Students receive abundant information about their educational performance, but how this information affects future educational-investment decisions is not well understood. Increasingly common sources of information are state-mandated standardized tests. On these tests, students receive a score and a label that summarizes their performance. Using a regression-discontinuity design, we find persistent effects of earning a more positive label on the college-going decisions of urban, low-income students. Consistent with a Bayesian-updating model, these effects are concentrated among students with weaker priors, specifically those who report before taking the test that they do not plan to attend a four-year college.