Evaluating an AI-mediated English conversation system for enhancing junior high school students’ speaking performance and self-efficacy in under-resourced contexts

Weak English proficiency is a significant challenge for junior high school students in China’s under-resourced areas, where educational resources are limited, largely due to the historical neglect of speaking instruction. This challenge is further compounded by assessment disparities as English-speaking ability has not been consistently evaluated across regions. While developed regions increasingly adopt integrated English-speaking components in senior high-school entrance exams, under-resourced areas continue to assess only reading, writing, and listening, thereby perpetuating a notable regional gap in communicative competence and affecting learners’ confidence. To address this challenge, this study developed an AI-mediated English conversation system (AMECS) based on Bandura’s self-efficacy theory for junior high school students. The study found statistically significant improvements in all the speaking test tasks. Qualitative interview data indicated that students held positive views on the system’s functionality, task activities, effects on confidence, and alignment with classroom instruction. The findings suggest that the system is technically feasible and educationally adaptable, with empirical evidence supporting improved English-speaking performance and self-efficacy. Accordingly, this study provides a practical pathway for combining AI with self-efficacy theory to enhance English-speaking practice in similar under-resourced educational environments.