We compare the scientific productivity of PhD students who are hired from a fine-grained set of mutually exclusive affiliation types: a PhD supervisor's affiliation, an external affiliation from which the supervisor derives her coauthors, and an external affiliation with which the supervisor has no coauthorship ties. Using a novel dataset of science and engineering PhD students who graduated from two major Swiss universities, we find that the most productive PhD category is the one made of students who are affiliated with universities other than their supervisors’ affiliation, but from which the PhD supervisors derive their coauthors. This result suggests an inverted U-shaped relationship between PhD students’ productivity and the social distance from their supervisors. Additionally, we find evidence consistent with the role of supervisors’ coauthor networks in resolving information asymmetries regarding PhD talent.