Smiling has been conceptualized as a signal of cooperative intent, yet smiles are easy to fake.We suggest that contextually appropriate, dynamically engaged smiling imposes an attentional cost, thereby making engaged smiling a plausible "honest signal" of cooperative intent. To test this hypothesis, we analyzed data from 123 pairs of same-sex strangers having "getting-to-knowyou" conversations who subsequently played a one-shot prisoner's dilemma together. We calculated the strength of engagement in smiling using a cross-lagged auto-regressive model for dyadic data. We found that when an individual's partner (the signaler) tended to smile in a more responsive way, that individual (the receiver) was more likely to cooperate. Conversely, when a signaler tended to smile in a less responsive way, the receiver was less likely to cooperate. These effects were present over-and-above the effects of average levels of smiling and self-reported liking, which also predicted likelihood of cooperation. However, dynamically engaged smiling did not predict cooperation on the part of the signaler, suggesting that receivers weight the importance of engagement more highly than they should, or even that engaged smiling might be a manipulative display. These results illustrate how conversational dynamics can influence evolutionary signaling.