This study examined how an instructional program integrating hands-on dye-sensitized solar cell (DSSC) experimentation with critically guided generative artificial intelligence (GenAI) inquiry influences university students' STEM–Education for Sustainable Development (STEM–ESD) literacy. An explanatory sequential mixed-methods design with an embedded quasi-experimental component was employed. Participants were 64 undergraduate physics education students in Bandung, Indonesia, allocated by intact class sections into an intervention group (n = 32) and a comparison group (n = 32). Across six weekly sessions the intervention group framed an energy problem, fabricated and characterized DSSCs sensitized with natural dyes, interrogated GenAI-generated explanations against their own measurements, and redesigned their cells using sustainability criteria. Data were collected through a validated five-dimension STEM–ESD literacy test administered before and after the program, laboratory notebooks, reflective journals, GenAI interaction logs, group discussions, and semi-structured interviews. Quantitative data were analyzed using baseline-adjusted regression models and effect-size estimation, qualitative data through thematic and epistemic coding with an epistemic-network representation, and the two strands were integrated using joint displays. Illustrative findings indicate the largest gains in sustainability literacy and evidence-evaluation practices, explained by an iterative cycle in which AI-generated explanations were treated as claims requiring verification rather than as authoritative answers.