lewn
Sign in

Ask a thousand people. In ten seconds.

Not one AI’s guess — a calibrated panel of demographically-grounded respondents tells you what people actually think.

n=100 → 2,500 respondentsbenchmarked against Pew & Gallup

Example · poll result

“How appealing is a 4-day work week?”

Asked 1,000 US adults · multiple choice

Very appealing58%
Somewhat appealing27%
Not appealing15%
lewn·aiasked 1,000 · benchmarked vs. Pew & Gallup

Why not just ask ChatGPT?

Ask ChatGPT “what would young men think of this?” and you get one answer — its single best guess, blended into one voice. It can’t tell you how many agree, where they split, or how close it actually is. Lewn samples a population.

ASK CHATGPT

  • —One blended answer
  • —The model’s “average” voice
  • —Disagreement is hidden
  • —Not calibrated to anyone
  • —Subgroups are guessed

ASK LEWN

  • A distribution across hundreds of people
  • Distinct, demographically-grounded respondents
  • Majority and dissent, both shown
  • Benchmarked vs. Pew & Gallup (≈±8 pts)
  • Cross-tabs by age, region, politics

ChatGPT is great when you want one smart answer. Lewn is for when you need to know what a population thinks — and how sure you can be.

Pricing

Free

$0/mo

n = 100

baseline volume

directional

Plus

$20/mo

n = 500

37× the free volume

indicative

Recommended

Pro

$70/mo

n = 1,000

200× the free volume

benchmarked · cross-tabs

Research

$200/mo

n = 2,500

1,000× the free volume

publication-grade · API

Frequently asked questions

What is Lewn?
Lewn (lewn.ai) is an AI tool that polls a panel of hundreds of demographically-grounded synthetic respondents and returns the full distribution of opinion — who agrees, who doesn't, and how it splits by age, region, and politics. It is the consumer-facing product from Lewsearch, benchmarked against real polls like Pew and Gallup.
How is Lewn different from ChatGPT?
Ask ChatGPT what an audience thinks and you get one blended answer — its single best guess, with no notion of how many people or which people. Lewn samples a population: hundreds of distinct respondents returned as a distribution, with subgroup splits and individual quotes, and the aggregate is benchmarked to real survey data (about ±8 points on calibrated questions). ChatGPT gives you the average voice; Lewn measures the room.
How accurate is Lewn?
Lewn runs on Lewis, the same calibrated model behind Lewsearch, benchmarked at 7.47% pooled error across 460 questions versus Pew, Gallup, PPIC, and official election results. Every result shows its own confidence; small-sample or un-benchmarked reads are labeled directional rather than dressed up as precise.
Is Lewn free?
Yes — a free tier polls a panel of 100 respondents with no signup. Paid tiers poll larger panels (up to 2,500 respondents) for tighter margins and stable cross-tabs.
What can I ask Lewn?
How an audience would react to a message, product idea, ad, or policy; which option a demographic prefers; or how opinion splits across age, region, income, and politics — answered in about ten seconds.