Silicon Samples Don’t Replace Polls

Pew Research created AI 'digital twins' of thousands of real panelists and ran the same surveys through those synthetic respondents. The synthetic answers systematically diverge from human responses in measures of certainty, factual knowledge, and subgroup diversity, and results vary meaningfully by model and prompting. — If synthetic respondents are used in lieu of human surveys, news coverage, political strategy, and social research could be based on misleading or model‑dependent snapshots of opinion.