From ChatGPT to Classroom Practice: A Visualized Protocol for Generative AI-Supported Feedback, Critical Thinking, and EFL Writing Instruction

What this study set out to examine was whether a visualized, protocol-guided approach to ChatGPT-assisted writing instruction, termed VP-CGWIP, was associated with more favorable writing outcomes, sharper critical judgment, and more constructive learner perceptions than unstructured AI use in an EFL writing course. Compared, across a ten-week intervention and a four-week delayed follow-up, were two intact sophomore classes at a Hangzhou university, using a quasi-experimental, three-timepoint mixed-methods design in which writing performance was scored against the IELTS Task 2 rubric, critical thinking disposition was measured with the CTDI-CV, actual judgment ability was assessed via a standardized AI-feedback detection task, and process data were drawn from an AI Feedback Adoption Log and reflective journals (complete-case analytic sample: experimental n=43, control n=42). Larger gains in writing performance from baseline to immediate posttest, retained only in part at delayed posttest, were shown by the experimental group, guided by a seven-step closed-loop protocol combining structured Accept/Modify/Reject judgment, critical dialogue, and reflective journaling; markedly higher and more durable, at both posttests, was this group's accuracy in detecting flawed AI feedback. Concentrated in analytic and self-regulatory subdimensions, rather than spread evenly, were the gains in critical thinking disposition, and toward conditional modification and externalized rationale did the thematic analysis of journals and adoption logs point. Correlated with writing gains were perceived usefulness and behavioral intention, though not perceived ease of use. Because intact classes rather than individual-level randomization underlay the design, these findings are best read as exploratory associations rather than causal effects. Suggested nonetheless is that structuring how learners engage with AI feedback, rather than merely permitting or restricting access to it, may be a promising direction for EFL writing pedagogy, pending component-based studies and multi-site replication with individual-level randomization.