Meta-analysis of 17 studies finds generative AI tools boost students' self-regulated learning

Effects were positive across the cognitive, metacognitive, motivational and behavioral dimensions of self-regulated learning. A multilevel analysis with CR2-robust inference found no significant difference between these dimensions (p = 0.564), so no single dimension gained clearly more than the others.

The size of the effect varied considerably between studies, so results differ across educational contexts. The authors conclude that generative AI works best as a scaffold that supports learners' own self-regulation rather than replacing it.

The study synthesized 17 studies to assess both the overall effectiveness of generative AI for self-regulated learning and its effects on four dimensions: cognitive, metacognitive, motivational and behavioral. It used a random-effects meta-analysis, plus a multilevel random-effects model with CR2-robust inference to compare dimensions.

Moderator analyses found that experimental design and intervention duration were linked to significant differences in effect size. Subject domain, intervention type, tool used and personalization did not produce significant differences.

The authors say the findings suggest generative AI can contribute to developing self-regulated learning, and stress using it as a scaffold that supports rather than replaces learners' active self-regulatory processes.