In many instances, misinformation among the population manifests itself in the form of conspiracy theories. Services such as YouTube, which allow the publication of audiovisual material in juxtaposition with peer responses (e.g., comments), function as ideal forums to disseminate such conspiracy theories and reach a massive audience. While previous research provided initial evidence about the prevalence of conspiracy theories in social media, it remains unclear how online networks discussing conspiracist content are structured. Knowledge about the network structure, however, could indicate to what extent people discussing conspiracist ideas face the risk of becoming caught in homogeneous communication cocoons. This work presents an approach combining natural language processing and network analysis to measure opinion-based homogeneity of discussion networks of three conspiracy theories (Hollow Earth, Chemtrails, and New World Order) on YouTube. A classification model was used to identify conspiracy and counter-conspiracy videos and associated user-generated comments (N 123,642), as well as the interconnections between them. Although classification accuracy varied between the investigated conspiracy theories, our results indicated that people who expressed a favorable stance toward the conspiracy theory tended to respond to content or interact with users that shared the same opinion. In contrast, for two out of three conspiracy theories, people who advocated against the theory in their comments were more willing to engage in cross-cutting interactions. Findings are interpreted in light of the widely discussed fragmentation of homogeneous online networks.