AI Hallucinations in Human–AI Collaboration: A Grounded Theory of Employee Use in Organizational Work

Generative AI outputs often contain hallucinated content, yet little is known about why employees adopt such content at work. Drawing on semi-structured interviews with 33 employees across multiple industries and procedural grounded theory, this study develops an explanatory framework of the factors and mechanisms that shape employees’ adoption of hallucinated AI outputs. Task demands, individual characteristics, technological cues, and organizational arrangements operate through two mechanisms: cognitive credibility judgments and trust-laden affective reliance; organizational arrangements also shape whether employees resolve their doubts. Adoption takes three forms that an explicit decision rule separates: unconscious adoption, cognitive adoption, and motivation-driven adoption. The study shows that hallucinated content enters work products knowingly as well as unknowingly. It moves beyond treating hallucinations solely as technical failures or discrete information-adoption decisions and explains how plausible but flawed AI outputs enter organizational work products and decisions. The findings also offer implications for AI system design, employee training, and organizational risk management.