From Incoherence to Verification: Epistemic Friction and Confidence in Evaluating AI Health Information

The increasing use of artificial intelligence (AI) for health information seeking raises concerns about exposure to inaccurate or misleading content. Although prior research has documented the prevalence and consequences of health misinformation, less is known about how individuals detect and respond to problematic AI-generated information. Drawing on Epistemic Vigilance Theory (EVT), this study examines whether exposure to incoherent health information influences verification-oriented information seeking through epistemic friction, defined as perceived inconsistency between encountered information and prior knowledge. Data were collected from U.S. adults with diabetes (N = 200) using an experimental design in which participants were exposed to AI-generated information about a fictitious antidiabetic drug that either aligned with or violated established medical knowledge. Exposure to incoherent content increased epistemic friction, which in turn was associated with greater verification-oriented information seeking intention. Notably, this indirect relationship was moderated by perceived evaluation confidence, such that the effect was stronger among individuals with lower confidence in their ability to evaluate health information. These findings suggest that epistemic friction may play an important role in responses to health information presented as AI-generated and that subsequent verification intentions may vary according to individuals’ perceived evaluative confidence. The study extends EVT to AI-mediated environments and highlights the importance of promoting effective responses to health misinformation.