Abstract
This study employs a meta-analytic approach to synthesize empirical evidence on the impact of AI-assisted learning on the agency of foreign language learners. The overall synthesis indicates positive associations between AI-assisted learning and multiple dimensions of learner agency; however, the magnitude of these associations varies substantially across studies. Among these outcomes, engagement shows the largest pooled effect size (r = 0.648), whereas enjoyment demonstrates the smallest (r = 0.392). Due to extreme heterogeneity, these pooled estimates serve only as descriptive summaries of the literature rather than evidence of robust effects, as variability significantly constrains their interpretability. Moderator analyses and meta-regression were conducted to explore potential sources of this heterogeneity. Although subgroup analyses reveal that neither learners' first-language background, educational level, nor AI tool type significantly accounts for between-study variability, indicating that contextual factors likely shape outcomes in complex ways. These findings underscore the robust potential of AI-assisted learning while emphasizing the importance of investigating specific conditions-such as task design, teacher intervention, and learner profiles-that optimize agency development. Future research should move beyond global effect estimates toward context-sensitive strategies for maximizing AI's impact on learner agency.