Social media platforms have been playing an increasingly crucial role in shaping social interaction and information exchange. While facilitating communication, they possess the central power that determines which content users are exposed to. In this paper, we examine how platforms’ information distribution mechanisms affect a society’s ability to aggregate information correctly and efficiently. We consider four criteria that algorithms can favor in curating content: globally most engaging (GE), locally most engaging (LE), globally closest (GC), and/or locally closest (LC). That is, for a given user, platforms can promote beliefs that are most engaging and/or are most similar to her pre-existing belief; beliefs can be sorted globally (among all users) or locally (among the user’s connections). We simulate a social learning model where agents form beliefs by learning from each other and from the information allocated by the platform. We find that while exposing users to engaging beliefs considerably increases learning bias, feeding personalized information based on belief similarity does not affect collective wisdom significantly. All mechanisms except GE slow down the speed for agents to converge to a consensus.