Generative AI Use and the Student Well-Being Paradox in Higher Education: Efficiency-Gain Retention, Performance Escalation, and Relative Competition

Generative artificial intelligence (GenAI) can reduce the time and cognitive effort required for university learning, yet technical efficiency does not necessarily translate into improved student well-being. This article develops and tests a conversion account of this paradox by distinguishing efficiency gains that remain available as student-controlled resources from gains that are associated with perceived increases in performance expectations, competitive pressure, verification work, and dependence. A sequential three-study mixed-methods program was conducted with undergraduates in Jiangxi Province. Study 1 used codebook thematic analysis of 42 interviews to identify six mechanisms linking immediate efficiency to resource retention or perceived pressure. Study 2 surveyed 1268 students across eight universities. GenAI use was strongly associated with perceived efficiency gains, but also with perceived performance-standard escalation and verification burden. Positive retention and negative competition pathways were similar in magnitude, yielding a nonsignificant total association with well-being; working-time retention strengthened the efficiency–well-being association. Because Study 2 was cross-sectional, these pathways are interpreted as indirect associations rather than causal mediation. Study 3 used a 2 × 2 experiment (N = 320) and found that AI access improved task quality and reduced completion time under both performance-norm conditions, whereas gains in autonomy, retained time, and state well-being were evident under the stable norm but not statistically distinguishable from zero under the escalated norm. The findings distinguish technical affordance from context-dependent conversion and suggest that responsible AI integration should consider whether efficiency gains remain available for learning, recovery, and student-directed use alongside productivity.