Prior research has shown that time spent in school does not close the achievement gap between students with low and high socioeconomic status (SES). We examine the effect of combining increased instruction time with information to teachers about their students' reading achievements by using a randomized controlled trial. We find that the teachers' baseline beliefs are more important for low-SES students' academic performance, that the intervention makes the teachers update these beliefs, and—not least—that the intervention improves the reading skills of low-SES students and thereby reduces the achievement gap between high- and low-SES students. The results are consistent with a model in which the teachers' beliefs about the students' reading skills are more important to low- than high-SES students, while at the same time, the teachers' beliefs are subject to information friction and Bayesian learning.