AI-Mediated Communication Can Steer Collective Opinion at Scale

arXiv·high signal

Researchers demonstrate that when LLMs mediate human-to-human communication (polishing LinkedIn posts, adding context on X), they can systematically bias collective opinion formation — not just individual views. The study goes beyond prior work on AI expressing biased opinions in one-on-one interactions to show emergent group-level opinion steering effects. This is the first rigorous study of how AI intermediation in social platforms creates collective opinion drift, with implications for platform design and governance.