YouGov validates AI twins against real panelists
YouGov and OriginalVoices built 1,008 LLM-based “E-twins” from identity-verified panelists and asked 693 of those people the same new questions in a recontact survey within 48 hours. YouGov reports that adding panelists’ ratings of more than 4,000 brands and entities increased person-level agreement within one point on a five-point scale from 78% to 90%. The twins also reached 90% agreement on fiction books, a topic outside their training data, while 75% of panelists said the twins’ open-ended answers described their own views. YouGov calls on simulation providers to validate twins against their corresponding people, test unseen topics, publish their methods and misses, and publicly predict real-world outcomes.
The study tests each AI twin against its corresponding panelist rather than relying only on aggregate comparisons, offering a proposed approach for checking whether individual simulations reflect real respondents. YouGov also argues that shared validation standards are needed because public datasets may already be present in language models’ training data and aggregate comparisons can conceal twins that revert to averages or stereotypes.

New York City, United States, November 10th, 2026, TechnologyWire
- YouGov created 1,008 “E-twins” and tested them against 693 real, government-ID verified people they were built from, asking both the same new questions
- Twins’ results were validated through a recontact survey of the panelists they were based on within 48 hours
- Twins trained on panelists’ ratings of 4,000+ brands and entities matched their human to within one point on a five-point scale 90% of the time, up from 78% on profiling data alone
- Twins hit the same 90% on a topic they were never trained on
- YouGov calls on the industry to agree on validation standards, and challenges every simulator to predict real-world outcomes in public