One AI-generated individual from a target audience, built from real evidence rather than imagination. A synthetic user has personality traits, cognitive biases, memory, and behavioral patterns drawn from market data and research, and you interview them one at a time the way you would a real person.
A useful synthetic user is more than a chatbot told to play a role. It has a profile, and the profile shapes every answer. In Candor that profile has four parts.
Personality. Scores on the five OCEAN traits (openness, conscientiousness, extraversion, agreeableness and neuroticism), drawn from published population research for the audience’s region and occupation. A low-agreeableness user pushes back harder in an interview. The OCEAN model essay explains how the scores change answers.
Cognitive biases. Tendencies such as loss aversion or status quo bias, each with a strength from 0 to 100, set from the ranges for the user’s group.
Memory. A background, habits and constraints, beliefs and preferences, a way of talking, and a handful of past decisions. Within a study, a synthetic user also remembers what they said in earlier interviews, so you are talking to the same person each time. The participant memory essay covers how that works.
Labels on the details. Each detail in the profile says where it came from: Grounded (something you can check sits behind it), Derived (it follows from the evidence without a source stating it) or Assumed (nothing pointed to it, and Candor says so).
Three places. Research you upload, such as interview transcripts, survey exports or support tickets, comes first when you provide it. Uploading is optional. Web research about your audience fills in the rest, and published population data sets the spread of personality and demographics so uncommon profiles appear about as often as they do in real life.
This is the difference between a synthetic user and a prompt. Ask a model to “be a nurse” and you get its average idea of a nurse. Build the nurse from evidence about nurses and you get answers you can trace. The evidence essay walks through the full pipeline, and how Candor works shows each stage.
The three get confused because they share a job: helping a team understand a customer. They do it in different ways.
| Persona document | Synthetic user | Real participant | |
|---|---|---|---|
| What it is | A written sketch of a typical customer | An AI-generated individual you can interview | A recruited person |
| Can you ask it a question? | No | Yes, as many as you like | Yes, within the session you booked |
| What it is built from | The team's research and judgement | Evidence about the audience, published personality data and population figures | Their own life |
| What its answers are | Nothing to answer with | An estimate of how a type of customer reasons | A fact about that one person |
They are strongest early, when you have more questions and options than budget. Teams use them to find which problems matter to a market (problem discovery), check whether a problem they believe in holds up (problem validation), and compare concepts, names, messages and prices described in words (concept testing).
They fall short in three places. They cannot tell you what one named person will do. They do not react to mockups, prototypes or images. And they do not replace real customers for a decision you will be held to. Treat a synthetic finding as a lead to confirm. What synthetic research skeptics get right is an honest account of the limits, and how to spot convincing nonsense shows how to catch a fluent answer with nothing behind it.
For the method as a whole, read what synthetic user research is. Related terms: a synthetic persona, a synthetic respondent and a digital twin, which is not the same thing.
Synthetic users are AI-generated people who stand in for members of a target audience during research. Each one has a personality, cognitive biases, a background and opinions, built from evidence about the audience rather than invented from a single prompt. You interview them the way you would interview a real customer, and their answers are shaped by that profile. Teams use them to explore a market, test concepts in words and pressure-test assumptions before spending money on real recruiting.
In UX and product research, synthetic users means simulated research participants: AI-generated people you question in place of, or before, real users. The term covers a wide range of quality. At one end is a chatbot told to act like a 35-year-old nurse. At the other is a participant assembled from evidence, with personality scores, biases and memory that shape every answer. When you evaluate a tool, ask what each synthetic user is built from and whether you can check it.
They are accurate enough to tell you how a type of customer tends to reason, and not accurate enough to replace real people for a final decision. Accuracy depends on the evidence behind them. A synthetic user built from real research about your audience gives more useful answers than one built from a prompt. Candor labels each profile detail as Grounded, Derived or Assumed so you can see which parts rest on evidence. Confirm the findings that would change your plan with real customers.
Often, but not always. Some tools use synthetic persona for an individual you can interview, which is the same thing as a synthetic user. Others use it for a summary sketch of a whole group, closer to a traditional marketing persona. Candor builds individuals, and calls them synthetic users or synthetic participants to make that clear.
Not in Candor today. Candor's synthetic users react to things described in words: a concept, a name, a message, a price, a problem statement. They do not look at mockups, click through prototypes or react to images. For usability and visual design questions, test with real people on a usability platform.
Start with a free project, or walk through Candor with us first.