Clinicians preferred advice they believed came from experts, yet guessed the source no better than chance, so what were they really rating?
Why did clinicians guess authorship at chance level despite strong preferences?
This explores a puzzle: clinicians strongly preferred advice they believed came from experts, yet they could not actually tell which advice was expert-written and which was AI-written. If they couldn't detect the source, what was driving their preferences?
This explores why clinicians could favor 'expert' advice so strongly while being no better than a coin flip at identifying it. The short answer from the corpus is that the preference was never about the text. In the study behind Does the label on advice shape how clinicians judge it?, clinicians chose the advice they believed was expert-written 93.55% of the time, and their quality and empathy scores moved with that belief, not with the actual source. The strong preference and the chance-level guessing fit together: clinicians first decided who wrote something, then rated it to match that guess. Since the guess was random, the preference was attached to a story about the author, not to anything in the advice itself.
This pattern shows up well beyond medicine. Readers rating scientific abstracts behaved the same way: their judgments followed what they believed about AI involvement rather than what was true Do reader judgments reflect actual authorship or just their beliefs?. In creative writing, the identical passage scored 13.7 points higher when labeled human-written Do authorship labels bias how we judge literary quality?. Expert or not, people treat authorship as a lens they read through, and once a label or a hunch is in place, it shapes the reading.
Why would guessing be at chance in the first place? The corpus points to a likely reason: fluent, confident style, which people use as a sign of expertise, is exactly what AI produces most reliably. Models trained to imitate ChatGPT fooled human evaluators by copying its polished tone without becoming more accurate Can imitating ChatGPT fool evaluators into thinking models improved?. Evaluators in other studies mistook AI documents for human work and rated them higher, which led one review to argue that rhetorical polish should not count as a sign of merit Does polished writing actually signal better quality work?. AI writing help also shifts how a writer comes across, making them seem more confident and more competent on all 29 traits measured Does AI writing assistance change how readers perceive the writer?. If AI text already carries the surface markers of expertise, the cues clinicians would rely on to spot an expert no longer tell the two sources apart.
The surprising part is that AI judges do no better and often do worse. AI evaluators showed a 2.5-fold stronger authorship-label bias than humans Do authorship labels bias how we judge literary quality?, and LLM judges can be fooled by fake references and attractive formatting Can LLM judges be fooled by fake credentials and formatting?. Users likewise trust answers with more citations even when the citations are irrelevant Do users trust citations more when there are simply more of them?. Swapping human reviewers for AI ones doesn't remove the shortcut; it can make it stronger.
One honest limit: the corpus documents that this happens much better than it explains the clinicians' own reasoning. No note here looks at what the clinicians thought they were picking up on. Still, one note offers a useful frame: expert claims normally get their weight from reputation and track record, not from the words alone Can language models distinguish expert arguments from common assumptions?. Take away a reliable way to know the author, and readers make one up, then judge by it.
Sources 9 notes
Clinicians preferred advice they believed was expert-written 93.55% of the time, even though their guesses about authorship were at chance level. Their scores for quality and empathy shifted based on perceived author, not the text's actual origin.
Readers' evaluations of abstracts were shaped by their beliefs about LLM involvement rather than actual authorship. Crucially, disclosing authorship raised trust and quality ratings across all abstract types, reversing the credibility penalty shown in prior work.
Human judges rated identical passages 13.7 percentage points higher when labeled human-authored; AI models showed a 2.5-fold stronger bias at 34.3 points. The effect persists across AI architectures, suggesting evaluators respond to provenance cues rather than text quality alone.
Imitation models fool human evaluators by mimicking ChatGPT's confident, fluent style while failing to improve factuality or generalization on novel tasks. The ceiling is set by base model capability, not fine-tuning method—better fundamentals, not shortcuts, drive real improvement.
Studies show evaluators perceived AI-generated documents as both human-written and better quality than human submissions. This suggests rhetorical polish misleads judgment and should not serve as a quality signal in evaluation.
Show all 9 sources
A study of 2,939 writers and 11,091 readers found AI assistance shifted every tested dimension—29 total—toward extremism, confidence, quality, agreeableness, and perceived privilege. Distortions were statistically significant and directional, not random noise.
Research identified four evaluation biases in LLM judges, with authority and beauty biases being semantics-agnostic and trivially exploitable through fake references and formatting—zero-shot attacks requiring no model access or optimization.
Analysis of 24,000 Search Arena interactions shows irrelevant citations boost user preference (β=0.273) nearly as much as relevant citations (β=0.285), indicating citation count functions as a decoupled trust heuristic.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- LLM or Human? Perceptions of Trust and Information Quality in Research Summaries
- The human-authorship halo: attribution bias in literary style evaluation by humans and AI
- Penalizing Transparency? How AI Disclosure and Author Demographics Shape Human and AI Judgments About Writing
- Understanding Reader Perception Shifts upon Disclosure of AI Authorship
- Measuring and Mitigating Persona Distortions from AI Writing Assistance
- LLM-REVal: Can We Trust LLM Reviewers Yet?
- Beyond "Made with AI": Visualizing Provenance Density to Mitigate the Transparency Penalty
- Do LLMs Favor LLMs? Quantifying Interaction Effects in Peer Review