© 2020, Springer Nature Limited. Health researchers are encouraged to involve patients and the public in each stage of the research process. Patient and public involvement (PPI) seeks to redress an apparent ‘democratic deficit’ in the production of health research. While many claims have been made of the benefits of involving and increasingly co-producing research with patient and public members, it has proven difficult to measure its impact. To inform understanding of why involvement initiatives ‘fail’ or ‘succeed’, this paper focuses on the relational dynamics of PPI and co-production. Through an examination of the discourse of relationships in UK health research guidance it is argued that while emphasis is placed on ‘sharing power’, there persists limited understanding of the constraints of institutional structures and how power asymmetries come to be performed, reinforced or resisted in PPI and co-production settings. To reshape and rethink existing practices I examine the theories that underlie PPI and co-production referring to Habermas’ (The theory of communicative action, Polity Press, Cambridge, 1987) and Mouffe’s (Social Research 66:745–758, 1999; Politica Comun, https://doi.org/10.3998/pc.12322227.0009.011, 2016) contrasting perspectives on the role of public participation in democratic designs. I suggest that power differentials cannot simply be eradicated or bracketed by guidelines and rhetoric, and that PPI and co-production must be understood as profoundly relational and interactional processes.