Text-Level Effects of AI-Supported Revision in EFL Writing: Evidence from Score Changes and Revision Patterns

This study examines the text-level effects of AI-supported revision in EFL writing through evidence from score changes and revision patterns. Thirty Vietnamese EFL university students wrote an initial draft without AI support and then revised the same draft with ChatGPT. The drafts were assessed using a 10-point analytic rubric, and the revisions were coded by type. The results showed a modest but statistically significant increase after AI-supported revision, t(29) = 3.01, p = .005. The mean score increased from 6.5 to 6.7, with a mean gain of 0.25 points. However, the changes were uneven: 19 students improved, six showed no change, and five received lower scores. Most revisions involved grammar correction, vocabulary replacement, and sentence clarification. Idea expansion and organization changes were less frequent. The findings suggest that AI-supported revision mainly affects surface-level features of EFL writing. Therefore, AI-supported drafts should be examined through both score changes and revision patterns, not only through final text quality.