AI Anxiety and Student Academic Outcomes in Higher Education: A Systematic Review and Integrative Process Model

The rapid integration of generative artificial intelligence (AI) into higher education has been accompanied by a growing psychological cost: AI anxiety. How this anxiety shapes academic outcomes, and under what institutional conditions, has not been systematically synthesized. This review addresses that gap by integrating 16 peer-reviewed studies (N = 7,295 students across nine countries/regions), identified through Web of Science and Scopus (2020–2026) and reported following PRISMA 2020 guidelines. Eligible studies measured AI anxiety as a named construct and reported a genuine academic outcome, distinguishing this corpus from the broader technology-acceptance literature. Quality was appraised using the Mixed Methods Appraisal Tool, and findings synthesized thematically. Three patterns emerge from a predominantly correlational evidence base. First, AI anxiety is functionally dual rather than uniformly harmful: its effect depends on the dimension (task-focused learning anxiety tends to be maladaptive; future-focused job-replacement anxiety more often adaptive) and on whether students appraise it as challenge or hindrance. Second, self-efficacy is the most consistently evidenced correlate, buffering anxiety's harmful effects across contexts, though whether it precedes, mediates, or moderates anxiety cannot be resolved from cross-sectional data. Third, institutional design—mastery-oriented instruction, governance congruence, and accessible support—is consistently associated with lower anxiety and better outcomes, though whether it actively compensates for elevated anxiety is advanced as a hypothesis for future testing. These patterns are consolidated in an Integrative Process Model specifying propositions for mediation, moderated mediation, and institutional compensation, repositioning AI anxiety management from an individual burden to a structural, institutional responsibility.