Ask any audience anything. Get benchmark-tested answers in minutes.

FAQ

Frequently asked questions

What is Lewsearch?

Lewsearch is an AI market-research platform. Instead of fielding a human survey, it polls panels of calibrated synthetic respondents to test messaging, ad copy, value props, and survey questions — returning percentages, crosstabs, and a client-ready PDF in under a minute. Its panels are benchmarked against real Pew, Gallup, PPIC, and election data.

How accurate is Lewsearch?

Across 460 independent survey questions from four panels (a Pew/Gallup/election-canvass benchmark, PPIC California, and the UT/Texas Politics Project), the production stack achieves 7.47% pooled mean absolute error, 5-fold cross-validated. On the Texas state panel it reaches 4.88% calibrated MAE and 3.43% on Texas political approval questions. On a truly held-out, pre-registered 22-question set sourced after training froze, it scores 10.68% overall — we publish both numbers so the real-world spread is visible.

How is it different from a traditional survey panel?

Speed and cost. Lewsearch returns a calibrated read in under a minute versus the days or weeks a human panel takes, at roughly 1% of traditional panel cost — with crosstabs by age, region, and politics, plus a client-ready PDF.

What can I test with Lewsearch?

Email subject lines, ad and landing-page copy, value propositions, survey questions, brand perception, and policy or issue support — anything you would otherwise put to a human research panel.

How does the calibration work?

Raw model outputs pass through Dirichlet calibration (Kull, Perelló-Nieto, Filipović et al., NeurIPS 2019) fit on held-out folds of the benchmark pool. The same calibrator ships to every customer — there is no per-panel retuning after the fact — and raw and calibrated error are published side by side.

Is Lewsearch free to try?

Yes. The sandbox runs a directional read with no signup. Paid tiers add larger panels for tighter margins, full crosstabs, analyst commentary, and a branded PDF report.

What is lewn, and when should I use it instead?

lewn (lewn.ai) is our consumer companion: a free, conversational way to ask a calibrated panel a single question and get a directional read in about ten seconds — no PDF, no setup. It runs on the same engine as Lewsearch. Use lewn for a quick gut-check; use Lewsearch when the decision is high-stakes and you need a client-ready report, multi-question studies, larger or private panels, crosstabs, an API, or team workflows.