Recommender systems for video on demand (VOD) services such as Netflix have elicited highly polarized reactions.

For many commentators, these systems – which use algorithms to suggest content likely to interest viewers on the basis of their prior viewing histories –  represent a fundamentally new way of connecting cultural objects and human beings. Computer scientists, business gurus, and feature writers swoon over the ability to scale the provision of cultural recommendation using big data.

In contrast, academics and activists sustain suspicions of filter bubbles and object to how such computational processes seem bound to confirm rather than challenge or develop taste. For these passionate interlocutors, algorithmic recommendation represents the end of humanist criticism as we have known it, the death knell of the Arnoldian “best which has been thought and said.”

Curiously, however, both the vociferous champions and vehement critics share a common first-principle assumption: that VOD recommender systems are effective, powerful, unprecedented, and widely used. Based on a long-term research project, this paper seeks to overturn this consensus, using, among other avenues of inquiry, the analysis of industry discourse and a mixed-method empirical audience study of VOD users.