Synthetic Consumer Research: Where It Helps and Where It Does Not
A practical, evidence-led guide to using synthetic consumers for exploration, scenario testing and research prioritisation.
Direct answer
Synthetic consumer research uses statistical or generative models to estimate how people may respond to questions, offers or scenarios. It is most useful for rapid exploration, hypothesis generation and prioritising what to test. It should not be treated as automatic proof of demand or a universal replacement for real customers.
Key takeaways
- Use simulation to explore more possibilities before expensive fieldwork.
- Accuracy is specific to a task, population, measure and validation benchmark.
- A responsible workflow combines synthetic exploration with proportionate human evidence.
What synthetic consumer research actually produces
A synthetic consumer is a modelled response, not a recruited human. Depending on the method, it may be grounded in demographic distributions, historical research, behavioural data, qualitative interviews or other evidence. The model then estimates responses under new questions or scenarios.
The category includes different approaches that should not be collapsed into one label. A lightly prompted persona, a statistically generated sample and an interview-grounded digital twin have different evidential foundations and failure modes.
Where it creates practical value
Synthetic research is strongest when the purpose is exploration: compare early concepts, pressure-test a questionnaire, reveal potential objections, examine segment variation and decide which hypotheses deserve real-world research.
BCG describes a hybrid role for synthetic panels and reports a tuned beverage conjoint exercise that predicted real consumer choices with 92% accuracy, while warning that synthetic panels are less suitable for radically new products and do not replace traditional research.[1]
Why one accuracy percentage is insufficient
Accuracy depends on what is being predicted. Matching a survey distribution, reproducing an individual’s answer, predicting an experimental treatment effect and forecasting an actual purchase are different tasks.
Stanford researchers built agents from two-hour interviews with 1,052 people. On the General Social Survey, those agents reproduced responses at 85% of the accuracy with which participants reproduced their own answers two weeks later. The careful wording matters: the benchmark was human test–retest consistency, not a claim of 85% accuracy for every consumer decision.[2]
Use the right evidence for the decision
Greenbook’s 2026 guidance recommends treating synthetic respondents as an augmentation to research, especially for early-stage exploration and questionnaire development, while retaining validation, transparency and human judgment for consequential decisions.[3]
- Low-cost reversible decision: synthetic exploration may be sufficient.
- New category or unfamiliar behaviour: increase qualitative discovery.
- Large media, inventory or pricing commitment: validate with real-world evidence.
- Sensitive or underrepresented population: require stronger provenance and bias testing.
- Policy, health or welfare decision: use independent review and proportionately stronger safeguards.
A credible hybrid workflow
Begin with the decision and define what the simulation is allowed to establish. Document grounding data, assumptions, model and prompt settings. Run variants, inspect disagreement and identify the conclusions most sensitive to those choices.
Then collect human or behavioural evidence where the decision requires it. Compare the results, publish the gap and update the model. The value of synthetic research is not that it removes people from research; it is that it helps teams learn more before asking people fewer, better questions.
Frequently asked questions
What is synthetic consumer research?
It is research that uses modelled consumers or respondents to estimate reactions to questions, products, messages or scenarios.
Can synthetic consumers replace real consumers?
Not universally. They can accelerate exploration and hypothesis testing, but novel, emotional, high-stakes and consequential decisions usually require real human or behavioural validation.
How accurate are synthetic respondents?
There is no universal percentage. Accuracy depends on the task, population, grounding data, model, measure and benchmark used for validation.
Sources and further reading
- Want Consumer Insights Faster? AI Can HelpBoston Consulting Group
- Simulating Human Behavior with AI AgentsStanford HAI
- Synthetic Respondents Explained: What They Are, How They Work, and When to Trust ThemGreenbook