Why Hide AI Use? Psychological Configurations and Explainable Machine Learning Evidence from Marketing Work
Artificial intelligence (AI) is increasingly embedded in marketing work, yet employees who use AI tools may not always disclose AI’s role in producing their outputs. This study examines AI disclosure silence, defined as employees’ intentional withholding of information about the use, role, or contribution of AI tools in work-related outputs after AI has already been used. Unlike AI avoidance or resistance, this construct concerns post-adoption concealment; unlike general employee silence, it focuses on the hidden technological contribution behind visible work. Drawing on Conservation of Resources Theory and Psychological Safety Theory, the study investigates how threat-based conditions, safety and governance conditions, and AI-related capability are associated with AI disclosure silence. Data were collected through a two-wave survey of 635 marketing employees who actively used AI tools at work. The analysis combined measurement validation, Necessary Condition Analysis (NCA), fuzzy-set Qualitative Comparative Analysis (fsQCA), and explainable machine learning. The findings show that no single condition operated as a strong necessary bottleneck. Instead, AI disclosure silence appeared through multiple pathways involving AI anxiety, fear of negative evaluation, perceived creativity threat, perceived job insecurity, low trust in management, weak psychological safety, and unclear AI policy. SHapley Additive exPlanations (SHAP)-based interpretation further indicated that fear of negative evaluation, AI anxiety, perceived creativity threat, and trust in management had the strongest model-based predictive relevance. The study contributes to workplace AI and employee silence research by positioning AI disclosure silence as an emerging post-adoption disclosure construct. It also highlights the need for clear AI disclosure norms, non-punitive managerial responses, AI-assisted authorship guidelines, and psychologically safe AI-governance practices. The findings should be interpreted as configurational and predictive evidence rather than causal effects, and further scale validation across sectors and cultures is encouraged.