EFL learners’ epistemic agency in AI-supported digital storytelling: a rhizomatic analysis

Generative AI (GenAI) expands English as a foreign language (EFL) learners’ resources for multimodal composing, a practice frequently operationalized through digital storytelling (DST) to express personal and cultural narratives. However, because algorithmic outputs risk linguistic or cultural flattening, GenAI integration raises critical epistemic concerns for culturally appropriate meaning-making. This study investigates how 25 university EFL learners enacted epistemic agency while producing culture-themed AI-supported DST videos, a task designed to surface GenAI’s limitations and necessitate human intervention. Data comprised 25 artifact-elicited retrospective interviews, 12 video products, and evaluations from six EFL instructors. Grounded in rhizomatic theory, a two-layer analytic framework mapped foundational nodes, asignifying ruptures, and learners’ epistemic moves within the non-linear composing process. Findings reveal that while learners conditionally enacted acceptance for routine AI efficiencies, asignifying ruptures (e.g., factual inaccuracies, cultural bias, and aesthetic limitations) catalyzed an active epistemic shift. To repair these breakdowns, learners enacted critical evaluation, verification, and adaptation, dynamically expanding their networks through social, cultural, and technical nodes. Although this active network enabled learners to reassert their narrative voice, evaluations of the final DST products revealed that seamless multimodal orchestration proved challenging. Theoretically, observing agency through a rhizomatic lens yields an agency-anchored framework that translates abstract human-AI interactions into empirically mapped decision work. Pedagogically, the findings redefine practical AI literacy as algorithmic gatekeeping, urging educators to scaffold learners in mobilizing external networks to critically assert human authorship over machine generation.