We offer a novel test of whether non-binding goals set ahead of a task are effective motivators, taking into account that individuals in principle could easily revise these goals. In our setting, subjects either set a goal some days prior to an online task (early goal) or right at the start of the task (late goal). Two further treatments allow for (unanticipated) explicit revision of the early goal. Consistent with early goals being self-regulation tools, we observe that (i) early goals are larger than late goals; (ii) subjects in early goals treatments work more than those in the late goal treatment, despite evidence of substantial downward goal revision in the former. A secondary contribution of our paper is that our design addresses a treatment migration problem present in earlier studies on goals that stems from the fact that subjects in a `no goals' control condition may privately set goals.