Synthetic Respondents Miss Nuance

Large language models can produce plausible survey answers, but they systematically misrepresent certainty, factual knowledge, and the diversity of responses across demographic subgroups. That makes them unreliable stand‑ins for real survey panels, especially for timely or high‑stakes questions. — If synthetic polling is treated as equivalent to real polling, journalists, pollsters and policymakers could be misled about public opinion and decision‑making signals around elections, public health and policy.