FlashInsight

Glossary

Synthetic persona

Also called AI persona · persona brief

The briefing given to the model: the traits, attitudes and characteristics that steer its answers. A specification, not a person.

A synthetic persona is a specification: the set of traits, attitudes and characteristics handed to a model so that its answers come from somewhere. It does not answer — it frames the answer. The synthetic respondent is what speaks once that frame is in place.

The distinction pays for itself immediately. Two personas can be equally vivid, coherent and plausible, and be wrong in exactly the place where the decision turns — because the quality of a briefing is not judged by how well it reads, but by what it reproduces of an observed population.

The common usage, and why we do not follow it

In everyday market-research talk, “synthetic persona” usually means the respondent itself, and sometimes any AI-assisted study whatsoever. We hold the narrower sense because it carries the only question a buyer needs answered: where did this profile come from? A persona written from a client brief does not have the same standing as a profile drawn from a population calibrated on public statistics. Both are legitimate. They do not compare.

What a buyer can check

The provenance of the profile: free-text brief, or draw from a calibrated population? The audit trail: can you get from an answer back to the profile that produced it? And the spread: when twenty personas are interviewed, do you get twenty points of view, or twenty phrasings of one?

Frequently asked questions

What is a synthetic persona?

It is the briefing given to a language model: the traits, attitudes and socio-demographic characteristics that frame its answers. It does not answer; it steers the answer. What speaks afterwards is the synthetic respondent. Well built, a persona is drawn from a population calibrated on public data; badly built, it is a portrait written by hand.

Is a synthetic persona a real person?

No. It is a specification: a plausible profile described by its characteristics and attitudes, matching no one in particular. Its value lies not in how lifelike it reads but in what it reproduces of an observed population.

What is the difference between a synthetic persona and a synthetic respondent?

The persona is the input, the specification handed to the model. The respondent is the output, what the model says once that frame is in place. You check different things on each: where the profile came from for the persona, how far the answers spread for the respondent.

How do you know synthetic personas are genuinely diverse?

By measuring the spread of their answers against real respondents with the same profile, on the same question, using a public survey. If the simulated answers are clearly tighter, the personas resemble each other too much, however vivid they are. That is under-dispersion, and it is measured market by market.

See also

  • Synthetic respondent — An instance of a model asked to answer questions as a human would, calibrated on data describing a real population. The output — the unit a sample is made of.
  • Synthetic population — The generated set taken as a whole, built to match the known statistical margins of a territory. The term comes from microsimulation, not from research.
  • Concept test — Putting an idea that still lives on paper — a concept, a product, a name, a promise — in front of its audience, before anything is committed.

Further reading

Back to the glossary