Partisan patterns of compliance with public health measures are a feature of early COVID‐19 responses. In many cases, these differences in behaviour relate to pre‐existing group identities. However, in times of rapid societal change, novel opinion‐based groups can emerge and provide a new basis for partisan identification and divergent collective behaviour. Here, we use network methods to map the emergence of opposing opinion‐based groups and assess their implications for public health behaviour. In a longitudinal study, we tracked public health attitudes and self‐reported behaviour in a sample of UK participants over four time points. Network visualisation reveal a rift in attitudinal alignment over time and the genesis of two distinct groups characterised by trust, or distrust, in science (Study 1a; N = 253). These groups also diverge in public health behaviour. In a brief follow‐up study (N = 206), we find that this opinion polarization partially reflects underlying societal divides. We discuss implications for opinion‐based group research and public health campaigns.