A synthetic respondent is a model-generated stand-in that answers research questions as if it were a member of a target group. No human sat behind the response. No panelist clicked a survey link. That's the first sentence every explanation should keep.
The category also attracts a small zoo of labels—synthetic consumers, AI personas, digital twins, silicon samples. Some are loose synonyms. Others imply extra data, continuity, or individual modeling that may not exist. Read the method, not the nickname.
The short answer
Key takeaways
- Synthetic respondents generate research-like answers; they aren't recruited people.
- Persona prompting, data-grounded simulation, interview agents, and SSR scoring are different construction methods.
- Use synthetic output for low-stakes exploration and screening, then move important questions to humans or behavior.
A definition without the sci-fi smoke
Synthetic respondents are outputs from computational models—often large language models—conditioned on an audience description, persona, data source, or research context. They may answer survey questions, take part in simulated interviews, rate concepts, or produce text that another method turns into a score.
They can be fast, repeatable, and cheap to generate. They can also reproduce model bias, overstate demographic differences, sound more certain than people, and miss lived experience. Both halves belong in the definition.
Sources: NIQ: The rise of synthetic respondents in market research, Nuremberg Institute for Market Decisions: Leaving Insight to Digital Twins?, Kantar: What is synthetic sample?, When Synthetic Users Fail: A Cross-Domain Benchmark
The labels aren't perfectly interchangeable
| Term | Usually suggests | Question to ask |
|---|---|---|
| Synthetic respondent | A generated answerer in a study | How was it constructed? |
| AI persona | A character or segment profile used for responses | Is it grounded in data or just prompted? |
| Synthetic consumer | A generated buyer in consumer research | What purchase context is modeled? |
| Digital twin | A richer representation tied to a real system or data | What exactly is it a twin of? |
| Silicon sample | A model-generated stand-in for a human sample | What population claim is being made? |
Sources: Nuremberg Institute for Market Decisions: Leaving Insight to Digital Twins?, Kantar: What is synthetic sample?
Four common ways they're built
- Persona prompting: describe a person or segment and ask a model to respond in character.
- Data-grounded generation: add customer, market, behavioral, or other context to the model workflow.
- Interview agents: run structured or adaptive conversations with generated personas.
- Semantic scoring: collect free text, then map it to a rating distribution using embeddings and anchors.
Sources: Maier et al.: LLMs Reproduce Human Purchase Intent via SSR
Good uses and bad uses
| Reasonable early use | Bad replacement |
|---|---|
| Brainstorm objections | Customer testimony |
| Pretest survey wording | Representative prevalence estimate |
| Screen rough concepts | Sales forecast |
| Generate interview hypotheses | Lived-experience research |
| Stress-test an audience assumption | Proof that a segment exists |
Sources: When Synthetic Users Fail: A Cross-Domain Benchmark
PaperPMF as one concrete example
PaperPMF generates personas from an audience brief, asks each one twice for a natural-language purchase reaction, and converts the text into five-point probability distributions with semantic similarity. The free preview aggregates 10 generated respondents; the separate full report uses 100.
That's a synthetic concept diagnostic. It isn't a human survey, a population sample, or proof that the product will sell. Keeping that noun phrase intact prevents a surprising amount of nonsense.
Sources: PaperPMF methodology
Sources and verification
Product details are based on official documentation reviewed on August 4, 2026 unless noted. Features and pricing can change; verify them with the provider before making a purchase.
- NIQ: The rise of synthetic respondents in market research — Commercial-industry definition and cautions; updated February 2026 and reviewed 2026-08-03.
- Nuremberg Institute for Market Decisions: Leaving Insight to Digital Twins? — 2026 marketing-research study of synthetic and human responses; reviewed 2026-08-03.
- Kantar: What is synthetic sample? — Commercial-industry comparison and adverse findings; reviewed 2026-08-03.
- When Synthetic Users Fail: A Cross-Domain Benchmark — 2026 preprint on social-attitude and cross-cultural tasks, not product purchase intent; reviewed 2026-08-03.
- Maier et al.: LLMs Reproduce Human Purchase Intent via SSR — Underlying SSR preprint; reviewed 2026-08-03.
- PaperPMF methodology — First-party product documentation; reviewed 2026-08-03.
