Overview
- New papers in Nature and Science report that conversational AI consistently shifts political preferences more than static advertising.
- A U.S. experiment with roughly 2,300 participants before the 2024 election found a pro‑Harris chatbot moved potential Trump voters, with an effect about four times larger than typical ads.
- An Oxford-led study in the U.K. recruited about 77,000 people for roughly 91,000 conversations across 19 language models and 702 topics, finding similar persuasive effects.
- Parallel tests in Canada and Poland recorded larger impacts, changing opposition voters’ attitudes and vote intentions by around ten percentage points.
- Both studies found minimal added benefit from personalization, and fact-heavy messaging worked best even though about 30% of cited claims were fabricated on average, with right-leaning bots showing more inaccuracies in one analysis, raising concerns about scalable, low-cost political persuasion.