Synthetic research

Synthetic research is market research in which the answers come from simulated respondents, built from a real population, instead of being collected from a human panel. It uses the usual tools, surveys, interviews, focus groups, and serves to explore and compare options before fieldwork.

The word covers very different practices, from asking a chatbot to role-play a customer to running a calibrated survey on thousands of profiles. What separates them is not the model. It is where the respondents come from, and whether their answers are checked.

The essentials

• Synthetic research simulates answers; it does not measure an opinion.
• Everything starts with the population: profiles anchored in real data, not invented by the model.
• Answers are checked before they are read: spread, bias, stability.
• It is read on the gaps between options, and confirmed in the field when the decision is a large commitment.

How does synthetic research work?

1. Start from a decision. What has to be decided, and between which options: two concepts, three names, five taglines.
2. Build the population.Profiles anchored in the market’s official statistics, the Census in the United States, INSEE in France, Eurostat in Europe, then enriched with large social surveys for attitudes and values.
3. Draw the sample. A synthetic panel of profiles, with quotas, like a conventional sample.
4. Give each profile its persona. The persona is the briefing that describes who is answering: situation, habits, attitudes.
5. Ask the questions. A quantitative questionnaire, often in cells where each group sees a single option, or a qualitative guide for interviews or an AI focus group.
6. Generate the answers. A language model answers for each persona, separately. What comes out is a synthetic respondent.
7. Check. Do the respondents disagree as much as real people? Are the answers too positive? Does the ranking of options hold on a second run?
8. Read and decide. The gaps between options, the objections, the verbatims in their original language. Then, if the decision is a large commitment, confirmation with real people.

How is synthetic research different from traditional research?

What can synthetic research tell you, and what can it not?

It can give you a reasoned ranking of options, the objections that recur, where a message is not understood, and a way to explore thirty routes instead of three. It cannot give you a market share, a figure you can defend to a third party, or a prediction of what one specific person will do. Typical uses are the concept test, product concept optimisation, the ad pre-test, copy testing and name testing.

How do you know if synthetic research is reliable?

Look at the spread of answers first. The most common defect does not show in the averages: simulated respondents who agree with each other too much, which pulls options artificially close together. That is under-dispersion. We measured it against national surveys in eight markets: 0.20 to 0.76 of human spread before correction, 0.96 to 1.07 on the attitudinal layer and 0.85 to 1.00 on the personality layer after. The details and our limits are on our Evidence page.

What should a buyer check?

• Which sources the population is anchored in, by name and date.
• Whether the spread of answers was compared with a human reference, market by market.
• Whether the ranking of options holds on a second run.
• Whether the deliverable states clearly that the answers are simulated, as the ICC/Esomar Code revised in 2025 requires.