AI-mediated oral practice and the dynamics of anxiety, performance, and self-regulation: A mixed-methods comparison of instructional modes

AI-mediated oral-practice tools offer scalable, low-stakes opportunities for second-language (L2) speaking practice, yet evidence of their pedagogical and affective impact remains limited. This quasi-experimental study investigated whether AI-mediated instruction enhances foreign-language anxiety (FLA), oral performance, and self-regulated learning (SRL) compared with blended and traditional human-taught instruction. A total of 150 learners (50 per group) from the English Department of Batna 2 University completed a six-week intervention integrating the AI pronunciation and fluency application ELSA Speak. Preand post-tests measured FLA, oral performance, and SRL. ANCOVAs controlling for baseline scores assessed adjusted group differences, and qualitative thematic analysis of reflective journals provided explanatory depth. Results revealed significant main effects of instructional mode on all outcomes: FLA (F(2,146) = 32.47, p .001, partial η² = .31), oral performance (F(2,146) = 23.68, p .001, partial η² = .25), and SRL (F(2,146) = 31.02, p .001, partial η² = .30). AI-mediated learners reported markedly lower anxiety and higher performance and SRL than both comparison groups, with large standardized effect sizes (Hedges' g ≈ 1.1-1.3). Qualitative findings identified four interrelated themes-emotional regulation, self-regulatory growth, pronunciation and fluency development, and technological motivation-highlighting how supportive feedback and autonomous practice fostered engagement and confidence. These results demonstrate that AI-mediated oral practice can meaningfully reduce anxiety, strengthen self-regulation, and improve L2 speaking outcomes within a short intervention. The study advances understanding of AI as an active pedagogical agent and supports its thoughtful integration alongside human instruction.