Why do people trust the same medical advice more or less just because of the label stuck on it?

This explores why identical medical advice earns more or less trust depending on whether it carries an AI label or a doctor's name, and whether that label changes anything beyond opinions. The short answer from the corpus is that the label strongly shapes how people *rate* advice, but it barely affects how they *use* it. That makes the label a weaker safeguard than it looks.

Start with what happens when there is no label. Neither patients nor clinicians can reliably tell AI-written medical answers from doctors' answers. In one study, 300 participants guessed at chance level and rated even low-accuracy AI answers as trustworthy enough to act on, sometimes more than real doctors' advice Can people tell AI medical advice from doctors' responses?. Clinicians did no better: blinded experts identified GPT-4 versus expert advice correctly only 45% of the time, and they rated GPT-4 as more emotionally empathetic Can clinicians tell GPT-4 advice apart from expert advice?. Outside medicine the pattern is the same. Unlabeled AI-assisted messages get the same default trust as human ones until someone discloses the AI's role Do readers trust unlabeled AI-written messages as much as human ones?.

So the trust gap comes from the label itself, not from anything in the text. Clinicians preferred advice they *believed* was expert-written 93.55% of the time, even though their beliefs about authorship were no better than coin flips. Their quality and empathy scores followed the perceived author rather than the actual one Does the label on advice shape how clinicians judge it?. The surprise comes when you check behavior. Radiologists rated AI-labeled advice lower, yet their diagnostic accuracy depended only on whether the advice was correct, not on who it was attributed to Does labeling advice as AI change how clinicians use it?. The label changes what people say about the advice. It doesn't seem to change how much the advice pulls their decisions.

That matters because the pull is lopsided. When advice labeled as AI output was wrong, experienced radiologists dropped from 82% to 45.5% accuracy, and inexperienced ones fell below 20% How much does wrong AI advice harm radiologist accuracy?. In an ICU simulation, misleading AI predictions hurt nurse performance nearly twice as much as correct ones helped Do wrong AI predictions hurt more than right ones help?. The same asymmetry shows up in AI fact-checking, where mislabeling true headlines lowered belief in them Does AI fact-checking actually help people spot misinformation?. Skepticism toward the label, then, does not protect people from bad content. Outside medicine the label's penalty can also be small: AI disclosure on a news article cost under 0.15 points on a 7-point scale Does disclosing AI assistance make readers trust articles less?.

The takeaway you may not have expected is that arguing over labels may be arguing over the wrong thing. If the label mostly shapes reputation and the content drives decisions, the more useful designs change how advice is delivered. One example is having AI point out which parts of a case deserve attention rather than handing over a verdict, which reduces anchoring on the machine's answer Can AI guidance reduce anchoring bias better than AI decisions?. One caveat: the evidence on default trust comes from snapshot studies. Whether people grow more suspicious of unlabeled advice as awareness of AI spreads remains an open question Does trust in unlabeled AI messages decline as awareness grows?.