Enhancing EFL learners’ speaking performance through ChatGPT-assisted practice: evidence from complexity, accuracy, and fluency measures

Abstract

This study examined whether ChatGPT-assisted speaking practice could strengthen English as a Foreign Language speaking development in an IELTS preparation context. Sixty-five intermediate Iranian learners participated in a ten-week quasi-experimental intervention, with one group receiving structured ChatGPT-supported speaking practice and a comparison group receiving teacher-led instruction. Learner performance was assessed through IELTS-aligned speaking tasks and analyzed using complexity, accuracy, and fluency indicators. The findings showed a clear pedagogical advantage of ChatGPT-assisted practice for syntactic complexity and fluency, while accuracy improved over time in both groups without a distinct between-group advantage. These results suggested that large language model-based interaction could create conditions that encouraged longer turns, richer language production, and faster speech delivery, particularly when learners received sustained opportunities for low-pressure, adaptive oral practice beyond limited classroom speaking time. At the same time, the pattern for accuracy indicated that generative AI should be integrated as a complement to, rather than a replacement for, teacher guidance, especially for form-focused feedback. The study contributed evidence from an underrepresented EFL context and supported a blended instructional model in which ChatGPT expanded speaking opportunities, promoted learner autonomy, and enhanced key dimensions of oral performance within communicative language teaching.