Human-Centered AI for Sustainable EFL Speaking Development: A Reflective Learning Intervention in Bangladesh

Sustainable educational technology should extend meaningful learning opportunities without displacing teacher judgment, learner agency, or equitable participation. This mixed-methods study examined a semester-long, technology-enhanced reflective learning intervention for undergraduate English-speaking development at a private university in Bangladesh. The intervention combined repeated speaking tasks, video-based self-review, teacher feedback, peer feedback, guided self-assessment, and supplementary ChatGPT-mediated prompts. Thirty-five students completed matched pre- and post-intervention speaking assessments across fluency and coherence, grammatical accuracy, vocabulary resource, pronunciation and intelligibility, and interaction and effectiveness; four focus-group discussions (FGDs) explored perceived processes, benefits, and challenges. Overall scores rose from M = 18.63 (SD = 2.44) to M = 21.97 (SD = 2.35), t(34) = 18.37, p < .001, with a very large within-group effect (d = 3.11). All five dimensions improved significantly. Thematic analysis indicated that replayable performance evidence and human feedback supported error noticing, goal setting, repeated practice, confidence, and participation. ChatGPT was useful for vocabulary exploration, alternative expressions, planning, and low-stakes prompting, but memorizing AI-generated scripts could weaken spontaneity and increase dependence. Sustainability was not directly measured; rather, the findings support a bounded sustainability interpretation. A low-material, reusable reflective cycle can extend practice beyond class and reduce dependence on continual teacher-produced resources, but its long-term viability depends on device and connectivity access, AI literacy, privacy safeguards, teacher oversight, and proportionate use that recognizes generative AI’s environmental costs. The study contributes a human-centered model in which AI augments a socially grounded reflective ecology instead of replacing it.