When people engage in rhythmic joint actions, they unintentionally increase their tempo. However, this phenomenon of joint rushing has so far been investigated only under very specific and somewhat artificial conditions. Therefore, it remains unclear whether joint rushing generalizes to other instances of rhythmic joint action. In this study our aim was to investigate whether joint rushing can also be observed in a wider range of naturalistic rhythmic social interactions. To achieve this, we retrieved videos of a wide range of rhythmic interactions from an online video-sharing platform. The data suggest that joint rushing indeed can also be observed in more naturalistic social interactions. Furthermore, we provide evidence that group size matters for how tempo unfolds in social interactions with larger groups showing a stronger tempo increase than smaller groups. Comparing the data from naturalistic interactions with data collected in a lab study further showed that unintended tempo changes in social interactions are reduced in naturalistic interactions compared to interactions in a lab context. It is an open question which factors led to this reduction. One possibility is that humans might have come up with strategies to reduce the effects of joint rushing.