Proceedings of the Annual Meeting of the Cognitive Science Society

AI hallucinations can reshape cognition when false content becomes useful for decisions. We test how AI-generated false claims become usable in reasoning and difficult to separate from their source. Across two preregistered studies (Study 1: N = 295; Study 2: N = 242), participants made a medication choice after reviewing standardized information and receiving an AI reply varying in hallucination severity. Study 1 shows that stronger hallucinations shifted understanding of a risk relation, changed onset-duration weighting, and increased source confusion, especially for a fabricated causal mechanism. Study 2 traced this pathway in speech: stronger hallucinations increased uptake during decision verbalization, reduced critical evaluation and reliance on original evidence, and weakened source reasoning during interaction recall. Later transmission and post hoc reflection were comparatively bounded. These findings locate hallucination risk at the transition from exposure to reason construction, where AI-originated content becomes useful in participants' explanations while provenance becomes difficult to track.