Effects of ChatGPT-assisted writing instruction on L2 writing performance

This mixed methods study investigated the effects of ChatGPT-assisted instruction on writing performance among Vietnamese first-year English majors in a 15-week writing course. It involved an experimental group (EG) that received ChatGPT-assisted instruction and a control group (CG) that received teacher-guided instruction without structured ChatGPT support. Participants’ performance was assessed with teacher-designed pre- and post-tests and a researcher-developed analytic rubric, complemented by four post-task reflections and semi-structured interviews with EG participants to explore perceptions and challenges. Quantitative results indicated that gains in lexical resource were significantly greater in the EG than in the CG (p = .001, r = 0.41), while no significant between-group differences were found in task achievement, coherence and cohesion, or grammatical accuracy. Within-group analyses showed improvement across all four dimensions in the EG and across three dimensions, excluding lexical resource, in the CG. Qualitative data revealed that learners appreciated ChatGPT as a scaffold for planning, drafting, and revising. However, students faced challenges of over-reliance on generic AI-suggested concepts, reluctance to adopt suggested vocabulary, stylistic misalignment, and concerns about authorship and integrity. This study offers a context-specific account of human-AI collaboration in process-oriented L2 writing instruction in Vietnamese higher education.