Artificial intelligence-driven adaptive learning platforms (AI-ALPs) have gained significant attention in educational technology for their capacity to address learners’ individualized needs. However, the impact of AI-ALPs on students’ academic achievement remains controversial, with empirical findings presenting inconsistent results. To address this issue, this study conducted a meta-analysis, synthesizing 47 effect sizes from 27 empirical studies. The main effects analysis revealed that AI-ALPs had a statistically significant but small positive impact on student academic achievement (Hedges’ g = 0.276, 95% CI = [0.197, 0.356]). Further analysis of moderating effects revealed: (1) No significant difference between multimodal and text-based interactions in improving academic performance; (2) Feedback modality significantly moderated learning outcomes, with mixed feedback outperforming both detailed and simple feedback; (3) AI-ALPs had positive effects across all academic levels, with the strongest impact observed at the secondary school level; (4) The platform was more effective in science and engineering domains than in humanities and interdisciplinary fields; These findings provide a strong empirical foundation for understanding the functioning of AI-ALPs across varied educational contexts. In addition, this study offers theoretical guidance and practical insights for designing precise instruction and enhancing human-computer collaboration.