<p>As generative artificial intelligence (AI) becomes increasingly embedded in higher education, a key educational question is not simply whether AI is used in design classrooms, but how it is instructionally integrated. This study examines how three instructional configurations in university design education—traditional instruction, a chatbot-assisted condition, and a fully integrated artificial intelligence-generated content (AIGC) condition—are associated with students’ AI-supported design product performance, self-efficacy, and role clarity / reverse-coded role ambiguity. Grounded in social cognitive theory and a role ambiguity perspective, the study conceptualizes AI use not as a binary presence-or-absence variable or a single technological feature, but as a set of instructional configurations that differ in AI availability, functional affordances, workflow integration, and pedagogical support. A quasi-experimental design was implemented in an undergraduate book design course at a comprehensive university in Hunan, China, involving 100 visual communication design students. Quantitative data were collected through pre- and post-intervention measures of design-related performance and self-efficacy, a post-intervention measure of role clarity / reverse-coded role ambiguity, and explanatory semi-structured interviews with a subset of students. The results showed that the fully integrated AIGC condition was associated with significantly higher AI-supported design product performance than the traditional and chatbot-assisted conditions, as well as the highest level of role clarity, corresponding to the lowest level of role ambiguity. However, because generative AI was available during the design process, this outcome should be interpreted as product-based task performance under AI-supported conditions rather than as direct evidence of independent learning gains. Traditional instruction and the fully integrated AIGC condition did not differ significantly in posttest self-efficacy, but both were associated with more favorable self-efficacy outcomes than the chatbot-assisted condition. The qualitative findings further suggested that these differences were related to continuity of workflow support, students’ sense of task control, and the clarity of human–AI role boundaries. The study contributes to research on AI-supported design education by comparing pedagogically meaningful instructional configurations, examining design product performance together with cognitive-affective perceptions, and providing ecologically relevant evidence from an authentic university classroom. The findings suggest that the educational value of AI in design education depends less on tool availability itself than on how AI is pedagogically embedded in design tasks, feedback processes, workflow structures, and human–AI role relationships.</p>