Multimodal GenAI-supported personalization: engagement, autonomy, and L2 performance

Purpose: Rapid advancements in multimodal Generative Artificial Intelligence (GenAI) are reshaping language learning by offering adaptive, interactive, and personalized support across text, audio, visual, and hybrid modalities. Although personalization is widely recognized as a key feature of GenAI-enhanced instruction, empirical evidence clarifying how such systems relate to learners' engagement, autonomy, system perceptions, and perceived language development remains limited. This study investigates the relationships among GenAI-based multimodal personalization, cognitive engagement, affective engagement, learner autonomy/agency, system usability and trust, and self-reported L2 performance in GenAI-supported learning environments. It further examines the extent to which personalization, engagement dimensions, autonomy, and system usability and trust predict learners' self-reported L2 performance.Design/methodology/approach: Data were collected from 831 EFL learners using a validated multilingual questionnaire and analyzed through structural equation modeling (SEM).Findings and Originality/value: The findings indicate significant associations among personalization, engagement, autonomy, system perceptions, and performance. GenAI-based multimodal personalization positively relates to both cognitive and affective engagement as well as learner autonomy. Additionally, cognitive engagement, affective engagement, learner autonomy/agency, and system usability and trust significantly predict self-reported L2 performance. The results underscore the importance of adaptive multimodal design and positive system perceptions in supporting meaningful engagement and perceived language development. The study contributes to a clearer understanding of how GenAI-supported personalization functions within L2 learning environments and offers practical implications for the design of learner-centered AI systems.