Test your idea on 1,000 simulated minds before you spend a single euro.
Describe your idea. Windrose Atlas simulates how your target customers would respond, and gives you a ranking, a direction, and hypotheses worth testing. Not made-up percentages.
Atlas: because every run opens a map of a population — parameterised from documented cross-cultural frames, or drawn from your own customers.
FIRST RUN FREE · NO CARD
RECORDED · YOUR VOTE SHAPES WHICH AUDIENCES WE BUILD NEXT.
FIRST RUN FREE. NO CARD.
Describe your idea in plain language.
No survey design, no brief. One paragraph is enough.
We simulate a population, not an average.
1,000 personas respond through a cognition model trained on millions of real human decisions from published behavioural experiments. They are parameterised from documented cross-cultural frames, or clustered from your own customer data when you bring it. Counterbalanced †2, distribution-first.
You get a compass, not a verdict.
Which variant is stronger, for whom, why, and exactly what to test in the real world next.
Any AI can print "your conversion will be 23%." It will be wrong.
Simulated research has a dirty secret: the numbers look precise and aren't. Peer-reviewed work †1 shows that language models do not reliably reproduce human psychology, absolute predictions drift, distributions collapse, and confident decimals hide guesswork. We built Windrose Atlas on that finding, not against it.
Read the full argument →A compass doesn't tell you where you'll land. It tells you which way to walk.
This page was tested on Windrose Atlas before you saw it.
We simulated three versions of this very headline on a population of founders. The one you're reading ranked first. After launch, we'll publish the real numbers next to what we simulated, direction, not decimals. That's the loop: simulate, then verify.
Read the build story →A simulation is an experiment, not a substitute for data.
These aren't made-up personas standing in for real people. Each run is a controlled experiment, built on documented frames — and the point isn't the raw output. It's what happens when your real outcomes come back and we set them side by side.
1,000 minds, parameterised from published frames, respond under controlled conditions — a designed experiment, not a generated shortcut.
When your real outcomes arrive, we place them next to the experiment. That comparison — not the raw run — is where the knowledge and the context live.
Nothing is thrown away. Every checked experiment joins a growing base of evidence that makes the next read more grounded. Not a one-off report — a method that compounds.
Two ways in. One instrument.
One paragraph in plain language. The population is parameterised from documented frames: Hofstede dimensions today, WVS-grade survey data as it lands on the changelog. Results read RUNG: SIM-RANK, CALIBRATION: NONE, and the label says so out loud.
Run without data →Upload a CSV or XLSX. Personas are clustered from your real customers, not invented. Any dimension your data actually labels is validated against it and reads RUNG: REAL. The simulation itself passes a blind validation and is calibrated, real to sim, never the other way.
Upload your data →The evidence ladder isn't a promise. It's a mechanic: want a higher rung than SIM-RANK? Bring outcomes.
Cheaper than one lunch. Radically cheaper than being wrong.
The real cost isn't the subscription, it's the quarter you spend building the wrong thing. Here is what a read costs against what a wrong bet costs.
Anchor scale, all four tiers, per-country pricing and billing live on the pricing page.
Reserve your founder price before we open.
The founder price is for people who back this early. Reserve now and it stays your rate for as long as you stay a member, even when public pricing goes up.
- →Founder price held for as long as you stay a member, even when public pricing goes up.
- →First access when runs open, ahead of the public queue.
- →No card now. We only ask when you decide to run for real.
We're getting the first runs ready. Founder pricing will be reservable right here shortly — check back soon.
Your idea never leaves the building.
The behavioural oracle runs on our own hardware. An external frontier model designs scenarios and second-guesses the ensemble, and it receives aggregates only, never your raw records. Your inputs are not training data. Deletion is a protocol, not a promise.
How we handle your data →Frequently asked, honestly answered.
What is simulated market research?
Simulated market research tests an idea against a simulated population of personas instead of recruiting real respondents. Windrose Atlas builds 1,000 personas, parameterised from documented cross-cultural frames or clustered from your own customer data, then reports the direction of an effect and the ranking of your options, never absolute numbers.
Is AI market research accurate?
No honest tool can promise accurate absolute numbers, language models do not reliably reproduce human psychology. Windrose Atlas reports only what it can defend: which option is stronger and for whom, each answer carrying an evidence label that shows where it can be trusted.
How much does simulated market research cost?
Your first read on Windrose Atlas is free, with no card. Full pricing opens with sign-ups, soon. A traditional market research study typically costs between €5,000 and €15,000.
How is Windrose Atlas different from a survey or focus group?
A survey or focus group gathers a handful of real opinions over days or weeks. Windrose Atlas simulates 1,000 personas in minutes to tell you which conversations to have first, it narrows what to test, it does not replace real fieldwork.
Can AI predict what customers will do?
It cannot tell you exact outcomes, and any tool claiming a precise percentage is guessing. Windrose Atlas instead points you toward the stronger direction and the hypotheses most worth testing with real customers.
Who is Windrose Atlas for?
Founders, solo builders, and small direct-to-consumer teams weighing several ideas without a research budget, people who need a direction fast and want to know exactly how much to trust it.
What is Windrose Institute, and what is Windrose Atlas?
Windrose Institute is the studio; Atlas is one of its instruments — the one for simulated market research. Its sibling, Windrose Logbook, turns recorded therapy sessions into clinical notes entirely on one computer. All products follow the same institute principles: evidence labels, no fabricated precision, the machine proposes and a human decides.
Does Atlas share my data with other Windrose products?
No. Each instrument is separate: nothing you upload to Atlas reaches Logbook or any other product, your inputs are never training data, and deletion is a protocol, not a promise.
How is this different from just asking ChatGPT or Claude?
An assistant gives you one articulate, agreeable answer. Windrose runs a designed experiment: a counterbalanced factorial over 1,000 personas from a cognition model trained on real human decisions, resampled 100 times, blind-validated, and calibrated against real outcomes as they return. Assistants are trained to be rational and agreeable, which is exactly what customers aren't. The full argument is on our Why-no-percentages page.
Can my AI agent use Windrose Atlas?
Soon. Windrose Atlas will expose an MCP endpoint with a typed epistemic contract: your agent will call test_idea, receive a signed ranking with an evidence label, and return real outcomes via submit_real_outcome, which raises the calibration rung for its next question. Details on the For-agents page.