AI-Generated End-of-Module Practice: The Performance Lever Your Training Hasn't Activated Yet

Your learners have just finished a module. They answered exercises, received feedback, made progress. What if this exact moment, right after learning, while the memory trace is still fresh, turned out to be the most strategic one for making what they just learned stick? An impact study conducted at Didask measured precisely the effect of a personalized exercise, automatically generated by AI at the end of each module. The results are clear: it boosts learner performance, and it costs course designers nothing.

What our impact study measured

To carry out this study, just under 200 learners took part in a training course split into two parts. Half of the learners had the AI activated for the first part of the course but not the second, and vice versa for the other half.

The AI condition included several types of interaction: conversational open-ended questions built into the course, spontaneous interactions with the assistant, and practice automatically triggered at the end of each module. After completing their training, learners took a performance test, our key measure.

The study then identified which types of interaction best predicted the results. This is where the signal becomes particularly clear.

End-of-module practice: the strongest predictor of performance

End-of-module practice significantly predicts learner performance (β = 0.316, p = 0.007). The more modules a learner completes with practice, the higher their performance, regardless of their initial motivation (β = 0.424, p = 0.024).

Learners who engage with this practice score significantly higher than those who don't (p = 0.0003).

An effect independent of motivation

In our models, learners' initial motivation is not statistically significant. It is actual engagement with end-of-module practice, not the starting profile, that explains the performance gain.

Why it works: active manipulation and a metacognitive pause at the right moment

This result can be explained by what happens cognitively at the end of a module. The learner has just absorbed new information, and what's needed now is a moment to pause and actively work with it: rephrasing it, putting it in context, testing their understanding in a situation slightly different from the course.

This is exactly what end-of-module practice triggers. The learner doesn't reread: they answer, they justify their answer, they receive personalized feedback. This moment of forced activity acts as a metacognitive pause, letting the learner check what they have actually understood, not just what they feel they have retained.

Studies have repeatedly shown that exercises interspersed throughout a course produce better results than a single test at the end (Kornell & Bjork, 2008). End-of-module practice embodies this principle: it breaks the effort into several distributed moments, rather than concentrating it in a final assessment.

The spacing effect in practice

Our study also confirms that learners who start practicing early in the course get better results than those who concentrate their use toward the end of training (r = 0.271, p < 0.001). Practice distributed over time is more effective than a massed revision effort, even at an equivalent volume of use.

What this means for course designers: zero workload, maximum impact

On most platforms, designing exercises that are relevant, varied, and paired with immediate feedback is the most time-consuming part of instructional design. It's also the part that non-specialist teams tend to neglect for lack of time.

In Didask, end-of-module practice is generated automatically by AI from the content provided by the subject-matter expert. The questions are conversational, adapted to the learner's level, and paired with personalized feedback. The designer has nothing to produce manually for this step: the AI handles it, drawing on the same cognitive principles documented in this study. If you'd like to learn more, other articles cover Didask's impact on training.

This is what an AI grounded in cognitive science means in practice: not a tool that generates content faster, but one that generates content that works better, because it's designed to activate the right learning mechanisms.