POPULATION INTELLIGENCE
Synthetic focus groups.
Built to be questioned.
An AI agent runs the study mid-task over the CrowdOS MCP server, on every plan, Free included. The panel is drawn from census data and disagrees with itself — not one model's opinion. See who leans in, who pushes back, and why.
APPS BUILT ON CROWDOS
Directional opinion, and a stimulus under pressure.
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.
PULSE
Ask the crowd. Get answers.
Ideal when you need fast, clean directional signal across a population.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.
HOW IT WORKS
How a study runs.
01
A persistent respondent population.
Archetypes across 24 markets read real news and evolve on a recurring cycle; every study draws a fresh, demographically-weighted panel from them.02
Three times a week, from the world.
Three times a week the population runs a cycle. Real news in, real reactions out — so it arrives at your study with the same week already processed.03
Write your question
Plain language. Yes/no, ranking, open opinion — the moderator agent reframes it into a structured research brief before fielding. The instrument is assembled by a moderator model with a structured research brief, not a freeform prompt.04
Review the findings
Sentiment distribution, demographic crosstabs, key arguments, dissenting minority, verbatim respondent excerpts. PDF and JSON export.
METHODOLOGY
An LLM persona is a costume for one prompt. Ours is a database row.
These methodological principles separate population intelligence from general-purpose LLM prompting — and every one is instrumented on every study.
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.
01
Census-grounded sampling.
Every respondent is drawn from published census distributions — US ACS, UK ONS, AU ABS — never convenience-sampled.02
OCEAN Big Five personality.
Five validated trait scores per respondent that materially change how each one reasons, argues, and responds.03
Behavioral context, not labels.
Routines, media diets, life events — the prompt carries a full persona, not a demographic label.04
Instrumented anti-sycophancy.
Every study instructs respondents to be critical, contradictory, and specific — against the LLM default to please.05
Multi-provider model diversity.
Panels rotate across an ensemble of frontier models, so no single model's bias becomes the panel's bias.06
A persistent respondent population.
Archetypes across 24 markets read real news and evolve on a recurring cycle; every study draws a fresh, demographically-weighted panel from them.
WHY YOU SHOULD BELIEVE THE NUMBER
Calibrated against real polling — and we publish the table.
We ran the crowd against 25 questions Pew Research actually asked Americans, applied per-topic calibration, and measured the gap. Anyone can claim accuracy; the per-question table is public so you can audit it. The latest run is the first Grade A in this benchmark's history.
- Pew parity
- 93.3%
- Mean error
- 3.36pp
- Rank accuracy (Pearson r)
- 0.984
See the question-by-question breakdown on the benchmarks page, or read the full methodology.
METHODOLOGY FAQ
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.
APPS BUILT ON CROWDOS
The same crowd, four ways in.
PULSE
Ask the crowd. Get answers.
Opinion measurement across census-grounded respondents. Sentiment distributions, demographic crosstabs, verbatim quotes. Vote, debate, or compare.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.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.MCP / API
From an agent, on every plan.
On every plan, Free included, an AI agent runs a study mid-task over the CrowdOS MCP server. Enterprise accounts also query the same crowd over REST, with keys the organization owns, returning structured JSON.
Positioning
Not a general-purpose chatbot.
A general LLM produces one voice optimized for helpfulness. Research audiences are nothing like that. The platform produces the distribution — skeptical, cynical, enthusiastic, confused, disengaged.
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.
The respondent population here is pre-assembled and pre-contextualized. Fielding is measured in seconds, not weeks.
METHODOLOGY FAQ
Audience research requires the distribution.
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 accurate is it?
We ran the crowd against 25 questions Pew Research actually asked Americans, applied per-topic calibration, and measured the gap. Anyone can claim accuracy; the per-question table is public so you can audit it. The latest run is the first Grade A in this benchmark's history.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.
Ask the crowd yourself.
Run a study in your browser, browse the live population, or connect your AI agent to the same crowd.