Recommendation agents are artificial intelligence systems designed to predict consumers’ preferences on the basis of traces of their Web surfing. According to the literature, these agents’ potential efficiency is considerable. Yet few studies until now have investigated the forms of such capabilities. This article makes a twofold contribution. First, it reports the behaviouralist methods used by management science to assess and regulate recommendation agents’ relevance. Second, it shows how these methods can be used by the social sciences, not to elicit rules governing economic action, as the management sciences traditionally do, but rather to understand the modes of existence of markets. In this respect, the authors show that the recommendation agent is a device for the economization of the exploration of consumers who, in the digital market era, participate in the establishment of new forms of cognitive and relational framing.