How parties and politicians relate themselves to social groups is a question at the heart of comparative politics. However, studies addressing this question are limited in scope because they rely on time and resource-intensive manual content analysis to quantify group appeals. We argue that automated methods can alleviate this constraint to comparative research. This paper makes the first necessary step in this direction. We propose two scalable strategies for measuring how parties and politicians relate themselves to different social groups in their rhetoric. Our measurement strategies cover two common scenarios. In the first scenario, a researcher wants to extract and compare references to a pre-defined (set of) social group(s). We present dictionary expansion methods that enable measurement in this scenario and illustrate their use for quantifying British parties' appeals to the unemployed and the working class, respectively. In the second scenario, a researcher is interested in identifying all groups referenced in a corpus of political texts. We propose a strategy combining crowd coding and supervised learning in a novel text annotation and token classification setup that addresses this second scenario. We apply this strategy to extract the group appeals of British mainstream parties between 1945 and 2019. By presenting two scalable measurement strategies for identifying social groups mentioned in political texts, we enable novel comparative research on social group appeals in political rhetoric.