From the makers of Lewsearch

The Lewsearch Report

Consumer and public-opinion data from calibrated synthetic panels: published before the human benchmarks, then scored against them in public. Every issue ships with methodology, raw data, and the error we got wrong last time. Free to cite with attribution.

Issues

Released every Wednesday

Longer studies, ~4 per month.

Pulses

Posted throughout the week as news breaks

Short reactive reads, 5–10 per month.

Issues vs Pulses

  • Issues release every Wednesday, pre-registered against a human benchmark. Wins and losses count.
  • Pulses are timely reads when news breaks. Outcomes are appended; they are not graded.

Subscribe · free

New issues, pulses, and public scorecards. No spam, unsubscribe anytime.

Issues

Released every Wednesday

Issue score tally

2 win/0 loss/1 pending

Wins and losses require a gradeable Issue with a pre-registered match to an external benchmark or vote. Pulses and instrument mismatches (e.g. Issue 01 P2/P3) count neither as wins nor losses.

Upcoming

  • August 6, 2026 · Issue 04 — topic TBD
  • August 13, 2026 · Issue 05 — topic TBD

Pulses

Posted throughout the week as news breaks

1 · Simulated, disclosed

Findings come from Lewsearch's synthetic panels: AI respondents grounded in U.S. census demographics and calibrated against real surveys. No humans are interviewed, and every page says so.

2 · Predict, then verify

Where a human benchmark exists (NRF, Pew, election results, council votes), we publish first, freeze the numbers, and append a public scorecard when the ground truth lands.

3 · Published error

The pipeline runs at 7.47% pooled mean absolute error across 460 cross-validated benchmark questions. Full methodology →

Citing this data

All findings are free to republish with attribution to “The Lewsearch Report” and a link to the piece. Describe the data as a calibrated synthetic panel with published error rates, not a traditional human survey. Press inquiries: jeffrey@lewsearch.com