Synthetic personas, explained

A synthetic persona is an AI-generated profile of a hypothetical person — their demographics, situation, attitudes and behaviour — used to simulate how part of a market might react to a question, a product idea or a price. Unlike the two or three archetypes pinned to a wall after a research project, synthetic personas are generated in bulk, so you can ask a question to three hundred of them and read the distribution of answers rather than a single story.

How a synthetic population is built

1. 1. Define the audience. Describe who you want to hear from in plain language — a market, a region (a single country or global), and any refinements such as income band, life stage or category usage.
2. 2. Sample the attributes. Each persona’s demographics are drawn from distributions, not invented freehand, so the population as a whole has a realistic shape instead of three hundred near-identical people.
3. 3. Generate the persona. The sampled attributes become fixed facts for that individual; the model fills in context, motivations and constraints around them.
4. 4. Keep the population persistent. The same people answer every question in the study and any follow-up, so their reasoning stays consistent across the whole project.

Grounding: the part that decides quality

Left unsupervised, language models are agreeable and repetitive: they praise whatever you describe and converge on the same phrasing. Two corrections matter most.

• Anchor to real data. Where a published benchmark exists for a question, draw each respondent’s outcome against the real percentage first and let the model explain that outcome — rather than inventing the number.
• Force distinct reasoning. Each answer should be justified from that persona’s own specific attributes, separately from its sentiment, which suppresses both flattery and copy-paste answers.

Where synthetic personas are genuinely useful

• Pressure-testing positioning or naming before you commit to copy.
• Early price sensitivity reads on a concept that does not exist yet.
• Writing better questions before you pay to field a real survey.
• Comparing how an idea lands across several markets in one pass.
• Finding objections early, then following up with the same respondents.

The limits — read this before you rely on it

Synthetic respondents are simulations. They reflect patterns in training data and in the distributions used to build them, not lived experience. Qualitative themes tend to hold up better than precise numeric scores, and repeat runs will vary. Use them to shape decisions upstream and to decide what deserves real fieldwork — never as evidence for a public claim, a regulated statement or a safety-critical call.

Common questions

What is a synthetic persona?

A synthetic persona is an AI-generated profile of a hypothetical person — demographics, context, attitudes and behaviour — used to simulate how a segment of a market might respond to a question, a concept or a price. It is not a real person and holds no real individual's data.

How are synthetic personas different from traditional personas?

Traditional personas are a handful of hand-written archetypes summarising past research. Synthetic personas are generated at population scale — hundreds at a time — each with its own attributes, so you get a distribution of answers rather than a single illustrative profile.

Are synthetic personas accurate?

They are directional. Accuracy depends on how the population is constructed and whether responses are grounded in real distributions. Treat qualitative themes as the stronger signal and numeric scores as indicative, then confirm anything decision-critical with real fieldwork.

When should you not use synthetic personas?

Avoid them for regulated claims, published statistics, safety-critical decisions, or anything where you need evidence from real people. Use them upstream: before you spend money on recruiting a real sample.