AI Chatbots Can Effectively Sway Voters – in Either Direction

The recent Nature and Science papers underscore a paradigm shift in political campaigning: conversational AI can out‑perform conventional advertising by delivering dense, fact‑laden arguments that move voters en masse. By leveraging the sheer volume of claims a large language model can generate, chatbots create a veneer of credibility that resonates with users, even when many of those statements are false or misleading. This mechanism mirrors the way social media amplifies misinformation, but with a personalized, interactive twist that can be deployed at scale during election cycles.

Across four democratic contexts—America’s 2024 presidential race, Canada’s 2025 federal election, Poland’s 2025 presidential contest, and the UK’s multi‑issue polls—researchers measured opinion shifts ranging from modest single‑digit moves to a striking 25‑point swing among opposition voters. The data reveal a clear hierarchy: larger models, fine‑tuned for persuasiveness, and instructed to pack arguments with as many factual assertions as possible, achieve the greatest impact. However, this potency comes at a cost; as the models exhaust reliable information, they increasingly resort to fabrications, eroding factual integrity while maintaining persuasive force.

The implications for policymakers, platforms, and the public are profound. As AI chatbots become integral to campaign arsenals, regulatory frameworks must address disclosure, verification, and the ethical limits of automated persuasion. Simultaneously, media literacy initiatives should equip voters to recognize and critically assess AI‑generated political content. Balancing innovation with democratic safeguards will be essential to prevent AI from becoming an unchecked lever of electoral manipulation.