AI or not to AI in Resource-Constrained Contexts? Exploring the Impact of AI-Mediated Feedback on the EFL Speaking Proficiency of Bangladeshi Undergraduate Learners

Abstract

Developing English-speaking proficiency remains challenging in Bangladesh because teacher-centred instruction, large class sizes, and limited opportunities for individualised feedback constrain the speaking practice of learners. Although AI-mediated feedback has shown promise for supporting language learning, its effectiveness in resource-constrained English as a Foreign Language (EFL) context remains underexplored. This study investigated the impact of SmallTalk2Me, an AI-mediated speaking platform, on the speaking proficiency, fluency, and perceptions of Bangladeshi tertiary-level EFL learners. Using a one-group pretest–posttest pre-experimental design, 28 undergraduate students completed IELTS speaking tests before and after a three-week intervention, while speaking fluency was measured using words per minute (WPM) and the perceptions of learners were examined through a post-intervention survey. The findings showed significant improvements in IELTS speaking scores (5.75–6.38, p <.001, d = 0.93) and speaking fluency (84.50–93.10 WPM, p <.001, d = 1.15), alongside positive perceptions of AI-mediated feedback. However, variation in learning outcomes, concerns regarding feedback reliability, and the lack of human interaction suggest that AI-mediated feedback is most effective when integrated with contextualised learning opportunities and teacher support. The study contributes empirical evidence from a resource-constrained EFL context and highlights the potential of AI as a complementary tool for speaking development rather than a replacement for teachers.