Learning Styles, Engagement and Anxiety in AI‐Mediated Writing: A Multimodal Feedback Study

ABSTRACT Artificial intelligence (AI) tools now permeate English academic writing. However, evidence on how feedback modalities align with student differences and with psychological mechanisms remains limited. Prior work often reduced learning styles to simple matches with delivery modes and treated learning engagement and writing anxiety as peripheral. A unified account that connects multimodal AI feedback, perceptual styles, engagement, anxiety, and writing performance has remained scarce. This study addressed these gaps through a cross‐sectional survey of 358 Mandarin L1 undergraduates from two Malaysian universities that measured text‐based (TF), linguistic (LF), and interactive feedback (IF); four perceptual styles (visual, auditory, kinaesthetic and tactile); learning engagement; writing anxiety; and writing performance. Instrument quality exceeded conventional thresholds; exploratory and confirmatory factor analyses supported construct validity, and a structural equation model with tests for mediation and moderation estimated the proposed paths. TF showed no direct effect on performance, whereas LF and IF exhibited positive effects. All styles predicted performance, with kinaesthetic being the strongest, followed by visual, tactile, and auditory. Engagement increased under all feedback and style inputs and served as a mediator; anxiety strengthened most paths to performance, with no significant interaction for the tactile style. The evidence rejected a simple style‐tool match and supported a layered model in which discourse‐orientated and interactive feedback, aligned with embodied activity, improved writing quality. Teachers and developers should stage support from text‐based cues toward linguistic and interactive guidance, pair kinaesthetic or visual tasks with those channels, and tune intensity to students’ anxiety profiles.