Often social scientists want to label whether text is populist or anti-elite in some sense. Traditional methods of content analysis tend to run into one of two problems. Labeling text by hand is taxing, limiting the scope of the analysis. Alternately, labeling text based on political-party affiliation elides variation within political parties and does not tend to work well for two-party systems. I use recent breakthroughs in natural language processing (NLP) combined with supervised learning to explore an alternative way of labeling text as anti-elite that avoids these constraints, allowing sentence-level categorization at scale.