Beyond Model Prompts: Developing Writer Agency in Japanese EFL Learners’ ChatGPT-Mediated Writing

Abstract Generative artificial intelligence (GenAI) tools such as ChatGPT are increasingly being used in L2 writing classrooms; yet research has prioritized perceptions or performance gains, with limited attention given to how interactions with AI evolve and shape writer agency. Guided by sociocultural theory, this qualitatively driven mixed-methods study examines Japanese university EFL learners’ changing engagement with ChatGPT in a semester-long academic writing course. Qualitative data included a teacher-provided model prompt, students’ self-generated prompts and justifications, and end-of-semester reflections. A supplementary quantitative strand compared changes in complexity, accuracy, and fluency between the focal group and a control group that completed the same course prior to the availability of ChatGPT. Thematic analyses show that early prompting closely mirrored the model and emphasized external norms such as native-like expression, whereas later prompts shifted toward learner-defined goals, including clarity and readability. Learners’ reflections further revealed contrasting orientations toward GenAI: roughly half of the participants described ChatGPT as a thinking partner that required selective uptake, while others expressed concern about diminished cognitive engagement, perceiving ChatGPT as a surrogate thinker. These patterns suggest that GenAI engagement depends on how decision-making responsibility is negotiated between writer and AI. Quantitatively, only fluency showed significant gains. Overall, the findings indicate that learners appropriated ChatGPT as a mediational tool while increasingly asserting agency and resisting standardization, underscoring the pedagogical value of fostering critical prompting and reflective, agency-aware AI use.