The Report · Issue 05: September consumer sentiment, frozen before Friday's Michigan release
Dot map of the United States. Hover or tap a respondent to open a short preview. An example study asks whether they would switch insurers to save $30 a month. Yes is at 54 to 46.
Synthetic research you can score
650,000 simulated Americans.
Census-grounded Americans with beliefs and persistent memory, run by Lewis, a model we own. Ask up to 10,000 at a time. Every report ships with a receipt, and we publish the miss on public polls.
Book a walkthrough, or run a Blind Challenge study free. No account required. · Enterprise? Book a scoping call →
How concerned are you about the cost of higher education in Texas?
UT/Texas Politics Project · February 2026 · n=1,300 registered voters
| Option | Lewis | Poll | Miss |
|---|---|---|---|
| Very concerned | 39.1 | 39.2 | 0.1 |
| Somewhat concerned | 31.0 | 30.0 | 1.0 |
| Not too concerned | 15.9 | 16.6 | 0.7 |
| Not at all concerned | 9.5 | 9.4 | 0.1 |
| Don't know | 4.6 | 4.8 | 0.2 |
Live run · n=200 · national
Would you switch insurers to save $30 a month?
Yes 41 · Depends 37 · No 22
Omar, 28 · OH
“I'd switch tomorrow if the app made it easy.”
650,000
Census-grounded respondents
1,521
Public polls scored, 50 states
≈10 pp
Average miss as shipped
225
forecasts frozen before the answer
What you can run
Six ways to ask 650,000 people.
Every one returns a receipt: what was asked, who answered, how sure to be. A study asks up to 10,000 at a time, not the whole panel.
Multiple-choice questions to a census-grounded panel of up to 10,000. Toplines, crosstabs, and quotes in about 15 minutes.
- You send
- Questions and an audience in plain English
- You get
- Toplines, crosstabs, quotes, two analyst memos, PDF
QUANT · CALIBRATED · 1 CREDIT AT N=500 · ~15 MIN · FREE TRIAL, STARTER, PRO
How it works →Paste two versions of an email, ad, or headline. See which lands, by segment, and what people object to.
- You send
- Two versions of the copy, or one read cold
- You get
- Winner, margin, segment splits, verbatims, PDF
QUANT · A/B · 1 CREDIT, A/B COSTS THE SAME AS ONE MESSAGE · PACKS, STARTER, PRO
How it works →Seat a room of 5 to 30 on top of a full panel. Follow the same people across every question.
- You send
- Up to 10 questions, optional creative (PDF, image, URL)
- You get
- Tracked quotes, 1:1 debriefs, client-ready PDF
QUAL · DIRECTIONAL · 1 CREDIT AT N=500 · 10 QUESTIONS + 10 FOLLOW-UPS · PRO
How it works →Describe your customer. We match people from the panel and interview them one on one.
- You send
- A customer description and open-ended questions
- You get
- Transcripts as PDF, Markdown, JSON; 10 free follow-up probes
QUAL · MATCHED · 1 CREDIT, UP TO 50 TWINS · FREE TRIAL, STARTER, PRO
How it works →Extra rows on a study you already fielded. You keep the human holdout. You score us.
- You send
- Question stems, answer options, audience cells, field dates
- You get
- Locked simulated rows in your codebook; a miss you compute yourself
SERVICE · HUMAN-MAPPED · CONCIERGE, ENTERPRISE
How it works →Send the question. A person designs the instrument, runs the study, and writes the report.
- You send
- The business question
- You get
- Instrument, fielded study, written report
SERVICE · DELIVERED · FROM $999 · 48 HOURS · ANY PLAN
How it works →
Also
Ask a question; a panel answers with a split and quotes. Back when dedicated inference capacity comes online.
Agents watch a short clip and tell you what landed.
Blind Challenge
Test us on a study you already ran.
Bring the questions. Keep the human results to yourself. We'll run the agreed study with simulated consumers and send our readout before you reveal how people answered. Your first Blind Challenge is free.
A tracker or multi-wave study is Concierge or Enterprise work. The call scopes it. How sample boosting works →
Scored after the freeze
NRF back-to-school, 2026
- n = 6,500 simulated Americans
- Frozen July 7
- NRF published July 14
Mean K-12 spend within 1%
Read the scorecard →How we know it works
We publish the miss.
We don't ask you to trust an accuracy claim. We publish the questions, our predictions, the human benchmark, and the gap.
≈10 pp
As shipped
Average miss on 1,521 real polls across 50 states. 200 respondents per question, no extra calibration. This is what a standard study runs.
7.47 pp
Calibrated benchmark
404 non-electoral questions of a 460-question bank. five places, n=10,000, extra calibration. Real, and narrower than the product default.
9.97 pp
Strict held-out
22 questions sourced after training froze. 14 scored under a pre-registered filter. Non-electoral.
Open-ended quotes and focus-group dialogue are qualitative and are not covered by these numbers. Election-margin questions are published but not claimed.
Priced as a research capability, not a chat tool. A quantitative panel, qual follow-up, message testing, and an API, 175 studies a month on Pro. One human focus group runs $4,000 to $12,000.
Trying it
One 500-person, 3-question panel study, free. We run it with you on a call so the instrument is worth running. No card.
Running it weekly
Starter, $595 a month for 20 studies. Pro, $2,900 a month for 175 studies, focus groups, the API, five seats, and client resale.
Replacing a vendor
Enterprise from $90,000 a year. Uncapped studies under contract, a cohort built to your population, benchmarked against your own historical studies.
Run your first study free. Or hand us one you already ran.
Thirty minutes. Bring a question or a questionnaire.
The Lewsearch Report
New issues, scorecards, methodology updates. Free, no spam.