This study draws on Social Cognitive Theory to examine the association between learners’ perceptions of generative AI (GenAI)-based pronunciation feedback and their willingness to communicate (WTC) in English, with English pronunciation self-efficacy considered as a potential mediator. Using a cross-sectional survey design, data were collected from a convenience sample of 1,701 Chinese university EFL learners. Covariance-based structural equation modeling and bias-corrected bootstrapping were employed to test the hypothesized relationships and mediating effect. The results showed that positive perceptions of GenAI-based pronunciation feedback were significantly associated with greater WTC in English. English pronunciation self-efficacy partially mediated this association, with the indirect effect accounting for 69.9% of the total effect, while the direct effect remained significant. These findings suggest that pronunciation self-efficacy is an important, but not exclusive, psychological mechanism connecting perceptions of GenAI-based feedback with WTC. GenAI-based feedback may also relate to WTC through other mechanisms, such as reduced communication anxiety and enhanced psychological safety. By conceptualizing GenAI-based feedback as a technology-mediated environmental influence and pronunciation self-efficacy as a personal cognitive factor, this study extends the application of Social Cognitive Theory to the Chinese EFL context and offers new insight into how intelligent feedback technologies may support learners’ readiness to communicate in English.