Effect of Disclaimers on the Public Perception of AI-Generated Medical Advice: Two Preregistered Survey Experiments

Large language models are increasingly used by the public as a source of medical advice, yet they can produce plausible but incorrect output. Disclaimers about the limitations of artificial intelligence (AI) are the most common safeguard and are also anchored in regulatory frameworks such as the European Union's AI Act. Whether disclaimers change how laypeople perceive and act on AI-generated medical advice, however, is unknown. We conducted two preregistered, randomized between-subjects survey experiments with US participants recruited via Prolific (study 1: N = 399; study 2: N = 396). Participants read four simulated dialogues in which a patient received medical advice from an AI tool and were randomized to see a disclaimer below the advice or no disclaimer. Study 1 used the standard disclaimer of a popular chatbot; study 2 used a stronger disclaimer that explicitly highlighted the higher risk of error compared with physicians and required participants to confirm having read it. Participants rated the advice’s empathy, reliability, and comprehensibility and their willingness to follow it. In study 1, the disclaimer did not significantly affect any outcome (all p > .17). In study 2, it did not significantly affect empathy, reliability, or comprehensibility (all p > .19). Willingness to follow the advice was slightly lower with the disclaimer (p = .039), but this effect did not remain significant after correcting for multiple testing. Furthermore, there was no significant difference between the two disclaimers. Disclaimers may thus serve mainly as transparency and liability measures and should not be relied on alone to protect users. Alternatives, such as communicating uncertainty within the response and strengthening AI literacy, deserve evaluation.