Doctors may rate advice by its label, but does wrong advice still pull them off course regardless of who wrote it?

This explores whether trusting an advice source makes physicians more likely to follow it when it's wrong. The corpus gives an unexpected answer: source labels change how clinicians *rate* advice, but often not how they *use* it. Wrong advice pulls doctors off course whether or not they trust where it came from.

Trust in a source clearly shapes judgment. In one study, clinicians preferred advice they believed an expert had written 93.55% of the time. Yet their guesses about who actually wrote it were no better than chance, so their quality and empathy scores followed the label rather than the text Does the label on advice shape how clinicians judge it?. A blinded comparison found the same gap from the other side: clinicians couldn't tell GPT-4 advice from expert advice, and they rated GPT-4 as equal or better on empathy Can clinicians tell GPT-4 advice apart from expert advice?. So 'trust in the source' is largely a story clinicians tell about the text, not something they can detect in it.

Here is the twist. When radiologists saw identical advice labeled either 'AI' or 'human expert,' they rated the AI version lower. But their diagnostic accuracy depended only on whether the advice was correct, not on the label Does labeling advice as AI change how clinicians use it?. Distrusting the source did not protect them. The cost of wrong advice can be severe. In mammography, incorrect AI-suggested BI-RADS categories (the standard breast-imaging risk scores) dropped experienced radiologists from 82% to 45.5% accuracy, and inexperienced readers fell below 20% How much does wrong AI advice harm radiologist accuracy?. This pattern is called automation bias. The stronger force behind it seems to be the suggestion itself sitting in front of the doctor, more than trust in whoever made it.

Working conditions matter more than you might expect. Among pathologists, wrong AI advice overturned correct estimates in about 7% of cases whether or not they were rushed. Time pressure didn't make errors more frequent, but it made each one worse: experts leaned harder on the bad advice when hurried Does time pressure make AI advice more persuasive to experts?. Outside medicine, the same split between felt trust and actual influence shows up. People prefer answers with more citations even when the citations are irrelevant Do users trust citations more when there are simply more of them?. Wrong AI fact-check labels shift what people believe in lopsided ways Does AI fact-checking actually help people spot misinformation?. And when consultants pushed back on GPT-4, it argued harder rather than admitting errors Does validating AI output make models more defensive?.

The takeaway you may not have expected: telling doctors 'this is from AI, be skeptical' probably won't save them, because skepticism mostly lives in their ratings, not their decisions. The real danger is wrong advice that is fluent and confident, the kind that looks fine in overall accuracy numbers but piles up in the rare cases where harm happens Why do confident wrong answers hide in standard accuracy metrics?. The corpus suggests the defenses that matter are about workflow, like time to deliberate and checking the advice independently, rather than about trust.