<p>Critical discourse analysis has traditionally relied on the dissection of a limited number of texts, leading to criticisms regarding the subjectivity and lack of empirical rigour in its findings. In response, the integration of corpus linguistics provided more robust quantitative methods, but these have primarily focused on the statistical significance of lexical patterns, with limited capacity for deeper interpretive analysis. Recent developments in generative artificial intelligence offer new possibilities for interpreting complex discourse phenomena. This paper addresses how a large language model can interpret emotions within a critical discourse analysis framework. In this regard, our experiment aimed to assess affective polarisation in a corpus of YouTube comments about Brexit. The method primarily involves the identification of emotion-oriented lexical markers through lexical semantic resources, followed by their interpretation and contextualisation via the large language model. The findings suggest that large language models can enhance the analytical process by uncovering latent discourse patterns and offering nuanced interpretations. However, human analysts should still play an active role in guaranteeing the accuracy and validity of results.</p>