The Poll of AIs: Why Our Synthetic Polls Now Ask a Panel of Models, Not One
Every AI has a house lean, just like every pollster. We made five models from five labs run the identical poll on the identical 6,400 simulated voters across eight certified elections. The panel average beat every single model — typical miss about 1 point — and when the models disagreed, that disagreement pointed straight at the one scandal race every single one of them misread. Both findings now run inside every Civly candidate poll.
Read the research
