We leverage the implicit cognitive exercise inherent to wage self-reporting in standard labor force surveys to propose a method for measuring workers’ attention based on clustered (unsupervised) machine learning methods, which is capable of addressing potential reporting biases or rounded responses. The application of this methodology to French data reveals that workers perceive their own wages with a degree of uncertainty ranging from 8.2\% to 12.4\%, which, through the lens of a simple rational signal extraction model, translates into estimates of workers’ attention ranging from 29\% to 84\%. Women and longer tenured workers achieve significantly higher levels of attention. In addition, the attention of the lowest paid 30\% of workers exhibits a cyclical pattern: it rises continuously during the ten days preceding payday, before falling immediately, as predicted by a simple model of end-of-month financial constraints with (cognitively) costly budget constraint monitoring.