Integrating AI literacy into the TAM-TPB model to explain students’ intention to use educational AI through MASEM approach

The Technology Acceptance Model (TAM), the Theory of Planned Behaviour (TPB), and AI Literacy (AIL) are widely used to study students' behavioural intention (INT) to adopt educational AI (EAI). However, existing evidence on their interrelationships remains inconsistent and often contradictory. To address this, we applied meta-analytic structural equation modelling to integrate TAM, TPB, and AIL into a unified framework, synthesizing data from 166 studies across 21 paths. The AIL-integrated model achieved the best fit and strongest explanatory power, with AIL emerging as the most influential predictor of INT (β = 0.408), followed by perceived usefulness (PU, β = 0.330). Mediation analysis confirmed that AIL influenced INT through PU, PEU, and attitude. Notably, PEU had a weak effect on attitude (β = 0.086), indicating its limited role in shaping students' perceptions of EAI. Subgroup analysis revealed that economic development and academic stage moderated key pathways, underscoring contextual influences on students' INT. Theoretically, this study advances TAM and TPB by identifying AIL as a decisive antecedent of intention, shifting attention from external tool utility to internal user capability. Practically, it highlights the need to cultivate AIL through teacher training, ethics integration, and experiential learning. In developing regions, priority lies in foundational literacy and social support; in developed contexts, in curricular integration and reflective engagement. At the K-12 level, social demonstration and introductory programs are essential for fostering students’ INT to adopt EAI, while in higher education, project-based courses and research opportunities strengthen competence and sustain adoption. • AI literacy is the strongest predictor of students' intention to use educational AI. • Integrating AI literacy with behavioral models greatly improves prediction accuracy. • Meta-analytic structural modeling resolves earlier inconsistent findings. • Economic and academic contexts moderate key paths in intention toward educational AI. • Factors shaping students' intention shift from AI tool value to self-capability.