Compare
Lewsearch vs Simile
Lewsearch publishes a pooled error rate against named public polls. Simile publishes validation volume, academic lineage, and a $200 million Series B, and does not publish a pooled MAE or a price list. The useful comparison is the rubric: published error, named benchmarks, n, and whether you can start without a sales call.
What each site actually says
Simile describes itself as a simulation company whose populations start with real interviews and are refreshed on new behavioral and news data. Its homepage states that it validates against real humans weekly, with over 7,000 evaluations covering distributional distance, effect size, and accuracy against observed behavior, and that an AI confidence model tags each result. It names CVS Health, Wealthfront, Gallup, and others as quoted partners. It does not publish a single pooled error rate with a denominator.
Lewsearch publishes one. 7.47% mean absolute error on 404 ex-electoral questions of a 460-question benchmark vs real polls (Pew, Gallup, UT/Texas Politics Project, PPIC). The scored file and the 5-fold cross-validation method are on the methodology page. Current market coverage is listed on coverage (27 live panels · U.S. national · 4 census regions (32 audiences)).
Worked example: one category, one n
On the UT/Texas Politics Project approval block (18 questions, three options, 1,000 Texas residents per item, 5-fold CV), Lewsearch's calibrated MAE is 3.43%. That is the strongest published category on the strongest published panel. Electoral-margin items on the same Texas panel are 12.23%. Both numbers sit on the same page. A buyer can see where the model is tight and where it is not.
Simile's site cites its founders' 2024 paper, Generative Agent Simulations of 1,000 People: interview-grounded agents reproduced participant responses at 85% of human self-retest accuracy. That is a published academic result on interview-grounded agents. It is not a production MAE against Pew or Gallup, and this page does not treat it as one.
| Rubric | Lewsearch | Simile |
|---|---|---|
| Published error rate | 7.47% MAE on 404 ex-electoral of 460 questions | Does not publish a pooled MAE |
| Named benchmarks | Pew, Gallup, UT/Texas Politics Project, PPIC, election canvass | Weekly human evaluations; academic papers cited on-site |
| n disclosed | 460 scored questions; per-panel n on /methodology | Over 7,000 weekly evaluations (homepage). No public-poll n |
| Public pricing | Yes, on /pricing | Does not publish pricing |
| Access model | Self-serve study plus demo | Request a demo |
| Company fact on their site | Columbus, OH. Swarmgram, Inc. | $200M Series B at $2B, July 30, 2026 |
How to use the table
Funding and partner logos are not accuracy. Simile is a larger company with a demo-led enterprise motion. Lewsearch is the vendor that posts the error file. If your procurement packet needs a scored question count against public polls, that cell is filled on one side and empty on the other. If you need interview-grounded enterprise simulation at Simile's stated scale, ask Simile for the evaluation pack they describe.
Sources
Competitor cells cite the vendor's own pages. Lewsearch cells cite /methodology.
FAQ
- Does Simile publish a pooled error rate?
- As of August 2026, Simile's homepage describes weekly validation (over 7,000 evaluations) and an AI confidence tag on results. It does not publish a pooled mean absolute error against a named public-poll set with a scored-question count.
- What accuracy number does Lewsearch publish?
- 7.47% mean absolute error on 404 ex-electoral questions of a 460-question benchmark vs real polls (Pew, Gallup, UT/Texas Politics Project, PPIC). The Texas state panel is 4.88% ex-electoral. The pre-registered held-out set is 9.97% ex-electoral. All figures are 5-fold cross-validated on the methodology page.
- How much has Simile raised?
- Simile's homepage, dated July 30, 2026, announces a $200 million Series B at a $2 billion valuation. That is a company fact, not a product accuracy fact.
- Which one has public pricing?
- Lewsearch publishes pricing. Simile's public site is demo-request. It does not publish a price list.
Check the work
Related
Canonical: https://lewsearch.com/compare/lewsearch-vs-simile