Hesham Aldamen, Mohamad Almashour, Mutasim Al-Deaibes, Rami Alsharefeen
Introduction: Speaking anxiety and limited opportunities for spontaneous English interaction remain persistent constraints for many EFL learners in Middle Eastern university settings. This interpretive qualitative study examines how a reflection-enhanced AI speaking design shapes learners' affect, motivation, and future self-concepts. Methods: Twenty-five Jordanian undergraduates completed a 12-week module in which they practiced weekly speaking with ChatGPT in voice mode (official interface; model label recorded at the time of use). A fixed system prompt configured a supportive, non-judgmental speaking partner, introduced the weekly topic, and triggered two brief in-task reflective questions. Data comprised weekly reflective journals, semi-structured interviews, and saved interaction transcripts. Results: Using reflexive thematic analysis, we identified three patterned orientations in learners' accounts and interaction traces: initial curiosity with cautious engagement, increasing emotional regulation and psychological safety during speaking, and strengthened future-oriented identity work linking present effort to imagined competent English-speaking selves. Discussion: Interpreted through cyborg psychology and complementary motivational frameworks, these findings suggest that embedding reflection into human-AI dialogue can support willingness to communicate and identity-related development alongside speaking practice.