49

The proliferation and calculability of Internet users’ speech on blogs, forums, and social networking sites seems to provide access to spontaneous opinion, taken directly from the traces of our everyday conversations. Since the 2000s, a range of start-ups and agencies have been developing human and algorithmic methods to draw on this wealth of material in order to provide a new way of measuring public opinion, purportedly more authentic than that measured by traditional surveys. By providing a sociological history of the online opinion survey market, this article studies the way in which a new regime of knowledge about opinion was constructed based on digital traces, and underlines the varied, contingent and situated nature of the epistemic projects that make use of big data. Based on interviews and an ethnographic study, I show the contrast between the companies that chose sampling of digital traces as an approach and those that, instead, pursued exhaustiveness in the capture of opinions voiced on the social web. In particular, I analyze the at once technical and epistemic challenges faced by online opinion actors that caused sampling approaches to fail, establishing instead the extensive and continuous monitoring of online conversation.