The Center for Media and Communication Research (CMCR) hosted a research talk by Professor Baohua Zhou from the School of Journalism, Fudan University on February 3, 2026. Professor Zhou presented his research titled “Simulating Public Opinion with Chain-of-Thought (CoT) in Large Language Models.”

In his talk, Professor Zhou explored whether large language models (LLMs) can be used to simulate public opinion on social issues beyond elections, with a focus on cross-national contexts. His study examined public health opinions and compared models developed in the United States and China, including ChatGPT and ZhipuAI’s GLM.

Professor Zhou highlighted the importance of prompting strategies in improving simulation performance. He explained that prompts encouraging step-by-step reasoning or role-playing help large language models better approximate real-world public opinion at the aggregate level, even though individual-level simulations remain less accurate. He further noted that, in generating responses, LLMs tend to draw on core demographic factors such as education, income, and age—patterns that align closely with established approaches in social science research.

Professor Zhou concluded that although LLMs cannot replace traditional surveys, especially for individual-level analysis, they offer promising tools for methodological innovation in public opinion research.