Synthetic audiences, grounded in a real market
A synthetic audience is a set of AI participants configured to reason like a demographic segment. Most of them are generic — a persona with an age and a job title. Ours are built on the demographics, retail habits, benchmarks and everyday register of one specific market, and the current economic and social climate of that country is injected before every session. First question is free, no sign-up.
What a synthetic audience actually is
A synthetic audience is a panel of AI-generated participants, each configured with an age, a city, an income band, an education level, a household situation and a way of speaking, that responds to the same stimulus the way a recruited panel would. You ask a question, they answer one by one, they read what the others said, and they disagree when a real person would.
The value is not that it is cheap. The value is that the loop closes in two minutes instead of three weeks, so you can afford to be wrong ten times before you spend money on being right once.
Why grounding is the whole difference
Most synthetic audience tools give you a persona sheet: 34, marketing manager, urban, values sustainability. That produces answers that could come from anywhere, because they do — they come from the model's average of everywhere.
A profile only becomes useful when it reasons with a specific cost of living, a specific set of shops, and a specific argument that is going on in that country this month. A Spanish participant weighs a price against Madrid rent and the supermarket basket, and names Mercadona, Wallapop and El Corte Inglés without thinking. A Mexican one reasons in pesos, months-without-interest instalments, Oxxo and Mercado Libre. A Chilean one talks about arriendo, the dividend in UF, Líder and Falabella.
None of that survives translation, which is exactly why a synthetic panel built for the US and asked to 'act Spanish' produces a tourist.
What you can do with one
It replaces the round of informal conversations you would have before commissioning fieldwork, not the fieldwork.
- Screen ten concepts, claims or names and keep the three that survive.
- Pilot a discussion guide before you pay for recruitment, and cut the questions that separate nobody.
- Find the vocabulary a category is actually described with, in the local language, and use it in your copy.
- Test a price and see what it gets compared against — which is rarely what you assumed.
- Compare two segments on the same stimulus without doubling the budget.
- Sense-check a market you are considering entering, before you commit a budget line to it.
What it is not
This is a language model interpreting an archetype. It is not a sample. It has no margin of error, it is not projectable to a population, it does not forecast demand or willingness to pay, and it is not evidence for a committee that requires human fieldwork.
When the decision is large, when the product is medical, financial or safety-critical, or when you need a real person to hold the packaging, commission research with humans. The correct use of this is before that investment, so you arrive at it with fewer hypotheses and better questions.
Pricing
The first question is free and needs no card and no account: one every 48 hours. Focus Sprint is €7.99 once for 60 synthetic responses — 10 questions to a panel of six — valid for 30 days, with the custom participant builder, memory across sessions and full export. Studio Pro is €59.99 a month for 600 responses — 100 questions to a panel of six — across up to 10 focus groups, with two parallel rooms, the time machine and market evidence with cited sources.
Focus Sprint is a single purchase, not a trial that turns into a subscription. Buying again tops up whatever is left and restarts the 30 days.
Recruited panel, generic synthetic panel, and this
| Recruited panel | Generic synthetic | Grounded synthetic | |
|---|---|---|---|
| Time to first answer | 2–4 weeks | Minutes | Under 2 minutes |
| Cost per session | Thousands | Subscription | From €7.99 |
| Local cost-of-living reasoning | Yes | No | Yes |
| Local retail and media references | Yes | Rarely | Yes |
| Speaks the local register | Yes | Translated | Native |
| Projectable to a population | No | No | No |
| Counts as formal evidence | Yes | No | No |
Frequently asked questions
Nobody can honestly give you a number, and anyone who does is selling. What we can say is what the output is for: directional signal about how a segment might react, useful for discarding weak ideas quickly and for sharpening the questions you will later ask real people. Treat every finding as a hypothesis with a validation plan attached.
Keep reading
Last reviewed: 2026-08-21. Participants are AI-generated synthetic archetypes, not real people.