Answers
Synthetic panels with living agents
A synthetic panel with living agents is a census-grounded set of simulated respondents that keep a persistent profile and a memory of prior answers, then get tabulated like a survey. Persistent memory means the same simulated respondent can be asked again. A fresh prompt with no stored memory is a different design. The calibrated 50-state benchmark: 7.07-point average miss on the best 80% of 1,521 scored questions (1,216 retained). 10.02 points across every scored question. A narrower April test on 5 places with 10,000 respondents scored 7.47 points on 404 non-electoral questions.
Persistent memory
Each Lewsearch record keeps a persistent demographic profile and a memory of prior answers, so related questions on the same panel stay tied to the same simulated respondent. Focus-group rooms carry that memory across the session. The panel is still the statistic: option shares you can score. See methodology.
Pew Research Center reported on Sept 30, 2026 that AI respondents "differed from their human counterparts by an average of 12 percentage points" across nearly 300 questions, after building a twin per panelist from demographics and political typology answers, on a question set that is not Lewsearch's 50-state file.
Field-date window
Evaluation is anchored to the midpoint of the source survey's field-date window. Period context is injected at inference so a historical item is answered from that window. That temporal discipline is what keeps a benchmark item from being answered with later news.
| Object | Memory | What you score |
|---|---|---|
| Living agent panel | Persistent profile and prior answers | Option shares vs a real poll (MAE in points) |
| Prompt with no stored memory | No persistent record | Score it only if the vendor publishes a file |
| Industrial digital twin | Machine or customer telemetry | A different object from a survey MAE |
The published miss
The calibrated 50-state benchmark: 7.07-point average miss on the best 80% of 1,521 scored questions (1,216 retained). 10.02 points across every scored question. MAE is an average miss in percentage points against published poll toplines. A narrower April test on 5 places with 10,000 respondents scored 7.47 points on 404 non-electoral questions. Coverage: over 650,000 simulated respondents. 650,000 census-grounded respondents · all 50 states and D.C. · 16 occupation panels.. The headline file is state-benchmark-per-question.csv (1,521 scored questions).
FAQ
- What are synthetic panels with living agents?
- A synthetic panel with living agents is a census-grounded set of simulated respondents that keep a persistent profile and a memory of prior answers, then get tabulated like a survey. Persistent memory means the same simulated respondent can be asked again. A fresh prompt with no stored memory is a different design. The calibrated 50-state benchmark: 7.07-point average miss on the best 80% of 1,521 scored questions (1,216 retained). 10.02 points across every scored question. A narrower April test on 5 places with 10,000 respondents scored 7.47 points on 404 non-electoral questions.
- What does persistent memory mean here?
- Each simulated respondent keeps a demographic row and a memory of prior answers, so related questions on the same panel stay tied to the same record.
- Do they get today's news?
- Inference is anchored to the source survey's field-date window. A February 2024 question is answered from a February 2024 state of the world. That keeps the published miss tied to the poll it is scored against.
- How accurate is the panel?
- The calibrated 50-state benchmark: 7.07-point average miss on the best 80% of 1,521 scored questions (1,216 retained). 10.02 points across every scored question. MAE is an average miss in percentage points against published poll toplines. A narrower April test on 5 places with 10,000 respondents scored 7.47 points on 404 non-electoral questions.
Check the work
Related answers
Canonical: https://lewsearch.com/answers/synthetic-panels-with-living-agents