To test in how far the Media Equation and Computers Are Social Actors (CASA) validly explain user responses to social robots, we manipulated how a bad health message was framed and the language that was used. In the wake of Experiment 2 of Burgers et al. (Patient Educ Couns 89(2):267–273, 2012. https://doi.org/10.1016/j.pec.2012.08.008), a human versus robot doctor delivered health messages framed positively or negatively, using affirmations or negations. In using frequentist (robots are different from humans) and Bayesian (robots are the same) analyses, we found that participants liked the robot doctor and the robot’s message better than the human’s. The robot also compelled more compliance to the medical treatment. For the level of expected quality of life, the human and robot doctor tied. The robot was not seen as affectively distant but rather involving, ethical, skilled, and people wanted to consult her again. Note that doctor robot was not a seriously looking physician but a little girl with the voice of a young woman. We conclude that both Media Equation and CASA need to be altered when it comes to robot communication. We argue that if certain negative qualities are filtered out (e.g., strong emotion expression), credibility will increase, which lowers affective distance to the messenger. Robots sometimes outperform humans on emotional tasks, which may relieve physicians from a most demanding duty of disclosing unfavorable information to a patient.