Experiments, benchmarks, and technical papers from Minds Research Lab on synthetic respondents, persona-conditioned AI, evaluation methods, and the limits of simulated evidence.
Total: 7
200 SINUS Milieu-Minds answered 24 questions on personal concerns. Per milieu, their answer distributions reached 91.7% closeness to a SINUS survey of 2,005 people, against 96.3% between two real survey waves.
An ANES pre/post-election comparison tests 300 matched people on 28 later answers, with separate results for response format and participant grounding.
A CES comparison evaluates 300 matched respondents on 27 binary, ordinal and multiselect questions, separating aggregate fit from individual prediction.
A target-separated GSS pilot compares four response conditions across 360 people and 17 questions, separating probability elicitation from profile value.
A five-country PISA comparison evaluates 600 students and ten held-out questions, with weighted distribution errors and explicit country and language limits.
What five public survey datasets reveal about synthetic audience accuracy, from a 301-Mind Gen Z test to an international PISA replication, and where grounding helps.
An outcome-blind validation compared 903 responses from 301 persistent Gen Z Minds with published UK Food Standards Agency survey distributions. Aggregate approximation reached 93.99%.
