The Impact of AI Assistance on the Academic Writing Self-Efficacy of Undergraduates with Different Writing Abilities

Against digital education background, generative AI has become common writing assistance for undergraduates. Existing studies mainly focus on second-language writing and ignore stratified comparisons across learners, while the dual effects of AI on academic writing self-efficacy remain understudied. Based on self-efficacy theory, this paper distributes questionnaires and divides respondents into three proficiency groups by CET-4 scores. Descriptive statistics, group difference tests and correlation analysis are adopted for empirical analysis. The results reveal prevalent AI over-dependence despite its polishing functions. Participants’ writing self-efficacy stays at medium-upper level yet hampered by writing anxiety. AI delivers short-term confidence promotion but impairs independent writing competence in the long run, with stronger influences on low-proficiency students. This stratified dual-path framework offers empirical references for standardized AI application in university writing teaching.