During everyday social interactions, we typically anticipate (or explain) others' behaviour according to their current mental states (e.g. their knowledge, beliefs and intentions). To date, very little is known about the time-course with which such perspective information influences communication. We report a novel interactive ‘visual world’ study examining these processes. Here, two communicators watched videos depicting transfer events and subsequently described these events to each other. Critically, on half the trials a screen blocked the speakers' (but not the listeners') view part-way through the video, establishing a discrepancy in the knowledge held by the two communicators. Eye-tracking analyses showed that listeners were rapidly sensitive to their partner's perspective, as evidenced by a significantly reduced reality-bias when speakers held out-of-date knowledge about a privileged transfer event. However, we also found that under these conditions, listeners suffered ongoing interference from their own knowledge of reality, which inhibited successful anticipation of the speaker's intended referents. ► We examined the time-course of adult perspective-taking. ► We employed eye-tracking in a 2-person ToM task. ► We found that people were rapidly sensitive to others' perspectives. ► But, we also found evidence of ongoing interference due to a ‘pull of reality’.