Converging evidence has shown that in situations of joint action, co-actors form predictive representations of one another’s upcoming actions and hold these in the same format as their representations of their own future actions. Recently, it has been suggested that a similar process occurs with utterances in cases of joint language production (i.e. conversation). The current study investigates this proposition using a joint consecutive picture-naming task to ascertain whether people form representations of their partners’ utterances. Further, as findings suggest that joint action co-representations can be modulated by social factors, the present study examines whether the same is true with utterance co-representations by testing pairs of close friends against pairs of strangers. The two primary hypotheses were: (a) subjects would make the least amount of errors and produce the fastest responses when their partner was silent, and (b) performance among pairs of friends would be better as compared with pairs of strangers. Marginal results were found in support of the first hypothesis, such that speakers produced almost significantly faster responses when their partner was silent than when their partner was also directed to produce a response. With regards to the second hypothesis, there was a nonsignificant trend towards friends making more errors than strangers. Findings lend support to the theory that interlocutors produce predictive utterance co-representations for one another. These co-representations might also be affected by social variables. However, further replication and elaboration is required before any concrete conclusions can be drawn.