How Generative AI Affects Chinese College Students’ English Academic Writing Skills

Department of Chinese Study, Faculty of Arts, Chinese University of Hong Kong, Hong Kong, 999077, China

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Generative AI (e.g., ChatGPT, Grammarly, QuillBot) is rapidly diffusing into English for Academic Purposes writing, yet its relationship with learners’ self-efficacy, perceived ability, cognitive load, and dependence remains unclear. This cross-sectional survey (N = 93 Chinese undergraduates) quantified one-week AI usage (days/hours, stages, tools), writing self-efficacy (SE), self-rated ability across four EAP dimensions (AB: content/argumentation; structure/coherence; evidence/citation; language/style), cognitive load (CL), feedback-literacy behaviors (editing/verification), and perceived AI dependence (DPN). Results show a significant group difference in AB: high-frequency AI users rated their academic writing ability higher than low-frequency users (one-way ANOVA, p = .041), and AI use time correlated positively with AB (r ≈ .30). Perceived dependence was higher among high-frequency users than lowfrequency users (t-test, p < .01), and AI use time correlated with dependence (r = .369, p < .001). A multiple regression predicting DPN from CL, SE, and AB was significant but small in magnitude (R² ≈ .08); cognitive load uniquely and positively predicted dependence (p = .025). Non-significant associations (e.g., SE with AI use) are discussed as design targets for future work. The author argues for calibrated, feedback-literate AI use and planned “AI-off” practice to preserve autonomous writing competence while harnessing efficiency gains.