✦ AI audience simulation
Village simulates a real audience: distinct people with real biases and real reactions. Paste your text, choose who's in the room, and get instant, honest feedback and scores before you hit send.
“The wedge is clear, but “1,200 users” tells me nothing without retention.”
How it works
Paste your text, choose who reacts, and read the room in three steps.
Drop in any text: an email, post, ad, or pitch.
Choose a built-in audience or build your own, and how they react.
They join, react, and score it, with their reasoning.
Audiences
Use a built-in panel or describe your own in plain English. Village generates a panel of distinct personas with real roles, demographics, and the biases that come with them.
Conversation
Once the reactions land, keep going. Question the whole room, or @mention specific personas directly.
The science
Real people, real biases, real reactions, validated against how humans actually respond.
Tested against real human outcomes, frontier-model predictions track closely, even on studies they'd never seen.
Each dot is a study. The tighter they hug the line, the better the model matched real people. Here r = 0.90, where 1.0 is a perfect match.
You get the whole distribution of reactions and scores. The disagreement is the signal, not noise to flatten.
Each bar is one persona's score; the dashed line is the room's average. Hover a bar to see its score.
Accuracy
In one of the largest tests of its kind, researchers compared what frontier models predicted against how real people actually behaved.
Correlation between AI predictions and real human outcomes, on studies the models had never seen before.
Stanford · 70 pre-registered experiments · 2025
Real participants those predictions were tested against, across 476 experimental conditions.
Stanford behavioral study · 2025
How well agents reproduced 1,052 real people's own survey answers, vs. those same people two weeks later.
Stanford generative agents · 2024
Figures from published Stanford research, incl. AI-simulated human subjects (2025) and generative-agent simulations (2024).
And the science keeps advancing. Still, Village is a fast, directional read that augments real research. It isn't a replacement for talking to people.
FAQ
Village is an AI audience simulator. You paste text such as a pitch, essay, ad, resume, or idea, and a panel of distinct AI personas reacts with honest feedback and scores, so you can read the room before you hit send.
A chatbot gives you one averaged voice. Village runs a panel of distinct personas with different biases who react, score, and sometimes disagree, so you see the real range of reactions your text will get instead of a single opinion.
Frontier models predict how real people react with surprising accuracy. In large Stanford studies, AI predictions correlated about 0.90 with real human outcomes. Village is a fast, directional read that augments real research, not a replacement for talking to people.
Yes, you can try Village for free. A free run shows you how your text lands; deeper features and more audiences come with a paid plan.