How does a synthetic survey work?
First the audience is defined: you choose attributes such as age, income, city and shopping habits, or use a ready-made sample that matches Türkiye’s population. Then the questions are written and put to every synthetic respondent in the sample, one by one.
Each respondent does more than tick an option. They also explain why they chose it, so the same study produces qualitative reasons alongside the quantitative distribution.
Which questions does it suit?
Synthetic surveys work best for questions that compare options.
- Price: at which price does demand fall quickly, and which price brings the highest revenue?
- Product: which feature influences choice the most (conjoint, MaxDiff)?
- Communication: which of two messages or two package designs is more convincing?
- Advertising: is the message remembered, and does it create purchase intent?
How accurate are synthetic respondents?
Published research shows that language models can reproduce the overall pattern of human answers in many classic surveys and experiments. The same studies show that accuracy is not equal across subgroups and that models need calibration.
At SCL, models are therefore compared with real consumer data, and results are presented as findings that point a decision in a direction, not as exact forecasts. Reading a synthetic survey next to a small real survey is the most robust approach.
How is it different from a conventional survey?
In a conventional survey, most of the time and budget goes into finding participants. In a synthetic survey the sample is ready, and dozens of scenarios can be put to the same audience in a single day. A conventional survey, on the other hand, measures what real people say, and it keeps its place for final validation.
| Synthetic survey | Conventional survey | |
|---|---|---|
| Respondents | AI agents with defined profiles | Real people |
| Time to results | Minutes | Days to weeks |
| Sample size | From one profile to thousands | Limited by budget |
| Reasons | Come with every answer | Need extra open-ended questions |
| Validation | Needs comparison with real data | Direct statements |
Frequently asked questions
What is a synthetic respondent?
A synthetic respondent is an AI agent that resembles a real consumer in attributes such as age, income, city and shopping habits, and answers research questions according to that profile.
What is the smallest sample for a synthetic survey?
You can interview a single profile. To see a distribution, samples of several hundred synthetic respondents are typical.
How long do synthetic survey results take?
It depends on the size of the study. Most surveys finish within minutes, and the report can be downloaded as a PDF or Excel file.
When should a synthetic survey not be used?
It should not be used on its own for entirely new categories the model knows very little about, or for decisions with serious legal or financial consequences. In those cases, read it together with research on real consumers.
Sources
The assessments on this page draw on published academic work on simulating survey respondents with large language models. You can read our summaries on the articles page. Read the article summaries.
Related pages
- What is synthetic data?
- What is a digital twin of a consumer?
- Try the demo
- Example study: price sensitivity in snacks
Last updated: September 2026