AI‐Enhanced Feedback and the Multidimensional Development of Oral Performance in EMI Instructors: A Randomized Longitudinal Study

ABSTRACT

This study examined whether AI‐generated feedback supports longitudinal change in selected indicators of oral performance among English‐medium instruction (EMI) instructors. In a randomized controlled longitudinal design, 100 university lecturers completed a ten‐week speaking program in either an AI‐feedback or practice‐only condition. Both groups produced weekly unscripted academic monologues. Week 1 provided a common no‐feedback baseline; only the AI group received subsequent reports containing the platform‐labeled Accuracy, Fluency, and Pronunciation scores. Linear mixed‐effects models showed steeper trajectories under feedback for Fluency and Pronunciation, whereas Accuracy showed a smaller, statistically uncertain differential trajectory. Residual and growth‐level associations were moderate but followed partially different configurations, suggesting that relationships within weekly performances need not describe corresponding change across participants. By tracing change across repeated speaking occasions and subsequent feedback opportunities rather than relying on endpoint comparisons, the study identifies differential longitudinal change in separately reported outcomes, with more qualified evidence that their effects differ. The findings extend AI‐supported speaking research to EMI instructor development by examining how report access adds to repeated practice across successive first attempts.