Speaker
Description
Argumentative writing is a cornerstone of academic literacy, yet remains challenging for EFL learners. Although ChatGPT shows promise for supporting it, research has focused on error-oriented feedback, leaving comparative configurations underexplored. This mixed-methods quasi-experimental study compared Socratic scaffolded and directive ChatGPT feedback on Vietnamese EFL university students' argumentative writing. Twenty-eight students in two proficiency-matched groups completed a four-week intervention of weekly writing tasks. Both conditions used the same model, differing only in feedback: scaffolding withheld correction, prompting learners to justify claims, supply warrants and address counter-arguments, whereas directive feedback supplied explicit corrections. Performance was assessed via pre/post-tests scored on Alnemrat et al.'s (2025) rubric, analysed through ANCOVA with pre-test scores as covariate, and inter-rater agreement was reported. Findings indicated that both models enhanced writing, but scaffolding better supported argument development, critical reasoning and independent revision, whereas directive feedback yielded greater accuracy gains. Interviews revealed that scaffolding fostered ownership of reasoning, while directive feedback was valued for immediacy. Conducted with institutional approval and informed consent, the study offers a principled basis for selecting AI feedback strategies.
Keywords: ChatGPT feedback, argumentative writing, Socratic scaffolding, EFL writing instruction, AI-assisted language learning
