Talking with Machines: Experimental Evidence on the Effects of AI-Supported Speaking Practice on Fluency and Anxiety

Abstract

This quasi-experimental study examined the effects of AI-supported speaking practice using ChatGPT Voice on oral fluency and speaking anxiety among 60 Jordanian EFL university students. Participants were divided into an experimental group engaging in AI-based sessions and a control group receiving traditional teacher-led instruction. Pre- and post-tests were administered to assess oral fluency using a speaking test and an analytic rubric, while speaking anxiety was measured with an adapted version of the ‘Foreign Language Classroom Anxiety Scale’. MANOVA results showed that the experimental group achieved significantly higher fluency scores across all speaking tasks compared to the control group. An independent samples t-test indicated that the AI-supported group reported significantly lower levels of speaking anxiety than the teacher-led group. Pearson correlation analysis revealed a strong negative relationship between post-test fluency and anxiety scores. MANCOVA confirmed that group differences remained significant even after controlling for pre-test performance. These findings suggest that AI-supported speaking practice can simultaneously improve oral fluency and decrease speaking anxiety among EFL learners. The results support theoretical views linking linguistic performance with emotional regulation, demonstrating the potential of conversational AI as a valuable tool for language learning. Pedagogical implications and directions for future research are discussed.