A calibrated world
Ask the population
Test a price, a message or a policy on a synthetic population built from census and survey data before you commit. Every answer comes with its interval, who said what, and the reasons people give.
A survey tells you what people said last month. A model tells you what they might say next month, if it was built from the same people.
census.world is a synthetic population: one profile for every resident, matched to the census and the national surveys on who they are, where they live, what they do, what they earn, how they vote. No profile is a real person. Together they match the published tables, and every check is public.
Ask them a question and a representative sample answers, one at a time, in character, before you commit to the price, the message or the policy.
How it works
Every profile starts in the census: state, age, sex, race and origin, reproduced exactly. National surveys add work, income, family, health, civic life. Where the record is thin the population says so.
Published tables are held out of the build and compared state by state after it. Every kind of question is also checked against a real survey, and the answer page says how far it sits.
The question is read for its population and its kind, a sample is drawn, each profile answers as itself, and the answer settles with its interval. Then some of them explain why, in their own words.
Every breakdown, the states, the market the answer implies, the kinds of people in it, every reason, and the next question asked of the same people.
Where this goes
Today it is a survey without the fieldwork: an answer in a minute, on the people you name, with the reasons.
Next it is a population you can follow. The same people asked again after the price changed, the message ran, the policy passed. Households that decide together. Other countries with the depth the United States has now.
Then a population that is tested against the world: launches, elections, markets, and corrected where it is wrong. The end is a standing model of what people will do, checked against what they did.
Read a finished report: what people would pay a month for a streaming service, or whether Ohio's self-employed would pay $50 a month for an AI bookkeeper.
Questions already asked
Yes-or-no, choices, scales, amounts, rankings, allocations and comparisons between two populations. Open one to see the answer settle, who said what and why.
Recently shipped
The population and the answers change most weeks. Everything that changed.
- September 6, 2026
- September 6, 2026
- September 6, 2026
- September 6, 2026
Questions people ask first
The short answers. The long ones are on the methodology page.
Is this the census?
No, and it is not affiliated with the US Census Bureau or any government. It is a synthetic population built from the census and other public statistics: one profile per resident, matched to the published counts, that you can ask questions.
Is this a survey?
No. Nobody is interviewed. A synthetic population built from public statistics answers, one profile at a time, through a language model. What you get is what a representative sample of that population says, with the sampling interval a survey of the same size would carry.
How accurate is it, and how do you know?
Two ways. The population is checked against official tables it was never fitted to, state by state; every check is on the validation page with its number. The answers are checked against real surveys, kind by kind: on yes-or-no questions from the General Social Survey they sit a median 17 points from real people’s within demographic groups and 8 points nationally, and every answer says the number for its own kind.
Is anyone in this data real?
No. Every profile is synthetic, built from aggregate statistics and anonymized survey records, with an id that carries no information about anyone. There is no real person to find.
What can I ask about?
Any population you can describe with the attributes the profiles have: state or country, age, sex, race and origin, education, employment, income, household, health, civic life. Opinions, choices, amounts, rankings, allocations and comparisons between two populations. Not yet: cities and counties, and your own customer lists.
What does it cost?
The first hundred profiles of every question are free. To run a question on the whole population, or on your own segments and geographies, tell us what you want to run and we come back within a day. Design partners get ninety days of everything, free, for real use and honest feedback.
Join the waitlist
Design partners get a trial: ninety days of running their own questions with us, the full report on each, and a direct line to the team, in return for real use and honest feedback on what should come next.