What does good synthetic data look like?
When data is personal, it’s precious, especially for research purposes. Typically, such data are securely held, strictly controlled and hard to access. Synthetic data - artificially generated data designed to reflect the characteristics of real data without directly reproducing the original records - could help researchers overcome some of these barriers.
A new study published in the International Journal of Population Data Science (IJPDS), explores how researchers who work with secure data view synthetic data, identifying the characteristics that matter most for its use in research.
Whilst quality frameworks exist for real data, none do for synthetic data. The study sought to change that.
The researchers interviewed 19 people from academic and public sector backgrounds with an interest in secure forms of data. Their experiences spanned education, health, social care and justice data, both from the UK and internationally.
Understanding their perspectives allowed the researchers to identify characteristics that they considered important, including:
- How well synthetic data represents real data, statistically and structurally.
- Whether it can represent real-world processes.
- Whether differences from the original data can make it useful in ways that real data may not be.
- If it is appropriate for its intended purpose.
- The reliability of the data provider.
- How easy it is to access.
- Confidence that the data is provided legally.
Four themes emerged: barriers to accessing real data; the versatility of synthetic data; factors influencing its adoption and usefulness; and the challenges posed by an under-developed synthetic data ecosystem.
These themes were unpacked by the research team in relation to existing frameworks for real data quality.
Emily Oliver, an author of the study and lead for synthetic data at ADR UK, explains:
“We concluded from the interviews that to optimise the usefulness of synthetic data, it needs to be created with clear intention and be of high quality. Knowing what you’re creating it for in the first place is central to both what you create and how you create it, and both these have ramifications for what you invest in that process and how you share it and look after it once you’ve made it.”
The authors also concluded that, like real data, synthetic data needs to be well supported, and capable of improvement. They highlight the importance of good documentation, clear information about intended uses and limitations, and opportunities for users to provide feedback.
Synthetic data should also be accessible and easy to discover, ideally alongside the real data on which it is based. Importantly, it must be reliably safe, with proportionate governance arrangements and clear responsibilities of both users and providers.
Ultimately, the study highlights the importance of designing synthetic data around its intended purpose. It also identifies a need for further evidence, including systematic testing of how useful synthetic data is for different research purposes.
By clarifying what researchers need from synthetic data, the study offers a foundation for developing approaches that make it a more useful and trusted tool for research.
Click here to read the full article

Emily Oliver, Head of Research & Capacity Building, ADR UK Strategic Hub, UK
Frayling L, Pattinson E, Oliver E. What Are the Useful Characteristics of Synthetic Data? A Qualitative Study of Researcher Perspectives. IJPDS [Internet]. 2026 Nov. 17;11(1). Available from: https://doi.org/10.23889/ijpds.v11i1.3406