Millions of simulated people online

Ask millions of people first.

Millions of simulated people you can interview — and watch as they operate a scanned twin of your product, then explain why they stay or churn.

link · idea · prd · question
no demo call · first study free · results in ~45 min
Why it's different

A panel that actually uses your product.

Most synthetic-user tools stop at conversation. Crowdproof agents get reached by your launch, click through a scanned twin of the real thing, churn — and then explain themselves.

Interview them

simulated
MR
Miguel ReyesSkeptics
41 · Toronto ON · UX lead · agent #112,406

“The pricing page buried the per-seat math. I closed the tab there.”

What would change your mind?
“A calculator with my headcount, and the overage policy in plain words.”

stratified interviews · transcripts attached in full

Watch them use it

twin-operated
twin capture · /pricing · 00:42 · clicked ‘Start free’

Replay any agent’s session on the scanned twin — every click, every hesitation, the exact moment they left.

128 replays per study · click paths + action logs
study typesLaunch simulation — flagshipMessage testPricing studyMarket scanConcept screen
How it works

One question in. One sim-week out.

Every study runs the same instrument — the example here is the flagship: a full launch simulation, 45 minutes of real time.

01

Ask

Type a question or paste a link, an idea, a PRD. Any live site — your product, a store, a competitor — gets scanned into an operable twin; the society gets briefed.

02

Simulate

Your question runs against 1,000,000 census-seeded agents. Launch studies diffuse through six simulated social channels.

03

Watch & interview

Agents operate the twin — click, stall, churn — then sit for stratified interviews about what they did and why.

04

Decide

You get a funnel with uncertainty ranges, churn reasons ranked by segment, and every transcript attached.

Run #213 · pricing page v2

Simulating
Scan product into twin
14 pages · approved by you
Seed society
1,000,000 agents · ACS-2024
Simulate diffusion
72 sim-hours · 6 channels
Twin walkthroughs
128 sessions replayable
Interview cohorts
stratified · n = 128
Build funnel report
P10–P90 across 9 seeds

seed 42 · ACS-2024 · SIR-social v2.3
CI: bootstrap, 10K resamples

The output

A funnel, not a vibe.

Reach → engaged → tried → retained: a count of simulated people at every stage, the drops between stages with the reasons ranked — and the uncertainty drawn on, never hidden.

run #213 · pricing page v2 · n = 1,000,000 · 72 sim-hoursP10–P90 across 9 seeds
h0h24h48h72−156K · −38%never engaged — one mention, no second touch−116K · −45%skeptics exit at the per-seat pricing table−66K · −47%no clear win inside the first sessionREACH412K± 18KENGAGED256K± 14KTRIED140K± 11KRETAINED74K± 9K · 18.0% of reach
reach
412K± 18K
−156K · −38%never engaged — one mention, no second touch
engaged
256K± 14K
−116K · −45%skeptics exit at the per-seat pricing table
tried
140K± 11K
−66K · −47%no clear win inside the first session
retained
74K± 9K
h0 → h72 · retained = 18.0% of reach
simulated · bootstrap CIs, 10K resamples · churn reasons from 128 attached interviews
Methodology & honesty

Simulated evidence, labeled as such.

Every artifact carries its provenance — simulated for interviews and funnels, twin-operated for walkthroughs. Nothing here pretends to be a human.

society1M agents seeded from US census (ACS-2024) with occupation, income, and media diet. Cohort composition printed on every report.
twinsAgents operate a scanned, versioned twin of your real product — not a description of it. You approve the twin before any run.
interviewsStratified across segments; transcripts attached in full, never summarized away.
seeds & rangesEvery study executes across multiple seeds. We report P10–P90 ranges, not point estimates. Wide ranges mean 'we don't know yet' — and we say so.
Read the methodology →methodology v2.3 · changelog public
retained % · 9 seeds · run #213
10%18%26%

reported: 18.0% (P10 14.2 – P90 22.4)

Pricing

Meter by depth, not by seats.

A credit buys simulation depth, not seats at a demo. A shallow reach check costs 1 credit; a full study — diffusion, twin walkthroughs, interviews, funnel — costs 10. Your first study is free.

reach map · 1→+ engagement · 2→+ twin walkthroughs · 5→+ interviews & funnel · 10

Traditional research

$20k–$60k/study
4–8 weeks per study

The manual route — agency or DIY: recruiting, panels, moderation, synthesis.

·Audience survey $6k–$15k
·Eight 1:1 interviews $4k–$12k
·Any iteration starts the clock again
for comparison only
Most common

Pro

$29/mo
60 credits / mo

For one researcher or PM validating a roadmap.

·First study free
·Twin walkthrough replays
·Run history & diffs
Start Pro

Max

$99/mo
300 credits / mo

For continuous, deeper testing.

·Custom cohorts & panels
·More seeds per run
·API access
Start Max

Team

$79/seat/mo
200 credits / seat · pooled

A shared workspace for product orgs.

·Shared twins & panels
·Reviewer seats free
·SSO & audit log
Talk to us

traditional range = typical published U.S. agency & panel rates, shown for comparison only · every plan reports P10–P90 ranges and full methodology — honesty is not a paid feature

FAQ

Questions, answered plainly.

Directionally, at funnel scale. We backtest against launches and campaigns with known outcomes and publish the calibration report quarterly. The claim is never person-level prediction — it's that the funnel shape, the leak points, and the segment rankings hold up. Where they're uncertain, the ranges say so.

A scanned, interactive replica of any live site — your own product, a storefront, even a competitor's — that agents can actually operate: pages, flows, pricing tables, empty states. We build it from whatever you submit, and you approve the twin before any agent touches it.

No. Every interview is simulated and labeled as such, with the agent's ID and model version printed on the transcript. Nothing in Crowdproof pretends to be a human.

A reach check returns in minutes. A full study — diffusion, twin walkthroughs, interviews, funnel — takes about 45 minutes at the default depth.

Yes. Twins, PRDs, and questions run in an isolated environment, are never used to train models, and can be deleted — along with every derived artifact — at any time.

Then the honest answer is 'not enough signal', and the report says exactly that — along with what would narrow it: more seeds, a deeper pipeline stage, or a tighter cohort. We'd rather you wait than decide on noise.