Signals related to self generated movements are broadcast to non-motor brain regions, and can be detected across wide regions of the forebrain in mice. Indeed, it has been observed that, even during the performance of a task, incidental movements, or ‘fidgets’, account for much more variance in neural activity than task-related movements and task-related variables. Some bodily movements or expressions, such as those of the face, reflect biologically significant expressions of internal state, that is, emotions, which have important relationships with ongoing cognitive processes such as decision-making. Therefore, it is possible that movements characterized as incidental actually express latent internal states that are in fact related to decision processes. Here, we trained mice to perform a probabilistic foraging task while video monitoring facial movements and simultaneously recording large ensembles of neurons in frontal and premotor cortical regions using Neuropixels probes. In this task, mice had to combine a sequence of successful and failed foraging attempts to compute a latent decision variable (DV) in order to time the duration of their foraging bout. After training, this DV predicted the mice’s behavior and accounted for a large degree of premotor activity, but consistent with a previous report, high dimensional facial movements dominated neural activity in premotor cortex and in all the other recorded brain regions. Remarkably, however, further analysis revealed that the explanatory power of facial movement was largely due to its correlation with the latent DV. The component of the movement uncorrelated with the latent DV had little predictive power. The premotor representation of the DV temporally preceded that of the movements, suggesting that premotor activity contributed to generating facial movements rather than reflecting proprioceptive feedback. These results show that seemingly ‘incidental’ bodily expressions can in fact reveal otherwise hidden task-relevant states.