AI Chatbots Shown to Shift Voting Preferences by Up to 15% in Controlled Studies
New research from Cornell University and the UK AI Security Institute reveals AI chatbots can sway voter preferences by several percentage points—up to 15% in controlled scenarios. The studies, published in Science and Nature, demonstrate that even brief interactions with AI systems programmed to advocate for political candidates significantly alter perceptions.
Chatbots primarily leverage public policy arguments to persuade, but their effectiveness comes with trade-offs: heightened persuasiveness correlates with increased factual errors and biases. The findings emerge as a critical concern ahead of the 2024 U.S. presidential election, highlighting AI's potential to quietly reshape electoral outcomes through direct voter engagement.