AI-Supported Information Seeking: The Role of Sycophancy and Biased Information on Attitudes on Public Policy

As AI-supported search becomes an increasingly common mode of information seeking, questions arise about how AI shapes public opinion on policies. Drawing on the Search Engine Manipulation Effect and the HAII-TIME framework, this paper reports findings from a pre-registered between-subjects experiment examining how AI information bias and sycophancy influence attitudes toward a lesser known policy of ranked-choice voting (RCV). Participants were randomly assigned to one of four conditions varying the bias of the information in search returns (pro vs. anti-RCV), and message tone (neutral vs. high sycophancy). Results indicate that information bias had a strong effect on RCV attitudes. Sycophancy produced a smaller but significant independent effect. Prior trust in AI weakened susceptibility to biased content, while machine heuristic beliefs amplified susceptibility to biased content; prior exposure to RCV buffered against both effects. These findings raise concerns about the role of AI in shaping public opinion on low-salience policy topics.