In a network-based system for recommending media content items based on user preferences, clients contact a server on a periodic basis, independent of the user. In addition to client-server interaction, clients also interact with one another in peer-to-peer fashion. Peers query one another and evaluate their similarity to each other in an interactive comparison of user preferences. When two clients are sufficiently similar, the interaction culminates in the originating client downloading content listings from the targeted peer to generate suggestions for their user. If the two clients are dissimilar, the query may be terminated, or the targeted peer may route the query to a second targeted peer. In addition to the lists of preferences, the originating client may download actual content items from the targeted peers.