Positive Perceptions Are Not Enough : Reflection and Performance in ChatGPT-Assisted EFL Writing

Abstract

This study examines Korean EFL undergraduates’ use of ChatGPT-generated feedback in a general education English writing course at a Korean university, focusing on learner perceptions, reflection depth, and independent writing performance. Sixty students completed an anonymous perception survey measuring affective evaluation of ChatGPT feedback and cognitive awareness of AI limitations. In addition, a subset of 36 students with complete reflection texts and final writing task scores was included in the reflection-performance analysis. The reflections were coded using a three-level rubric, and scores from the final writing task were used as an indicator of independent writing performance. The results revealed a mismatch between students’ positive evaluations of ChatGPT feedback and their relatively limited awareness of AI limitations. Although students showed relatively high affective evaluations, their cognitive awareness was lower, and the two dimensions were not significantly correlated. This suggests that favorable attitudes toward ChatGPT feedback did not necessarily correspond to a critical understanding of its limitations. Reflection analysis further showed that many students engaged with ChatGPT feedback at a descriptive level, while fewer demonstrated analytical or critical reflection. In addition, reflection depth showed a marginal positive association with independent writing performance (ρ = .320, p = .057), which can be interpreted as a tentative tendency. Overall, the findings suggest that positive perceptions of ChatGPT feedback alone are not sufficient to ensure critical engagement or independent writing development, highlighting the need for structured pedagogical scaffolding that helps EFL learners evaluate, interpret, and selectively apply AI-generated feedback in general education writing contexts.