AI-mediated teacher learning in Chinese lesson study: regulating cognitive demands in EFL pre-service teachers

Purpose This study, guided by the cognitive theory of multimedia learning (CTML), investigates how multimodal large language models (MLLMs) mediate the professional learning of English as a Foreign Language (EFL) pre-service teachers within Chinese lesson study (CLS). It aims to elucidate the cognitive and dialogic mechanisms through which AI-supported mediation shapes teacher learning processes and participation. Design/methodology/approach A three-month intervention integrating MLLMs into CLS was conducted with 50 EFL pre-service teachers at a Chinese normal university. Adopting an interpretive qualitative design, data were collected from longitudinal reflective journals, MLLM interaction logs, field notes and two rounds of semi-structured interviews with 5 purposively selected focal participants. Data analysis combined deductive coding based on CTML (intrinsic, extraneous and germane cognitive load) with inductive thematic analysis to capture emergent patterns of AI-mediated learning and interaction. Findings The findings reveal that MLLMs functioned as cognitive-dialogic mediators in three key ways: (1) reducing linguistic and organizational barriers, thereby lowering extraneous cognitive load; (2) redistributing cognitive resources to support meaning-making and pedagogical reasoning; and (3) enabling deeper engagement in collaborative knowledge construction and reflective inquiry within CLS. However, participants also reported tensions related to epistemic reliability and the risk of overreliance on AI-generated outputs, indicating a shift in cognitive demands from knowledge construction to evaluation and regulation. Originality/value This study advances the field by proposing a cognitive-dialogic framework that brings CTML into dialog with sociocultural perspectives on mediation, offering a nuanced account of how AI reshapes teacher learning in practice-based settings. It contributes empirical evidence on the role of MLLMs as mediating tools in CLS and provides design implications for cognitively and ethically informed AI integration in teacher education.