography by country, region or city, and response times.

Keyword density counts the literal words and phrases people typed. Not themes, not interpretation. Just what share of questions contained each term, with filler words stripped and variants grouped, in a bar chart or a treemap. Under twenty questions it labels itself Thin data rather than letting you over-read a sample of nine.

Topics is where interpretation begins, clustering questions into named themes once your organization has roughly two hundred of them, and saying so plainly until then.

Two lists come out of it. All topics ranks everything by volume or conversations. Content gaps is the one to read first: it ranks by a gap score built from answers that went wrong, meaning no source was found, the question got re-asked, or the answer was thumbed down, weighted by how many people asked. A gap of zero means nothing went wrong there.

That is a genuinely uncomfortable list, and it is the most valuable thing on the page. It is your document's failures, sorted by how many people hit them.

Each row carries a trend against the previous 30 days (New, Rising, Steady, Falling, Stopped, Dormant) and an answer-quality signal (Answered well, Mixed, Poorly answered, No signal), which together tell you whether a gap is getting worse or whether last month's fix worked.

Doc, the analyst

Docutrain Assistant introduces Doc, who is not the assistant your readers talk to. Doc's subject is not your document. Doc's subject is the questions your audience asked about it.

You question it in plain language, and it offers suggestion chips to start from, the first three of which are usually the right ones: What are people most confused about?, Where are the content gaps?, What changed recently?

It grounds its answers in the question list itself, quotes real examples, and separates what the data shows from what it is inferring. That last distinction is rarer than it should be in analytics tooling, and it is what makes the difference between an interesting paragraph and one you would repeat in a meeting.

Reports you can hand to someone

Generate new report produces a durable written snapshot of everything asked so far. It takes a minute or two, runs in the background so you can close the editor, and shows an indicator on the document's library row while it works.

The framing is deliberate and worth noticing: reports are written about your learners, not as a critique of your document. Five sections cover what people are trying to learn, where they get stuck, what prerequisite knowledge they seem to be missing, how they are using the material, and who they appear to be.

That is a more useful lens than a scorecard. "Readers arrive assuming they already know X" is something you can act on. "Document scores 6/10" is not.

Three charts follow. Themes by sentiment is a treemap sized by volume and colored by dominant feeling, across Curious, Confused, Skeptical, Enthusiastic and Neutral. Learner intent splits questions into Learn, Verify, Apply and Other, which tells you whether people are studying your material or checking it. Then a sentiment mix, and a themes table with frequency, sentiment, the learner gap, and verbatim example questions.

Every report is kept in a history with its date and question count, expandable inline, copyable, and downloadable as a PDF with the charts intact. Only platform administrators can delete one, so a report nobody liked cannot quietly vanish from the record.

Letting outsiders see it without an account

Report share links is the feature that stops this data from living only in screenshots.

Enter an email address and the recipient gets a link unique to them. No Docutrain account, no sign-in. It expires after 30 days and can be revoked at any point. Each invite shows its status, expiry, and how many times it has been opened, and resending issues a fresh link while invalidating the old one.

What they land on is a standalone page carrying your logo and the document's title, with a Read-only badge and a sidebar covering Overview, keyword density, reports, and Doc. They can ask Doc questions and export the latest report as a PDF. They cannot generate, delete, or change anything. Open a link after it expired and it says so plainly instead of showing a partially loaded page.

For a client engagement, a funder, or an internal stakeholder who will never log into your dashboard, this is the whole reporting workflow in one email.

Across a whole library

The dashboard's own Intelligence tab aggregates everything across your library, with a breakdown listing questions, satisfaction and last activity per document and per collection. Click any row to drop into that scope.

One honest footnote lives there: conversations spanning several documents count toward each one, so that column can add up to more than your overall question count. The collections view ends with a Multi-document (no collection) row, and those totals do reconcile. Most analytics products would let the discrepancy pass unmentioned.

Collections get the same toolset once saved, covering every conversation held inside them.

The habit that pays off is generating a report at milestones rather than at random: before and after a content revision, or at the end of a training cohort. Comparable snapshots tell you whether the rewrite worked. The raw conversation logs are always underneath when you want to read the exchanges behind a flagged theme.