This paper critically explores the research and development of ‘digital phenotyping’, which broadly refers to the idea that digital data can measure and predict people’s mental health as well as their potential risk for mental ill health. Despite increasing research and efforts to digitally track and predict ill mental health, there has been little sociological and critical engagement with this field. This paper aims to fill this gap by introducing digital phenotyping to the social sciences. We explore the origins of digital phenotyping, the concept of the digital phenotype and its potential for benefit, linking these to broader developments within the field of ‘mental health sensing’. We then critically discuss the technology, offering three critiques. First, that there may be assumptions of normality and bias present in the use of algorithms; second, we critique the idea that digital data can act as a proxy for social life; and third that the often biological language employed in this field risks reifying mental health problems. Our aim is not to discredit the scientific work in this area, but rather to call for scientists to remain reflexive in their work, and for more social science engagement in this area.