AI Fatigue in Human–AI Interaction: Conceptual Framework, Scale Development and Validation, and Associations with AI Engagement

Emerging from the accelerating pace of AI development, user fatigue arising from prolonged human–AI interaction has become an increasing concern in public discourse. This growing attention underscores the need to clarify why sustained engagement with AI systems can generate psychophysiological strain and how such exhaustion manifests in users. The present research develops a conceptual framework for AI fatigue and introduces the 15-item AI Fatigue Scale as a validated tool for examining its correlates. Across four studies with 720 participants, we generated and content-validated an item pool, identified and confirmed a four-factor (cognitive, emotional, physical, behavioural) higher-order model. The scale demonstrated strong internal consistency (α = .92), robust factor loadings, two-week test–retest reliability (ICC(2,1) = .65), and measurement invariance across sex. Convergent validity was supported through associations with general, clinical, and digital fatigue, and with AI-specific technostress, while discriminant validity was demonstrated against adjacent constructs including AI dependency, AI attachment, and critical thinking in AI use. Within the nomological network, greater AI fatigue was associated with more negative affect, more negative attitudes toward AI, higher neuroticism, and lower conscientiousness and extraversion. Greater AI fatigue predicted lower current AI use and stronger intentions to reduce use in the next three months, above and beyond general, clinical, and digital fatigue as well as AI-specific technostress. These findings provide a validated tool for examining AI fatigue and its underlying mechanisms, and establish an initial empirical foundation for how the phenomenon develops, manifests, and shapes users’ engagement with AI systems.