Medical misinformation more likely to fool AI if source appears legitimate, study shows