In a study investigating different ways to engage students in an individual preparation activity, we assessed learning from collaboration with a peer. Learning was measured at pretest, individual preparation, collaboration, and posttest. Two conditions were compared: generative preparation and nongenerative preparation. With no differences in knowledge at pretest, we found that students who prepared by engaging in the generative compared to the nongenerative preparation task performed worse to a statistically significant degree in both the preparation and collaboration phases. However, in a postactivity transfer task, the students performed to a similar degree as the comparison group evidencing "productive failure." In addition, students in the generative condition showed greater learning gains from the preparation to the collaboration phase. We introduce the notion that generative preparation may better invoke learning mechanisms that help students to learn from subsequent collaboration.