When choosing business partners, organizations ideally prefer partners with the most relevant expertise while avoiding those who also serve their competitors. Hence, in market networks, partner choice often presents a trade-off between accessing expertise and avoiding second-order competitive overlap. We propose that as competitive overlap increases, organizations’ fears of information leakage and concerns about access to resources lead them to select less expert partners. A matched sample analysis of 963,089 US patents with measures of expertise and competitive overlap constructed via a text-based, deep learning algorithm shows that the likelihood a client selects a patent law firm based on relevant expertise decreases significantly as competitive overlap with other clients of the law firm increases. However, when concerns about information leakage or access to resources are lower, in particular when the client and the law firm have a prior relationship, when the law firm is high status, and when few alternatives are available, this effect weakens. Finally, we show that when a client chooses a less expert partner, time to patent acceptance is greater and forward citations are lower, indicating that avoidance of competitive overlap may come at a significant cost.