Tracing Interactional Change in Japanese EFL Learners’ Interactions with a Teacher-Designed AI Voice Chatbot: A Pilot Study

Although AI-mediated conversation practice is increasingly common in EFL teaching, qualitative analysis of learner–AI interaction has been insufficient. This exploratory pilot study examined Japanese EFL learners’ interactional change. Ten university students completed three five-minute sessions with a teacher-designed chatbot on Google AI Studio. Logs were analyzed quantitatively and qualitatively. Effect sizes suggested a two-phase pattern of observed change: large effects (d = 0.88–1.13) from Sessions 1 to 2 marked an initial adjustment phase, while medium effects (d = 0.59–0.66) from Sessions 2 to 3 indicated slowed quantitative growth alongside continued qualitative elaboration, particularly in topic development. Qualitative analysis identified both constraints (vocabulary gaps, abstract questions) and affordances (explanatory expansions, analytical reflections, self-directed topic drift). Comparison with peer-to-peer data highlighted distinctive features: longer turns, near absence of backchannels, and more frequent language-related episodes. These preliminary findings offer initial guidance for teachers developing AI conversation tools using accessible platforms.