How synthetic polling results change based on the AI model

Survey experiments with Opus 4.6 and GPT-5.1 each paint a very different picture of U.S. public opinion, neither of which reflect reality.