Synthetic research uses AI-generated respondents, also called synthetic respondents or digital twins, to predict how real people would answer surveys, react to concepts or make choices. The best systems are grounded in real human data such as interviews, transactions and behavioral experiments. They are fast and cheap and can match real respondents well on familiar, quantitative questions, but accuracy drops on novel or complex questions. Today they work best as a complement to real research: to narrow options, not to make final calls.
How synthetic research works
Every synthetic research system follows the same basic loop:
- Ground: collect real data about people, such as survey answers, interviews, purchase history or psychometric profiles.
- Model: build AI agents that represent individuals or segments, usually on top of a large language model fine-tuned on that data.
- Ask: put a question, product concept, message or scenario to the agents.
- Aggregate and validate: analyze the answers and, ideally, compare them with real responses to calibrate trust.
The quality of step one decides almost everything. An LLM prompted with a persona description produces plausible answers that drift toward the average person. Agents grounded in real data about specific people perform far better.
How accurate are synthetic respondents?
Accuracy depends heavily on the question. Vendor benchmarks report strong agreement with real respondents on familiar quantitative trends. Independent replications of complex, multi-factor studies report much lower accuracy, and qualitative depth is weaker still. The honest summary: synthetic research is good at directional answers on well-understood topics and unreliable on anything truly new.
Where synthetic research works and where it fails
| Works well | Use with caution |
|---|---|
| Screening many concepts, names or messages quickly | Genuinely new products or categories |
| Finding unclear wording and likely objections | Small or underrepresented groups |
| Directional reads on familiar topics | Emotional, high-stakes or qualitative depth |
| Deciding what to test with real people | Final, irreversible decisions |
Synthetic research inside organizations
Most synthetic research targets consumers and voters. Inside companies, the same idea answers a different question: how will our own people react to a reorganization, a policy or a message? Here the grounding data is not purchase history but how each employee tends to decide and collaborate. Cairo Rosetta predicts individual choices from psychometric profiles, so leaders can prepare a change before it lands. See how to predict employee reaction to change and Simile vs. Aaru for the consumer side.
How to use synthetic research responsibly
- Ask vendors what real data grounds their agents.
- Validate on a sample of real responses before you trust results.
- Use synthetic results to narrow options, then confirm with real people.
- Disclose when results are synthetic.
- Never use simulated individuals to make decisions about real individuals.
Frequently asked questions
What is a synthetic respondent?
An AI-generated participant that simulates how a real person or type of person would answer a survey, react to a concept or make a choice.
Is synthetic research accurate?
It can be on familiar quantitative questions, especially when agents are grounded in real data. Accuracy drops on novel, complex or qualitative questions.
Can synthetic research replace focus groups?
Not yet. Most researchers use it to screen and refine ideas quickly, then validate the most important decisions with real people.
Who are the main synthetic research companies?
Behavior simulation companies include Simile and Aaru for consumer and market research, alongside many survey platforms adding synthetic panels. Cairo Rosetta focuses on employee behavior inside organizations.
Sources
- Park, J. S., et al., Generative Agent Simulations of 1,000 People, 2024.
- Agility PR, Qualtrics 2025 Market Research Trends, October 2024.
- CleverX, Synthetic respondents vs. real participants (2026).
- Greenbook, Synthetic Respondents Explained.
