Censorship and Steering in Platform Duopolies

Abstract: We analyze the strategic design of recommendation algorithms in online marketplaces, motivated by stylized facts from Amazon’s BuyBox. First, we show evidence that Amazon strategically suppresses the BuyBox recommendation (“censorship”) or selects prices above the minimum available offer (“steering”) in a way that carefully tracks its largest competitor, Walmart.com. To rationalize these facts, we construct a model of platform duopoly in which two platforms compete by managing buyer search incentives through information design. We characterize the equilibrium and demonstrate a dichotomy: the platform optimally engages in either censorship or steering, but never both simultaneously. In the censorship regime, the platform strategically withholds recommendations to discipline buyer search, effectively raising the rival’s perceived cost. In the steering regime, the platform recommends prices strictly above the competitive level, bounded only by the rival’s price, which acts as a stochastic reserve. We show that these algorithmic strategies endogenously replicate the effects of Most-Favored-Nation (MFN) clauses, generating upward pricing pressure on upstream firms even in the absence of vertical restraints.