Generative AI-Assisted Academic Writing Experience and University Students' Academic Self-Efficacy: Parallel Indirect Associations Through Academic Emotions.

Generative artificial intelligence (GenAI) is increasingly used in academic writing, yet its associations with students' broader academic emotions and capability beliefs remain unclear. Drawing on control-value theory and social cognitive theory, this cross-sectional study examined positive and negative academic emotions as parallel statistical pathways between GenAI-assisted writing experience and general academic self-efficacy. Participants were 1128 students from universities in eastern China. A covariate-adjusted parallel pathway model was estimated with 5000 bootstrap resamples. GenAI-assisted writing experience was positively associated with self-efficacy (total association B = 0.733, β = 0.670) and broader positive emotions (B = 0.778, β = 0.631), and negatively associated with broader negative emotions (B = -0.457, β = -0.298). The positive-emotion indirect association was 0.471 (standardized β = 0.431), 95% CI [0.409, 0.531], whereas the negative-emotion indirect association was 0.033 (standardized β = 0.030), 95% CI [0.018, 0.052]. The direct association remained positive (B = 0.229, β = 0.209). After positive academic emotions were controlled, the coefficient linking GenAI-writing experience with negative emotions changed from negative to positive, indicating a statistical suppression pattern. A sensitivity model excluding emotional-motivation items reproduced both indirect associations. Given the limited discriminant validity between positive emotions and self-efficacy and the concurrent self-report design, the larger positive indirect association warrants cautious interpretation. The findings indicate that GenAI-assisted writing experience is associated with broader academic emotions and efficacy beliefs while retaining a smaller residual negative-emotion component.