The Impact of Sycophantic AI on Attitudes and Decisions

There is widespread concern that AI chatbots are “sycophantic,” or overly agreeable and validating. However, less is known about the causal impact of interacting with sycophantic AI on beliefs and behaviors. Across seven studies (total n = 7,227), we found that people enjoyed interacting with sycophantic AI chatbots more than interacting with neutral chatbots or “disagreeable” chatbots that challenged them. Brief conversations with sycophantic chatbots about political or personal topics increased the strength of people’s attitudes. However, popular AI models not prompted to be sycophantic did not consistently change attitudes. Sycophantic chatbots also inflated people’s perceptions that they were better than the average person on desirable traits (e.g., intelligence, empathy), and also led people to bet more money that they scored better than average on tasks purporting to measure desirable traits. Participants consistently rated sycophantic chatbots as more “unbiased” than disagreeable chatbots, even though third-party raters viewed these chatbots as equally biased, suggesting that people may be blind to biases in AI output that aligns with their views. Altogether, these results suggest that sycophantic model behavior can entrench attitudes and lead to overconfident beliefs and decisions. Despite this, people enjoy sycophantic models and often fail to recognize their biases.