Politicians appeal to social groups to court their electoral support and secure their political survival. But quantifying which groups politicians refer to, claim to repre- sent, or address in their public communication presents researchers with challenges. We propose a novel supervised learning approach for identifying group mentions in political texts. We first collect human annotations to determine the exact text pas- sages that refer to social groups. We then fine-tune a Transformer language model for contextualized supervised classification at the word level. Applied to unlabeled texts, our approach enables researchers to detect and extract word spans that contain group mentions automatically. We illustrate the reliability, validity, and flexibility of our ap- proach in a study of British parties’ election manifestos, parliamentary questions, as well as German party manifestos. Our application demonstrates that our method en- ables highly reliable retrieval of group mentions at scale and new quantitative insights into group-based political rhetoric.