Generative AI-Mediated Yellow River Narratives in Sustainability Education: A Repeated Q-Method Comparison at Two Post-Exposure Occasions

Education for sustainable development (ESD) requires pedagogies that connect global sustainability goals with disciplinary, cultural, and professional contexts. Using repeated Q methodology, this study compared how 31 Chinese STEM undergraduates perceived a two-semester course activity integrating instruction and oral practice with the generative artificial intelligence (GenAI) conversational agent Doubao (app version 7.2.0). Participants completed identical 60-statement Q-sorts at two post-exposure occasions—after one semester (T1) and after two semesters (T2). Principal component extraction followed by varimax rotation produced three-factor solutions explaining 50% of variance in the rotated Q-sort solution at T1 and 45% at T2. A rule-based loading audit identified 28 defining Q-sorts at T1 and all 31 at T2. T1 viewpoints were Autonomous Cultural Translators, GenAI-Scaffolded Linguistic Pragmatists, and Career-Oriented Global Communicators; T2 viewpoints were Independent GenAI-Cautious Rehearsers, Critical Cultural Negotiators, and Iterative GenAI-Supported Revisers. Cross-time factor-array correlations were low to moderate (r = −0.053 to 0.452), and participant assignments were dispersed across the solutions. The pattern describes occasion-specific viewpoint configurations and selective GenAI appraisal rather than measured competency growth or a causal developmental trajectory. Without a pre-exposure baseline, a comparison group, or direct learning-outcome measures, the study cannot establish educational effectiveness or GenAI’s independent effect.