Answers
AI polling accuracy
AI polling is accurate only where the vendor publishes the miss. Lewsearch's best panel is 4.88% MAE on 97 Texas ex-electoral items. Texas political approval is 3.43% on 18 questions. Electoral-margin items on that same panel are 12.23%. Those three numbers belong in the same paragraph.
Approval is easier than margins
The UT/Texas Politics Project block is the cleanest published political set: governor, legislature, senators, economic direction, three options, 1,000 Texas residents per item, 5-fold CV. Aggregate approval MAE is 3.43%. The same production stack on electoral-margin questions on that panel is 12.23%. California approval on PPIC is 7.94% (15 items). Method is on methodology. Coverage (27 live panels · U.S. national · 4 census regions (32 audiences)) is on coverage.
Worked example: published before the hearing
On July 6, 2026, Lewsearch fielded a 600-respondent, census-matched read on the Prince George's County data-center pause and posted it before the July 7 Council hearing. The council passed the moratorium. The artifact is a public-opinion study, not a probability sample of recorded votes. It is useful as a timestamped directional read. It is not a claim that synthetic panels replace election polling. The report and JSON sit at /reports/prince-georges-data-centers.
A second public check used real-money markets instead of a human survey. On 98 resolved Trump-approval contracts, with no settlement leak, the panel's direction accuracy was 70.4% against Polymarket's 72.4%. The write-up is on /proof. The honest line there: real, not alpha.
| Read | n | Result |
|---|---|---|
| TX political approval | 18 Q | 3.43% |
| TX panel, ex-electoral | 97 Q | 4.88% |
| TX electoral margin | 23 Q | 12.23% MAE |
| Held-out political approval | 5 Q | 7.24% MAE |
| Polymarket direction (approval contracts) | 98 resolved | 70.4% vs market 72.4% |
FAQ
- How accurate is AI polling?
- On Lewsearch's best published panel, 4.88% MAE on 97 Texas ex-electoral items (UT/Texas Politics Project). Political approval on that panel is 3.43% (18 questions). Electoral-margin items on the same panel are 12.23%.
- Can an AI poll call an election?
- Electoral-margin questions are the hardest category Lewsearch publishes. Do not treat a synthetic panel as a substitute for a probability sample or a certified canvass.
- Did Lewsearch publish anything before a real vote?
- Yes. The Prince George's County data-center moratorium read was fielded July 6, 2026, and published before the July 7 Council vote. The council passed the pause. That is a public-opinion read, not a poll of recorded votes.
- How did the panel do against Polymarket?
- On 98 pre-registered Trump-approval contracts, direction accuracy was 70.4% versus the market's 72.4%. When the panel disagreed with the market (n=26), it was right 46% of the time. Receipts are on /proof.
Check the work
Related
Canonical: https://lewsearch.com/answers/ai-polling-accuracy