Nearly all online marketplaces face the challenge of ensuring the quality of sellers; this is particularly challenging in large marketplaces, where it may be infeasible to pre-screen all sellers. A growing trend in such settings is to use feedback from past transactions to infer the quality of new sellers. However, this process requires that existing buyers-the most valuable users of the platform-be subjected to matching to new, untrusted sellers, potentially leading to a bad experience. We investigate this tradeoff in a setting where existing buyers help uncover seller quality. We develop a stylized model, which captures the salient features inherent in such markets, while retaining analytical tractability. Our model uncovers a qualitative difference between exploration and exploitation, based on underlying network externalities between buyers and sellers in the market. In particular, in many settings, pure exploration achieves the optimal rate of successful matches.