A living model of humanity.Built to be questioned.
Simulate how people react to a policy, a product or a public message.
Census-grounded simulated respondents that keep up with the news between your questions. See who leans in, who pushes back, and why.
- Pew parity
- 93.3%
- markets
- 24
- person population model
- 8B
APPS BUILT ON CROWDOS
PULSE
Ask the crowd. Get answers.
Opinion measurement across 20–1,000 census-grounded respondents. Sentiment distributions, demographic crosstabs, verbatim quotes. Vote, debate, or compare. Median fielding time under two minutes.
Open Pulse→
SCREENING ROOM
Pressure-test any stimulus before you ship.
Upload a creative, product mockup, policy draft, pitch deck, or document. A representative panel reacts and explains why — with cohort breakdowns and expert lenses.
Open Screening Room→
ADS LABO
Predict ad performance before you spend.
Upload a creative and set Meta-style targeting. A synthetic panel predicts click intent, brand recall and sentiment, with cohort breakdowns by age, gender, income, and interest.
Open Ads Labo→
SEE IT ANSWER · BROWSER, AGENT OR API — SAME CROWD
Flip through real example studies below — then run your own in the browser, from your AI agent, or, on an Enterprise account, from code in three lines.
EXAMPLE STUDY · 200 RESPONDENTS · US GENERAL POPULATION · 8.4s
“Should we ship a 4-day workweek?”
“Cutting a day would force me to actually plan. I’d trade Friday for that.”
“Four days of my clients’ problems don’t fit in four days. Someone still answers the phone on Friday.”
THE REQUEST · ENTERPRISE API · POST /api/v1/developer/simulations
curl https://api.crowdos.ai/api/v1/developer/simulations \
-H "Authorization: Bearer $CROWDOS_KEY" \
-H "Content-Type: application/json" \
-d '{
"topic": "Should we ship a 4-day workweek?",
"mode": "voting",
"population_size": 200,
"demographic_preset": "us_general_population"
}'METHODOLOGY
These methodological principles separate population intelligence from general-purpose LLM prompting — and every one is instrumented on every study.
I
Census-grounded sampling.
Every respondent is drawn from published census distributions — US ACS, UK ONS, AU ABS — never convenience-sampled.II
OCEAN Big Five personality.
Five validated trait scores per respondent that materially change how each one reasons, argues, and responds.III
Behavioral context, not labels.
Routines, media diets, life events — the prompt carries a full persona, not a demographic label.IV
Instrumented anti-sycophancy.
Every study instructs respondents to be critical, contradictory, and specific — against the LLM default to please.V
Multi-provider model diversity.
Panels rotate across an ensemble of frontier models, so no single model's bias becomes the panel's bias.VI
A persistent respondent population.
4,000+ archetypes across 24 markets read real news and evolve on a recurring cycle; every study draws a fresh, demographically-weighted panel from them.
ACCOUNT PLANS
Studies are metered against a prepaid wallet at the same per-respondent rate on every plan. Plans add monthly credit, unlock larger panels, and raise MCP limits. Every app is available on every plan.
Free
$0
pay-as-you-go
- ·$5 welcome credit on signup
- ·Pulse panels up to 20 respondents
- ·All Screening Room tiers available
- ·Same per-respondent pricing as every plan
Pro
$25
per month
- ·$25 monthly wallet credit
- ·Pulse panels up to 500 respondents
- ·Pricing and ranking studies from your AI agent
- ·Higher MCP rate limits
Max
$100
per month
- ·$100 monthly wallet credit
- ·Pulse panels up to 1,000 respondents
- ·The highest MCP rate limits
- ·Invoicing and custom audience segments
METHODOLOGY FAQ
How does this differ from prompting a general LLM?
Each respondent is a persistent persona sampled from census distributions — OCEAN traits, occupation, life stage, media diet — with a demographic identity that survives across studies. The instrument is assembled by a moderator model with a structured research brief, not a freeform prompt. Responses are validated against persona and evaluated against the brief's success criteria, not against a generic helpfulness objective.
How does synthetic respondent data compare to human panels?
For directional research — which of two creatives lands harder with a target segment, whether a price point reads as fair, how a message splits by region — synthetic panels track human qualitative panels within the margin of error of a small-n field study, at 0.01–0.1% of the cost. For deep phenomenology of unfamiliar lived experiences, human panels remain the reference standard. The platform flags when a study falls into the latter category.
What is the data-handling posture for uploaded creative?
Creative files are stored in an isolated bucket with row-level security and deleted automatically about 24 hours after upload. Uploaded material is not used for model training. Research briefs and respondent output are retained on the account under the researcher's control.
Can audience definitions be customized?
Yes. The crowd spans 24 markets across 6 continents, with starter templates for common cohorts (US Gen Z, UK metro, Singapore tech workers, Brazilian urban middle-class, etc.) and a custom-audience builder for any country, age range, gender, income tier, education level, and political leaning. Available across all apps and via the API.
Is there an API?
Yes, for Enterprise accounts: the same crowd that powers Pulse, Screening Room, and Ads Labo is queryable over REST, with keys the organization owns, returning structured JSON. On every plan you can also run studies from an AI agent such as Claude or Cursor through the CrowdOS MCP server, at your plan's limits. See the Enterprise and MCP pages for details.
Query the crowd today.
Sign up free, no card. Your welcome credit covers your first studies — or connect your AI agent and run studies from there.
Get started free→