Trusting the machine as a partner: Human-AI collaboration and hallucination risk

Purpose This study examines the psychological and relational mechanisms underlying employees' trust in and collaboration with generative AI. Integrating relational demography theory and the transactive memory system framework, it develops a socio-cognitive model explaining how relational alignment and cognitive coordination shape trust and collaborative outcomes. Design/methodology/approach Data were collected from 517 professionals using LLM-based AI tools via online survey. PLS-SEM was employed to test the proposed model and hypotheses. Findings Personality similarity and cognitive coordination significantly increased trust in AI, which sequentially enhanced AI efficacy, knowledge sharing, and preference for AI collaboration. Anthropomorphism and workstyle similarity were non-significant. Hallucination risk negatively moderates the relationship between AI efficacy and knowledge sharing, underscoring the role of epistemic reliability. Originality/value This study advances human-AI collaboration research by integrating emotional, cognitive, and relational pathways of trust formation and identifying hallucination risk as a critical boundary condition in generative AI contexts.