The social anatomy of AI anxiety: gender, generations, and technological exposure

Introduction Public anxiety surrounding artificial intelligence (AI) carries significant clinical, educational, and policy implications. However, evidence regarding the multidimensional structure of AI-related anxiety and its demographic and experiential correlates remains fragmented. This study synthesizes validated measures into a coherent framework to examine how psychological and sociodemographic factors shape AI-related anxieties. Method A cross-sectional survey of adults (N = 1,151) assessed nine dimensions of AI-related anxiety --general AI anxiety, technoparanoia, technophobia, AI interaction anxiety, job-replacement anxiety, sociotechnical blindness, cybernetic-revolt fear, technology self-efficacy, and AI learning orientation --adapted from established scales. Dimensionality was evaluated using common-factor exploratory factor analysis (principal axis factoring, Promax rotation; KMO = .89; Bartlett's p < .001), supported by parallel analysis and scree inspection. A 70/30 hold-out confirmatory factor analysis assessed structural validity. Reliability (Cronbach's α, McDonald's ω), composite reliability (CR), and average variance extracted (AVE) were calculated to examine internal consistency and convergent validity, while discriminant validity used the Fornell -Larcker and HTMT criteria. Group differences were tested using t-tests and ANOVA with Holm -Bonferroni correction and effect sizes. Hierarchical regression models controlled for age, gender, marital status, employment, and AI-use status. Results The nine-factor structure was supported (64.17% variance explained). CFA indicated good fit (CFI = .943, TLI = .936, RMSEA = .045 [90% CI .041 -.049], SRMR = .046). All scales demonstrated strong reliability (α, ω ≥ .80), convergent validity (CR ≥ .83; AVE ≥ .51), and discriminant validity. After correction for multiple comparisons, gender differences remained for technoparanoia, AI learning orientation, and AI interaction anxiety (small effects, Cohen's d ≈ .18 -.21). AI users exhibited higher general AI anxiety, technoparanoia, and sociotechnical blindness (d ≈ .17 -.29). Age-group differences were non-significant. Hierarchical regression showed that sociotechnical blindness and technoparanoia were the strongest positive predictors of general AI anxiety, while technology self-efficacy and AI learning orientation were negative predictors. Discussion AI-related anxiety is a reliable and multidimensional construct, driven more by psychological dispositions and technology experience than by demographic characteristics. The findings suggest actionable pathways for mitigating anxiety, including targeted AI literacy initiatives, strengthening self-efficacy, and transparent communication regarding sociotechnical impacts. These interventions may support informed and equitable AI integration across clinical, educational, and policy contexts.