How do we decide whether a statement is literally true? Here, we contrast participants' eventual evaluations of a speaker's meaning with the real-time processes of comprehension. We record participants' eye movements as they respond to potentially misleading instructions to click on one of two objects which might be concealing treasure ("the treasure is behind thee, uh, hat"). Participants are less likely to click on the named object when the instructions are disfluent. However, when hearing disfluent utterances, a tendency to fixate the named object early increases with participants' autism quotient scores. This suggests that, even where utterances are equivalently understood, the processes by which interpretations are achieved vary across individuals.