There is an ongoing debate, both in philosophy and psychology, as to whether people are able to automatically infer what others may know, or whether they can only derive belief inferences by deploying cognitive resources. Evidence from laboratory tasks, often involving false beliefs or visual-perspective taking, has suggested that belief inferences are cognitively costly, controlled processes. Here we suggest that in everyday conversation, belief reasoning is pervasive and therefore potentially automatic in some cases. To test this hypothesis, we conducted two pre-registered self-paced reading experiments (N 1 = 91, N 2 = 89). The results of these experiments showed that participants slowed down when a stranger commented 'That greasy food is bad for your ulcer' relative to conditions where a stranger commented on their own ulcer or a friend made either comment-none of which violated participants' commonground expectations. We conclude that Theory of Mind models need to account for belief reasoning in conversation as it is at the center of everyday social interaction. for theoretical models of adult social cognition that fail to account for belief reasoning in conversation (e.g., Apperly & Butterfill, 2009; Heyes, 2014) and as a result leave out of their scope the bulk of the data observed in everyday social interaction. Here we argue that belief reasoning is pervasive in communication and therefore needs to be investigated and characterized in conversational settings, and not only in false-belief tasks. Does communication involve belief reasoning? A popular theoretical view, both in philosophy and psychology, is the hypothesis that attributing mental states to others (not only beliefs, but also intentions and desires) is too cognitively demanding to be the basis for real-time social interaction (e.g.