Conversation Simulation
Structured group discussions that change minds at the rate real discussions do.
Run structured group discussions with 10-500 AI participants. Chats can run simply for fast iteration, or in orchestrated mode for richer, more realistic conversational dynamics.
The orchestrated engine is built around one principle: a simulated discussion should change minds at the rate real discussions do. We tune and check it against large libraries of real recorded discussions and public debates with known outcomes. What that buys you:
- A private inner voice. Every participant thinks for itself before it speaks, and stays itself for the whole session. Operators can open the private trace and see the reasoning behind any turn.
- Human memory. Participants remember like people, and forget like people. That balance is tuned, and it matters more than you would guess.
- Mind changes you can trust. Panels of raw AI either agree in minutes or never move at all. Ours change their minds at realistic rates, and the right people do the changing.
- Evidence that behaves like evidence. Participants look things up when it matters to them, and what they find carries the weight it should. A document shown in the room counts for more than a claim about one.
- Private ballots or public votes. Run a jury with a show of hands, or a survey panel with secret ballots. What people say in the room and what they privately submit can differ.
- A room card. Post standing instructions everyone can see, and they stay in front of every participant for the whole session.
Example input: Topic: “inclusionary zoning requirements” · Participants: 30 · Duration: 90 min
Example conversation excerpt:
Marcus Webb We tried density bonuses in 2019. Developers took the bonus, skipped the affordable units. Enforcement is the whole game.
Dr. Priya Nair The Minneapolis data shows a 14% rent reduction in upzoned corridors within 3 years. The supply effect is real; the question is timeline.
Moderator We’re 20 minutes in. Rafael and Yolanda have been quiet, so I’m bringing them in on the enforcement question…
Real-time data integration: Participants can fetch and cite current statistics via the Kagi API (recent policy changes, economic data, regional statistics), scoped automatically to the discussion topic.
Project materials: Attach participant-facing materials (briefs, case facts, product specs, discussion guides, source packets) and optionally allow web research, so simulations draw from the same context your real audience would.
Next: Voting and results →
Request Early Access
We're onboarding a first cohort: academic and applied researchers, market research firms, enterprise AI and product teams, law firms, and policy organizations.
Early access participants receive preferred rates and priority onboarding.