The Role of Model Prompts and Feedback Seeking Behavior in ChatGPT-Assisted EFL Writing

The increasing use of generative AI (Gen-AI) tools in education is attracting scholarly attention, particularly regarding their impact on second language (L2) writing development. While Gen-AI provides cognitive and motivational support, concerns persist about overreliance and diminished learner autonomy. This study examines how prompt design affects the quality of AI-assisted writing revisions and learners’ feedback seeking behavior (FSB), which serves as an indicator of learner agency. Forty-three undergraduate EFL students were randomly assigned to one of two conditions: a direct feedback model prompt (DMP) or an indirect feedback model prompt (IMP). After writing initial drafts independently, participants revised their work using ChatGPT. The results indicated significant improvement within-group, especially in language use and vocabulary. Although there were no statistically significant differences in revised-draft scores between the groups, the DMP group performed better in mechanics and language use, with small to moderate effect sizes. Regarding FSB, while group differences were not statistically significant, effect size analysis showed that the DMP group was more engaged in seeking feedback. These findings suggest that prompt design can influence learner-AI interactions and enhance learner agency, providing valuable insights for integrating Gen-AI tools into L2 writing instruction.