Collective polarization rebuilds from scratch whenever interaction stops

Interventions designed to reduce collective polarization, tested in multiple large-scale studies, consistently produce only short-lived effects. Here we show why: collective polarization requires continuous interaction to persist, and interventions targeting individual beliefs address the component that does not move. This research embedded 4 million UK parliamentary speeches (2007–2025) in semantic space, measured collective focus via the spectral gap of the embedding covariance matrix, and decomposed each speaker’s trajectory into a dispositional baseline and an interaction-dependent component. Individual baselines proved near-invariant, unchanged by Brexit, a pandemic, or seven changes of government, yet collective divergence reset to zero each time structured interaction ceased. The pattern was anchored by a small structural minority, formed exclusively during rising collective attention, and was causally identified through converging natural experimental evidence. The resulting divergence can be disrupted by changing the structure of interaction, not individual minds.