syntheticperson.ai
In perspective · by thecognitivelab
Ask a million people.
Read the direction today.
We build a population that matches Brazil’s census and put your question to it. It tells you which way the market leans and which option wins — not the decimal. We publish how far off it lands.
Three steps.
That is the whole product.
You choose who
Region, class, generation, schooling, housing, work. Before we generate anything we tell you how many people that cut describes — and which parts of it we cannot count.
We build them
Each person gets a full interior, not a label: fears, income, dialect, plans, motivation. Coherent enough to sit still for an interview and explain themselves.
You ask
A price, a concept, a screenshot, a campaign. The whole batch answers at once — and you can open any single one of them and talk.
What you can do with it.
Test a concept
Put a product, feature or campaign in front of the population before you build or launch it.
Find the price
Probe willingness to pay across segments without burning a real launch to find out.
Cut the market
See who actually buys, and how the population splits on needs, attitudes and triggers.
Choose the words
Test positioning and copy variants at a scale no focus group reaches.
Interview one person
Open any individual in the batch and have a real conversation, one to one.
Read the room
Beside the answer, what went unsaid: the hesitation, the reservation, the thing a survey never captures.
The easy version of this
would be a lie.
Anyone can generate a thousand plausible people in an afternoon. They will agree with whoever made them, and nobody will be able to tell. That product is faster to build, cheaper to run, and impossible to trust.
So every population we build is fitted to official statistics, scored against studies whose answers we hid from it, and shipped with the ruler attached.
How we keep it honest.
No invented data
Every calibration figure names the table it came from. Where a licence has not been verified, we record the assumption and its date rather than claiming permission we do not have.
Privacy
Synthetic people represent real ones. Every free-text conversation passes a three-layer guardrail: requests for personal detail are refused at the gate, before any model is called, and formatted identifiers are scrubbed on the way out.
No caricature
Generated people are reviewed against class and income stereotyping before they reach the product. A population that flatters the brief is worse than no population.
Density limits
Cuts thin enough to approach an identifiable individual are refused. Where a dimension is unmeasured the estimate is treated as a floor, and the answer degrades to “unknown” rather than to a reassuring one.