A policy manual everyone has opened is not the same thing as a policy manual everyone understood. The gap between those two is where compliance training goes to die.

Docutrain closes some of it by writing a multiple-choice quiz from a document's own content. Questions come from the material, each one carries an explanation and page citations, and it takes about a minute of your time to produce.

The detail that makes it a real test

Answers are graded on Docutrain's servers.

That sounds like plumbing and is the whole difference. The answer key never reaches the reader's browser before they commit to a choice, so a curious person with the developer console open cannot read the correct answers out of the page. Select an option and it locks immediately, with no changing your mind after seeing the result.

Plenty of embedded quiz widgets ship the answers to the client and hope nobody looks. If you intend a quiz to mean anything, this is the property you need.

Taking one

When a quiz is available, a Quiz button sits among the document's chat tools, described as "Test your knowledge". No account is needed on a public document.

You choose how many questions to take, from 5 up to 20, and lengths larger than the available bank are not offered, so a short document simply shows fewer choices. Questions are drawn at random from the larger bank, so two attempts rarely match.

Each question shows its options, a progress line, and a row of dots. Answer, and feedback appears underneath straight away: correct or incorrect, the right option in green and a wrong pick in red, an explanation written from the document, and Page citation chips you can hover to see the source passage.

Those chips are the part that turns a quiz into a study tool. Getting a question wrong tells you that you were wrong. Being handed the paragraph tells you why.

Finishing opens a review screen with your score, a percentage, and a Missed only filter that collapses it down to what you got wrong, each with its explanation and page references. Attempts are unlimited, and every completed one is recorded.

Generating the bank

Quiz management lives on the document's Quiz tab. With nothing there yet, one button generates the bank.

You do not choose the size. Docutrain works it out from the document, roughly one question for every two sections of content, with a floor of 10 and a ceiling of 100, sampled across the whole document rather than concentrated wherever the model found it easiest. A typical document takes well under two minutes, and it runs as a background job with automatic retries, so you can close the editor and watch from the dashboard.

While it runs you get a real progress panel: a bar, a count like "12 of 40 questions done · 28 remaining", elapsed time with an estimate, and a cancel button. If progress pauses, a small indicator shows how long it has been since the last activity, which is the difference between waiting patiently and wondering whether it died.

Generating requires Pro or above. Existing banks stay readable after a downgrade, and only creating a new one is blocked. See plans and limits.

Editing, and the versioning underneath it

The generated questions are good, not perfect. Read them before you turn the quiz on.

Expanding a question shows its options with the correct one marked, the explanation, its page chips, and a version badge. Edit lets you rewrite the question text, change the options (between two and six) and which is correct, and rework the explanation.

Here is the part worth admiring. Saving an edit creates a new version, and the previous one is retired rather than deleted. So a reader's past attempt still shows exactly the question they were asked, in the wording they were asked it, months after you fixed a typo in it.

Anyone who has tried to audit a training record against a question bank that was edited in place will recognize why that matters. Your historical results stay meaningful.

Regeneration works the same way. It replaces the whole bank, which is what you want after retraining has changed the content, and old questions are retired rather than destroyed, so past attempts and statistics survive intact. It is normally limited to once every seven days, with the message telling you when the bank was last built and when the next run is allowed.

One caution: a reader partway through an attempt when you regenerate may be unable to submit it. They can retake, but avoid regenerating during a live training session.

Turning it on, and watching the numbers

Generating questions shows readers nothing by itself. A separate Quiz toggle in UI Options is what puts the button in the chat, and it stays disabled until questions exist.

Once people start, the statistics panel reports total attempts split between signed-in and anonymous, average score, highest and lowest, and questions per attempt.

The average is the one to watch, and to read carefully. A low average from an audience you know understands the topic is usually not telling you about your audience. It is telling you that the source material is confusing, or that the questions are drawn from a badly written section. Document Intelligence will help you localize which.

Four habits

Generate after training finishes, and regenerate after any retrain that changed the content.

Read the bank before switching the quiz on, and fix weak individual questions by editing them rather than regenerating everything, since regeneration throws away your improvements along with the problems.

Mention the quiz in the document's welcome message. A tool nobody notices is a tool nobody uses.

And watch the averages rather than the attempt count. Attempts tell you people found it. Scores tell you whether your document worked.