Examining ChatGPT-supported speaking practice among Saudi EFL students: a quasi-experimental study

Abstract

Speaking remains a persistent challenge for Saudi EFL learners, as large classes, limited exposure, and speaking anxiety restrict opportunities for practice. This quasi-experimental study examined whether ChatGPT-supported speaking practice was associated with higher adjusted post-test oral proficiency scores than those obtained with traditional instruction. Two intact groups of undergraduates (N = 42) participated over twelve weeks: an experimental group (n = 20) engaged in typed ChatGPT-supported conversational scaffolding integrated with spoken classroom role-play and task-based oral production, while a control group (n = 22) followed conventional instruction. Speaking performance was assessed using a validated analytic rubric covering eight criteria: fluency, pronunciation, grammar, vocabulary, coherence and organisation, interaction and engagement, task completion, and delivery and intelligibility. A significant baseline difference between groups was observed; therefore, analysis of covariance (ANCOVA) was used to evaluate post-test performance while controlling for pre-test proficiency. After adjustment, the experimental group achieved higher overall speaking scores than the control group (p < 0.001, partial η² = 0.60). Descriptive patterns across rubric dimensions suggested broader improvements in oral performance. The combination of typed ChatGPT-supported scaffolding, immediate feedback, structured speaking tasks, and teacher-mediated interaction may have increased opportunities for oral production in classroom contexts. The findings suggest that guided use of ChatGPT can complement traditional instruction and support the development of EFL speaking performance in higher education settings.