We study how people think others update their beliefs upon encountering new evidence. We find support for Martingale property: when two individuals share the same prior, one believes that new evidence cannot systematically shift the other's beliefs in either direction. We also find that when the two have different priors, people think that any information brings others’ expected posteriors closer to their own prior, but this adjustment is less responsive to information quality than theory predicts. We identify the primary cause of this insensitivity and discuss the implications of our findings for strategic games with asymmetric information, information design, and, more broadly, for understanding societal polarization.