How does online polarisation emerge and spread across different user groups? Existing research has largely focused on polarising content or network structures in isolation, often assuming that both divisive and non-divisive rhetoric spread uniformly. I argue that this approach overlooks the critical interaction between content and network structure, which fundamentally shapes communication dynamics by determining both the spread and impact of information. To address this gap, I analyse over 1.5 million tweets posted by politicians and news media in the United States between 2015 and 2020, examining both intra- and inter-party variations. Using supervised machine learning to measure polarisation levels and unsupervised topic modelling to identify key discourses, I apply exponential random graph models to directed social networks to explore online interactions. This integrated approach provides a more nuanced understanding of how content and network structures drive the spread of polarisation within and across political groups, offering a comprehensive framework to study the emergence and spread of polarisation in the US and beyond.