What do we mean by “synthetic research”?

In practice, synthetic research covers a range of approaches rather than a single method.

Some tools focus on AI-enhanced research repositories:

• These don’t create synthetic populations at all. Instead, they use AI to summarise, cluster and retrieve patterns from real research materials such as interview transcripts and notes. 
• They’re particularly useful for turning existing research into reusable assets, for example, extracting personas, quotes and themes from large back catalogues of qualitative work, speeding up manual analysis.
• Examples include Dovetail, Condens, Notably and Aurelius.

Other tools create client-data driven synthetic audiences: 

• These models are built from client datasets such as surveys or qualitative research transcripts, then extended using AI or statistical techniques.
• Because they are anchored in a specific evidence base, they can be useful for tailored messaging tests, exploring different options, or generating segment-specific insights.
• The key trade-off is data quality: these tools are only as good as the research they’re built from. As part of your planning you should consider how you’ll test & validate outputs.
• At The PSC we can build custom tools to create synthetic personas using your real, qualitative research transcripts, and there’s an increasing number of commercial products e.g. Electric Twin.

There’s also a growing set of general model-driven synthetic respondents: 

• These rely on pre-trained large language models or proprietary datasets to generate personas or responses without reference to a specific client dataset.
• These can be useful for rapid concept validation, early hypothesis testing, and are particularly powerful when no research baseline exists – but you need to consider how representative, or biased, these datasets may be.
• Example tools include Synthetic Users, SyntheticAudience.ai and Viewpoints.ai.

More specialised approaches include behavioural digital twins:

• These include models that are tuned using behavioural data to simulate decision-making (e.g. Brox.ai), offering potential for early UX testing with different groups, including those with accessibility requirements

And there’s also group or society simulators:

• These model interactions across populations of different agents to explore trends or message resonance (e.g. Artificial Societies).
• These approaches can be powerful, but their outputs often require careful interpretation and strong grounding in real-world evidence.

How can public services benefit from these new approaches?

Used appropriately, synthetic research can add real value in public service contexts. Organisations can better reuse existing evidence, drawing on rich (but often under-utilised) research archives. AI-assisted tools can help unlock long-term value from this material. Synthetic approaches can be particularly valuable for early hypothesis testing, generating initial feedback on ideas or personas before investing in full-scale workshops or research rounds. And by using tools with simple interfaces you enable broader access to insight, with faster synthesis makes qualitative insight easier to share across policy, design and delivery teams.

Being clear about limits

Synthetic research isn’t a shortcut to understanding people. Models can reflect bias, obscure uncertainty or appear more confident than the evidence supports. Some of the biggest potential – and risk – comes with small or hard-to-reach research groups, where ‘research fatigue’ limits multiple rounds of traditional research approaches. Synthetic research can help you make the most of limited resources, but with bigger risk, to often vulnerable groups, should you get it wrong.

How to start?

At The PSC, we help clients understand when synthetic research is appropriate – and when it isn’t, ensuring there is appropriate governance, transparency, and real-world validation at the centre of its use. We can help you build tools to better exploit your own research data, and help you deliver research that takes the best of all approaches. 

To find out more, get in touch at hello@thepsc.co.uk.

(1) – How The Times is Using AI to model synthetic focus groups from human audiences (accessed December 2025)