This work presents an interdisciplinary methodology for developing endogenous measures of group membership through analysis of pervasive linguistic patterns in public discourse. Focusing on political discourse, this work critiques the conventional approach to the study of political participation, which is premised on decontextualised, exogenous measures to characterise groups. Considering the theoretical and empirical weaknesses of decontextualised approaches to large-scale social phenomena, this work suggests that contextualisation using endogenous measures might provide a complementary perspective to mitigate such weaknesses. This work develops a sociomaterial perspective on political participation in mediated discourse as affiliatory action performed through language. While the affiliatory function of language is often performed consciously (such as statements of identity), this work is concerned with unconscious features (such as patterns in lexis and grammar). This work argues that pervasive patterns in such features that emerge through socialisation are resistant to change and manipulation, and thus might serve as endogenous measures of sociopolitical contexts, and thus of groups. In terms of method, the work takes a corpus-based approach to the analysis of data from the Twitter messaging service whereby patterns in users’ speech are examined statistically in order to trace potential community membership. The method is applied in the US state of Michigan during the second half of 2018—6 November having been the date of midterm (i.e. non-Presidential) elections in the United States. The corpus is assembled from the original posts of 5,889 users, who are nominally geolocalised to 417 municipalities. These users are clustered according to pervasive language features. Comparing the linguistic clusters according to the municipalities they represent finds that there are regular sociodemographic differentials across clusters. This is understood as an indication of social structure, suggesting that endogenous measures derived from pervasive patterns in language may indeed offer a complementary, contextualised perspective on large-scale social phenomena.