Is inner speech involved in sustaining attention, and is this reflected in response times for stimulus detection? In Experiment 1, we measured response times to an infrequently occurring stimulus (a black dot appearing at 1-3 min intervals) and subsequently asked participants to report on the character of their inner experience at the time the stimulus appeared. Our main preregistered hypothesis was that there would be an interaction between inner speech and task relevance of thought with reaction times being the fastest on prompts preceded by task-relevant inner speech. This would indicate that participants could use their inner voice to maintain performance on the task. With generalized linear mixed-effects models fitted to a gamma distribution, we found significant effects of task relevance but no interaction with inner speech. However, using a hierarchical Bayesian analysis method, we found that trials preceded by task-relevant inner speech additionally displayed lower standard deviation and lower mode (independently of the main effect of task relevance), suggestive of increased processing efficiency. Due to deviations from the preregistered sampling and analysis procedures, we replicated our findings in Experiment 2. Our results add support to the hypothesis that inner speech serves a functional role in top-down attentional control.