When better revisions do not mean better writers: Learning transfer and student engagement with automated feedback
Automated feedback can improve revised texts, but evidence that these gains transfer to later writing remains limited. This 16-week sequential explanatory mixed-methods study followed 28 Indonesian undergraduate EFL students across six writing tasks. Initial and Grammarly-assisted revised drafts were rated with the ESL Composition Profile and analyzed using linear mixed models. Qualitative evidence came from 54 critical screen-recording episodes (approximately 80 minutes) and video-stimulated recall interviews. Revised drafts obtained higher domain-level scores than initial drafts, F (1,1643) = 469.413, p < .001, approximate partial η² = .222. After scores were expressed as percentages of each domain maximum, the largest descriptive gains occurred in mechanics (30.72 percentage points), language use (17.86), and vocabulary (13.28), whereas content and organization changed minimally. The previous revised-draft score did not significantly predict the next initial-draft score after controlling for topic differences, F(1, 130.803) = 1.273, p = .261. The qualitative analysis showed that many successful revisions were produced through rapid acceptance, shallow processing, self-doubt, and reliance on Grammarly; reflective evaluation and authorial ownership appeared in fewer episodes. The findings distinguish within-task text improvement from near cross-task transfer and show why feedback uptake alone is insufficient evidence of writing development.