In research, an AI agent built to answer the way one specific real person would, made from that person's own data, usually a long interview plus a record of their choices. A synthetic participant is different: it is built from evidence about a type of customer and matches no real person.
A real person agrees to be twinned, then gives the data the twin learns from. That is usually a long recorded interview about their life, views and habits, sometimes with survey answers or a record of their purchases and behavior. The twin is then asked new questions and answers as that person would be expected to.
Several research companies build twins this way. Some keep a panel of consenting people who are paid to be interviewed and twinned. Others let a customer build twins from its own customer data, which puts part of the consent question on the customer.
Both answer research questions in place of a real person, which is why they get lumped together. The difference is where each one comes from.
| Digital twin | Synthetic participant | |
|---|---|---|
| Who it represents | One specific, real person | A type of customer, matching no real person |
| What it is built from | That person's own interview and behavioral record | Evidence about the audience, published personality data and population figures |
| Needs a person's consent to exist | Yes | No |
| Best at | Predicting what that person will say | Showing how a kind of customer reasons, and why |
Twins win at one job. In the study behind one twin company, agents built from two-hour interviews predicted those same people’s survey answers about nine points better than agents given demographics alone. If your question is what a named person will say, a twin of that person is the right tool.
A twin exists because a real person’s data went into it, and that brings questions a research or legal team will want answered. Who agreed to be twinned, to what, and when? If someone withdraws, does their twin disappear everywhere it was used? Who can see the record behind each twin? Could the results be used to target the real people behind them?
None of these is unanswerable, and the better twin companies publish their answers. They still add to a vendor review, which matters most in healthcare, insurance and financial services. See Candor for healthcare CX teams for how that plays out in a regulated setting.
Candor builds every participant from evidence about a type of customer, never from one person. The synthetic users in a Candor study get invented names and match nobody, so there is no consent chain to maintain and no personal record behind any of them. The trade is that Candor cannot predict what a named individual will say, and does not claim to. Why we don’t build digital twins of real people explains the choice and its cost in full.
For the method Candor uses instead, read what synthetic user research is and how evidence grounding works. For the related terms, see synthetic persona and synthetic respondent. For a side-by-side with one company that builds twins, see Candor vs Brox.
A digital twin of a customer is an AI agent trained to answer the way one specific person would. It is usually built from a long interview with that person plus a record of their choices and behavior, and the person has to consent to it. You can then ask the twin questions and get an estimate of what the real person would say. It is a copy of an individual, not a composite.
No, though the two are often lumped together. A digital twin copies one real person. A synthetic user, in the sense Candor uses, is built from evidence about a type of customer and matches nobody. Twins are better at predicting what a named individual will say. Synthetic users are built to show how a kind of customer reasons, without needing any one person's data or permission.
No. Every Candor participant is built from evidence about a type of customer: research you upload, web research about your audience, and published population data. There is no feature for building a participant from a named person or a profile. That means no consent chain to maintain and no personal behavioral record behind any participant.
In engineering and manufacturing, a digital twin is a virtual model of a physical system, such as a jet engine or a factory line, kept in sync with sensor data so engineers can test changes safely. Research borrowed the name for AI copies of people. The two share the idea of a model standing in for the real thing, and little else.
Start with a free project, or walk through Candor with us first.