Abstract
<p>With the rapid advancement of digital technologies, artificial intelligence (AI) has evolved from early intelligent tutoring systems to contemporary generative AI. Although AI features—such as personalization, adaptability, and instant feedback—have the potential to facilitate learners’ metacognition, empirical findings regarding the effectiveness of AI remain inconclusive. To address this gap, the present study employed a three-level meta-analysis of 54 studies (k = 121; N = 7,431) to examine the effects of AI interventions on metacognition. The results revealed a significant positive effect of AI on metacognition (Hedges’ g = 0.624, SE = 0.099, p &lt; .001, 95% CI [0.431, 0.817]). Moderator analyses demonstrated that the control condition, intervention duration, and data source significantly moderated this effect, whereas other study, methodological, participant, and AI characteristics did not. These findings quantify the overall impact of AI on metacognition and identify key boundary conditions, thereby providing valuable insights for future research and educational practice.</p>