An AI-generated profile of a customer that you can question. The term covers two different things: a sketch that summarizes a whole group, and an individual with their own personality, memory and history. Candor builds the individual kind, and now calls them synthetic users or synthetic participants, because an individual is what you actually interview.
The phrase is used for two things that behave very differently in research, and most confusion about synthetic personas comes from mixing them up.
The group sketch. One persona stands for a whole slice of the market, such as “budget-conscious young parents.” You can question it, but it answers as the average of that group. Ask it twice and you get the same view. It is quick to make and useful for brainstorming, and it smooths away the disagreement that makes real research worth doing.
The individual. Each persona is one person with a name, a personality, a set of biases, a background and a memory of what they have said. Several individuals come from the same group, and they differ. One trusts the brand and one does not. One cares about price and one about time. You interview them one at a time, and the value is in where they disagree.
Candor builds the individual kind. That is why the product now calls them synthetic users or synthetic participants: the word persona made many readers expect a group summary.
There are two broad approaches, and the difference shows up in the answers.
From a prompt. You describe the customer to a language model and ask it to answer in character. This is fast and free, and the answers lean toward the model’s stereotype of the description. Personas built this way also tend to agree with whatever you suggest, because language models are trained to be helpful.
From evidence. The persona is assembled from research about the audience. In Candor, that means your uploaded research when you provide it, web research about the audience, and published population data for personality and demographics. Each detail in the profile is labelled Grounded, Derived or Assumed, so you can see what rests on evidence and what was filled in. The case for building from evidence sets out why this matters more than which model you use.
Because the models behind them are trained to be agreeable. Left alone, a synthetic persona tends to like your concept, accept your framing and warm up to whatever you suggest. A persona that agrees with everything tells you nothing.
Candor works against this in three ways. Personality and bias scores give some participants a reason to be skeptical. Each participant carries a record of their beliefs and of any change of position, so their views stay consistent from one answer to the next. And in live interviews a second model checks each answer against the participant’s profile before you see it. The essay on people-pleasing participants goes into detail.
The pillar guide, what synthetic user research is, covers the method end to end. Synthetic interviews are not synthetic polls explains why individuals give you more than a group average. For the survey industry’s version of the idea, see synthetic respondent.
Synthetic personas are used to question a simulated version of your audience before, or alongside, talking to real customers. Teams use them to find out which problems matter to a market, test how a concept or message lands, compare names and value propositions, and gauge reactions to a price. They are most useful early, when you need to narrow many options down to the few worth taking to real people.
A marketing persona is a written document: a named composite with a photo, goals and frustrations, meant to remind a team who it is building for. You cannot ask it anything. A synthetic persona can answer questions. The individual kind goes further, with its own personality scores, biases and memory, so two personas from the same group can disagree with each other in an interview.
When a tool builds individuals you interview one by one, yes: synthetic persona, synthetic user, synthetic participant and synthetic respondent all describe the same thing. Candor used to say personas and moved to participants, because many readers hear persona and picture a group summary. Your search may say personas, and the idea is the same.
The quick way is to prompt a chatbot with a description and ask it to answer in character. That works for a demo, but the answers drift toward what the model assumes about the description. The more reliable way builds each persona from evidence: research about the audience, published personality data and population figures, with each detail labelled by where it came from. Candor does the second, starting from a description of your audience and any research you upload.
Start with a free project, or walk through Candor with us first.