Effects of Artificial Intelligence Feedback on Students’ Self-Regulated Learning in Higher Education: A Three-Level Meta-Analysis

Abstract

As higher education environments become digitized and complex, self-regulated learning (SRL) remains a pivotal cornerstone of student achievement. While artificial intelligence (AI) offers new possibilities for supporting SRL through feedback, its efficacy remains controversial. A three-level meta-analysis of 85 articles (k = 387; N = 9,564) published between 2015 and 2025 was conducted. The results demonstrated a significant, medium-sized positive effect of AI feedback on students’ SRL (g = 0.509). Dimension-specific analyses revealed that effects were stronger for the motivational and cognitive dimensions, but comparatively weaker for behavioral and metacognitive outcomes. Notably, the wide 95% prediction interval [-0.661, 1.678] indicates substantial heterogeneity, suggesting that AI feedback effects are conditional rather than uniformly beneficial. Hierarchical multilevel meta-regression identified feedback source and AI technology type as primary drivers of between-study heterogeneity. AI-led feedback yielded larger effect sizes than human-AI hybrid feedback, and generative AI dialogue systems and intelligent tutoring systems demonstrated stronger effects than non-generative AI dialogue systems and learning analytics systems. Feedback modality and academic subject yielded marginally significant or dimension-specific differences. Additionally, within-category analyses indicated that the effects of generative AI dialogue systems were not significantly moderated across conditions, with the exception of a marginal effect for feedback modality. However, within the learning analytics systems group, the study design demonstrated a significant moderating role. Overall, these findings not only quantify the efficacy of AI feedback on SRL outcomes, but also delineate its boundary conditions, offering empirically grounded implications to instructors, designers, and researchers.