We analyze the design of a mechanism to extract ordinal information dis-seminated in a social network. We show that friend-based ranking—the report by agents on the characteristics of their neighbors—is a necessary condition for ex post incentive-compatible and efficient mechanism design. We characterize the windmill network as the sparsest social network for which the planner can construct a complete ranking. When complete rankings cannot be achieved, expost incentive-compatible and efficient mechanisms arise when social networks are bipartite or composed of triangles. We illustrate these findings using real social networks in India and Indonesia.