What is it?

Synthetic panels use AI-generated responses to simulate human responses to surveys. The technology is intended to support survey modelling, testing, forecasting or augmentation.

Synthetic panels are typically developed in three ways:

1 )      Using interfaces surrounding a large language model (LLM), for example ChatGPT or Claude, to develop responses drawn from publicly available information.

2 )      Using machine learning to generate additional synthetic responses to a survey based on a smaller sample of real human responses.

3 )      Training an AI on human response data to deliver targeted insights in great detail. This often also uses open data sources, with the aim of focusing on specific market cohorts or segments.

Not to be confused with: synthetic personas

Synthetic panels generate results simulating a large sample of virtual respondents, while a synthetic persona is a representation of a consumer segment based on a generalised data set.

Why does this matter?

Synthetic panels offer the potential to scale and increase access to consumer opinions, in particular for niche audience groups that would otherwise be hard to cover in traditional research. Other areas of use include concept testing, survey simulation, segmentation modelling, questionnaire expansion and scenario analysis.

However, panels must be based on good quality data to be useful. Poor data could mean that inaccuracies creep into survey results. Likewise, using publicly available data to generate responses could also unintentionally draw on biased or skewed perspectives.

The challenge for synthetic data is whether it will ever be able to fully replicate the responses of a real human respondent. Currently, the consensus is that the technology is not yet at the level needed to be used as a direct replacement for humans in research.

What does it mean for research?

For research, potential benefits include increasing the speed of research, reducing costs associated with participant recruitment and management, and helping with early-stage idea testing.

Hasdeep Sethi, group AI lead at Strat7, said: “In quant research, it means creating new respondents (‘vertical scaling’) in a survey, or simulating what the same respondents would answer to new questions (‘horizontal scaling’). These can be outputs from digital twins, or more traditional statistical approaches. 

“Our research at Strat7 suggests that while the field is progressing fast, synthetic data should be seen as a complement to and not a replacement for primary research with real humans, especially in high-stakes or more complex research, where there is a significant commercial risk if the wrong decision is made. Regular validation tests and more transparency on underlying methodologies from vendors will help drive further adoption in this space.”