Synthetic Polls Misstate Certainty and Knowledge

When an AI model is asked to ‘be’ real respondents (a digital‑twin method), its answers diverge from real humans in predictable ways: it tends to display different levels of confidence and gives different answers on factual knowledge questions. That divergence persists across models and question types and can distort measures that rely on uncertainty or knowledge metrics. — If synthetic surveys systematically misstate how sure people are or what they know, they can mislead journalists, policymakers and researchers who rely on poll measures of public knowledge, uncertainty and issue salience.