Questions, answered.

Sibyl simulates an entire synthetic society to pressure-test a decision before you commit to it — a product, an investment, a technology bet, a campaign, a message, a strategy call. You get a verdict in under a minute: whether it's likely to work, why or why not, and how to improve it. It doesn't replace deep research; it removes the guesswork before you invest in it.

Anything you'd otherwise have to guess at: a product, an investment, a technology-adoption call, a business decision, a campaign, a message, a document, or a URL. “What should we invest in first?” “Should we adopt this now or wait?” “Which option has the best odds?” Validate a single idea, or rank several against each other — each gets its own full simulation.

A single chatbot gives you one confident opinion. Sibyl simulates an entire society — up to 500 people with different roles, beliefs, and social, economic, and political backgrounds — and reports not just the average but where they disagree and why. Seeded and methodical: the population, the reasoning, and the verdict are all persisted, so you can see exactly how the read was reached.

Yes, and we try to be honest about the limits too. Argyle et al. (Political Analysis, 2023) found LLMs conditioned on real demographic backstories reproduce nuanced, subgroup-specific opinion patterns — they call it “algorithmic fidelity.” A 2024 Stanford follow-up built agents from real interviews with over 1,000 people and found the agents reproduced their own survey answers about as consistently as the people did when asked again two weeks later. It's a genuinely contested field: other researchers in the same journal as Argyle et al. (Bisbee et al., 2024) found simulated respondents can under-represent real variation, and critics (Lin, 2024) warn against treating any single model's output as a stand-in for real people. That's exactly why every Sibyl verdict is labeled an estimate, not a guarantee, and why we're building an open calibration program against real outcomes.

We build a synthetic society whose people are given distinct roles, beliefs, and social, economic, and political backgrounds, then model how each of them reacts to your idea and aggregate it into one verdict — a rating, the spread of opinion, the reasons, and concrete improvements.

Sibyl gives you the central tendency of a diverse, simulated society — a fast, structured read, not a guarantee. The on-screen disagreement band reflects how much the society disagrees, not a statistical confidence interval. Verdicts are simulation estimates, not professional advice. We're building an open calibration program and will publish results as they mature.

Under a minute for a typical idea.

Inputs are redacted for PII before they enter the simulation, URLs are fetched behind an SSRF guard, and runs are isolated per account.

You're charged the real compute cost of each run, typically $2–$5, not a flat rate. Subscribe from $79/month for a usage balance (annual plans save 20%), or pay as you go — top up in advance, or save a card once and let your balance auto-refill as you spend. See the pricing page for details.

Yes — pay a flat $9 platform fee per run and bring your own model API key; you're billed for the model directly by your provider.

Sibyl is in private beta. We're onboarding a small first group soon.

By policy Sibyl does not cover regulated or high-stakes domains (politics, health, lending, hiring, legal, and similar).