AIGC-Assisted Critical Writing Feedback — A Double-Blind Randomised Controlled Trial at Fudan University

Researchers at Fudan University's Informatization Office ran a double-blind, pre–post randomised controlled trial with 259 eligible undergraduates (125 treatment, 134 control), screened for native Mandarin proficiency, intermediate English competency (CET-4 425–550) and no prior AI-writing-tool experience.

The trial tested whether AI-generated feedback could improve undergraduate critical-writing quality compared with traditional instructor-only feedback.

The treatment group received feedback from a fine-tuned Qwen-7B-Chat model customised for academic-writing evaluation over a four-week intervention; the control group received traditional instructor-only feedback.

Using difference-in-differences analysis and structural equation modelling, the study found a significant overall treatment-by-time effect on writing performance (β=0.149, p<0.001). Effects were strongest on essay organisation (β=0.311, p<0.001) and content development (β=0.191, p<0.001), with smaller but significant gains in language usage (β=0.070, p<0.05) and self-reported writing motivation (β=0.077, p<0.05). A technology-acceptance pathway analysis further linked perceived ease of use to perceived usefulness (β=0.326, p<0.001) and usefulness/attitude to actual system usage (β=0.431, p<0.001).

The trial was conducted at a single institution over one four-week cycle, so evidence of durability or transfer beyond this cohort and course is not yet available; the study itself does not report any classroom-scale rollout beyond the trial.