What is synthetic research?
Synthetic research uses a model to generate responses that stand in for those of people. A synthetic persona represents a described type of customer; a digital twin may use information about a particular person.
The responses are simulations, not testimony from consumers. Their usefulness depends on the information and method behind them and on what the research is intended to establish.
Synthetic work can make it easier to explore alternatives before commissioning fieldwork. Its cost advantage varies with preparation, validation and the scale of the real research being compared.
How accurate are synthetic audiences?
Accuracy varies substantially by task and method. Matching the ranking of product concepts is different from predicting an individual’s choice.
A PyMC Labs and Colgate-Palmolive preprint tested a method against 57 personal-care concept surveys containing 9,300 human responses. It reported concept-ranking performance at 90% of its human test-retest reliability benchmark. That is not 90% accuracy for predicting purchases. The researchers’ paper
Other work on digital twins found systematic differences between simulated and human responses. The findings concern the people, questions and models studied. Digital-twin research
These studies use evidence more than three months old. Rapid changes in models make them informative about specific methods, rather than a current rating of synthetic research as a whole.
What can synthetic research establish?
It can expose plausible reactions, compare concepts within a validated method and help identify questions for further investigation. It cannot establish that real customers actually hold the views it generates.
The distinction matters when the output is presented as a percentage, quotation or audience segment. A numerical answer can look like survey evidence even when no person was asked. Repeated model agreement can also reflect a shared assumption rather than a reliable finding.
Validation against relevant human evidence gives a narrower and more useful basis for interpretation than realism of presentation alone.
When is research with real consumers still relevant?
Real consumer research remains relevant when the decision requires evidence about actual attitudes, behaviour or circumstances that the simulation has not reliably captured.
The appropriate method depends on the question and the consequence of being wrong. Interviews, surveys, observed behaviour and experiments each answer different questions and carry their own limitations.
Synthetic and human research can be combined in several orders. Screening concepts before fieldwork is one option, but there is no basis for making that sequence mandatory for every category or decision. A strong research design makes clear which conclusions rest on people and which rest on simulation.