Test a policy on many lives at once

Spin up a panel of synthetic personas: different generations, incomes, genders, nationalities, cultural backgrounds, places. They debate your policy in three rounds. Then real people with those profiles tell you where the AI spoke for them, and where it got them wrong.

Request a panel Read an example debate

Synthetic voiceReal person

How it works

From a brief to the gap between AI and reality

BriefYou describe the policy, the place and the horizon year, and choose who sits on the panel.
Synthetic debateEach seat becomes one plausible person. They give openings, answer each other across their differences, and close with the condition that would win their support.
Real reactionsYou share a link. People pick the seat closest to them, rate how well it spoke for them and give their own stance. No account needed.
The gapYour results link compares synthetic and real stances seat by seat, and collects what the AI missed.

Who sits on the panel

Pick a panel type, or design your own

Each type varies one thing that shapes how a policy lands. The seats below are examples; we adapt them to your country and question.

Loading the panel…

Request a synthetic panel

We run the panel for you and email you two links: one to share with participants, one for your results.

Panel type

Number of seats

Add an imagined resident of the horizon year Someone speaks for a generation that can't vote yet, as in Future Design.

Policy title

What the policy does What changes, when, who pays and who benefits. The panel debates exactly this text.

Place

Horizon year

Debate language

Anything else we should know? Optional. Who you plan to invite to react, a deadline, a seat you must include.

Your name

Organisation Optional

Store my request and contact me about it, as described in the privacy notice.

What happens next

We read your brief and run the synthetic debate. If anything is unclear, we email you first.
You receive a participant link and a private results link.
You share the participant link with real people who match the seats.
Your results link shows, seat by seat, where the synthetic voices matched them and what they missed.

Good to know

Unlike the wind-tunnel, this service stores data on our server: your request, and the anonymous reactions of participants.
Participants never give their name. They are not asked for their gender, ethnicity or nationality; they only choose the seat that feels closest, or none.
The debate is written by Claude, an AI model, and is labelled as such everywhere it appears.

Questions first? Email connectwithus@humanitysolutions.net.

Use it well

What a synthetic panel can and can't tell you

Good for finding questions

Before you spend weeks on engagement, a panel shows which groups a policy may hurt, which conditions could win them over, and where the fault lines run.

Not a sample

Each persona is one plausible person, not the average of a group. A synthetic panel never replaces talking to the people affected; it helps you prepare for it.

Stereotypes are findings

When real people say "that's not us", that gap is the result. It shows where assumptions, the AI's and possibly your own, need checking.