This paper describes a new approach to defining genre. A model is presented that defines genre based on likability ratings rather than features of the content itself. By collecting hundreds of thousands of likability ratings, and incorporating these into a topic model, one can create genre categories that are interesting and intuitively plausible. Moreover, we give evidence that likability-based features can be used to predict human annotated genre labels more successfully than content-based features for the same data. Implications for outstanding questions in genre theory are discussed.