Designing Generative AI for Learning-Oriented Use: An Efficiency-Dominant Acceptance Perspective With Education for Sustainable Development as an Evaluative Lens

Generative artificial intelligence (GenAI) is increasingly used in higher education, raising concerns about whether it supports sustainable learning or primarily promotes short-term efficiency. This study examines how GenAI design cues shape students ’ learning-oriented use through underlying psychological mechanisms. Integrating the MAIN model (Modality, Agency, Interactivity, Navigability) with technology acceptance theory (UTAUT), we adopt a sequential explanatory mixed-methods design, combining a survey of 216 undergraduate students in China with follow-up interviews. Partial least squares structural equation modeling shows that GenAI design cues influence performance expectancy, effort expectancy, and social influence, with performance expectancy emerging as the dominant pathway. This asymmetric pattern forms an Efficiency-Dominant Acceptance Structure (EDAS), in which efficiency-oriented expectations disproportionately guide GenAI-supported acceptance and learning-oriented use. Rather than treating sustainable learning as a directly measured outcome, this study uses Education for Sustainable Development (ESD) as a normative evaluative lens to interpret whether this dominant acceptance logic aligns with broader educational goals. From an Education for Sustainable Development perspective, the findings highlight a tension between efficiency-driven adoption and deeper learning-oriented engagement, underscoring the need for design and pedagogical strategies that reorient GenAI use toward deep cognitive engagement.