Core Concepts
Synthetic Consumers vs. Digital Twins, what a panel and a study are, and how to read results responsibly.
This page explains the ideas the rest of the guide builds on. It's worth ten minutes — understanding what a panel is changes how much you trust a report.
The building blocks
| Term | In plain language |
|---|---|
| Research Agent | The AI assistant you talk to. It finds panels, designs studies, and runs surveys on your behalf. |
| Thread | One conversation with the agent, kept in your history. |
| Panel | Your audience: a defined, versioned group of AI consumers you survey. |
| Study | A survey design: a research question, its questions, and any image stimuli, aimed at one panel. |
| Survey | The study while it's actually running against the panel. |
| Report | The result: findings, charts, and sourced quotes, produced automatically when a study finishes. |
| Credits | The unit consumed when you run a survey or generate a panel. |
Two kinds of consumers
Every panel is made of one kind of research unit — never a mix.
Synthetic Consumer
An artificial consumer generated by AI from statistical population data (modeled on real census proportions) and then given depth through a guide-driven interview. A Synthetic Consumer does not represent any single real person.
- Strengths: fast to generate, scalable, good for broad and representative questions, and for exploring reactions to concepts, wording, or designs.
- Keep in mind: answers are modeled, not measured from a specific human. Treat them as well-grounded simulations, not ground truth.
Digital Twin
A representation of a real person who was interviewed in depth beforehand. The twin answers by drawing on that person's standardized interview.
- Strengths: especially lifelike, grounded in an actual individual's views — useful when you want realistic, specific voices.
- Keep in mind: a twin speaks for the person behind it, so a Twin Panel is only as broad as the people who were interviewed.
Which to use? Reach for a Synthetic Panel for breadth and representativeness, and a Twin Panel when you want depth and realism from known individuals. The agent will suggest a fit — you decide.
Panels are versioned
A panel isn't a loose list — it's a defined audience frozen into a version. Each version has a fixed member list and a vintage (the date it was generated). This matters because a report always names the exact panel and version it used, so you can tell precisely who was "asked".
Standard vs. Team panels
- Standard panels are generally available building blocks you can pick from.
- Team panels belong to your workspace — created by your team, manageable by your team.
How to read results responsibly
The platform is deliberately transparent about how solid each statement is. Reports carry small trust markers:
- Derivation — a recommendation reasoned from the results, not a measured number.
- Thesis — a statement offered without a direct supporting quote. Treat with care.
- Exploratory — an early signal or trend, not yet confirmed.
Two habits will keep you honest:
- Check the panel first. A result is only as meaningful as the audience behind it — its type (Synthetic or Twin), size, and vintage.
- Separate measurement from interpretation. Charts and quotes are what the panel actually produced; derivations and theses are the agent's reasoning on top. Both are useful — they're just not the same thing.
You'll see exactly where these markers appear in the Reports section.