Qualitative research is powerful but messy. A room full of opinions can be difficult to translate into actionable recommendations:

• How do you ensure quieter voices aren’t drowned out by the loudest?
• How do you make sense of hours of transcripts without losing weeks to analysis?
• How do you strike the right balance between speed and depth?

Synthetic user research offers a new approach. By feeding interview transcripts and focus group data into a large language model (LLM), we can generate “synthetic representatives” - AI-generated stakeholder profiles that act as composite characters, faithfully reflecting the views of specific groups.

These representatives are not fabricated personas. They are grounded entirely in real user voices. Their job is simple: to represent what a sector, profession, or user group is actually saying.

Through our pilots, we’ve developed a methodology for building synthetic user panels:

1. Prepare inputs – Gather and group transcripts by stakeholder type, ensuring the dataset is comprehensive and, where necessary, pseudonymised.
2. Generate representatives – Define the role, set strict rules (“do not invent views”), and let the model synthesise authentic, composite profiles.
3. Assign tasks – Ask representatives to appraise options, score success factors, or reflect on trade-offs, just as real participants would.
4. Extract recommendations – Triangulate synthetic outputs with human judgement and workshop results, ensuring validity and balance

Our early pilots have offered promising value to our clients:

• Speed – Insights emerge in hours, not weeks.
• Balance – Quieter or minority perspectives are weighted appropriately, not overshadowed.
• Scalability – Panels can be built across multiple user groups simultaneously.
• Value for money – Existing interview data is repurposed, extracting new insights from past projects

For one proof of concept, we compared synthetic scoring with results from a live workshop. The correlation was strikingly strong (up to 0.83), showing how representative these panels can be, even with minimal refinement.

This won’t replace the need to talk to real users, but we are excited by the opportunity to squeeze more value out of the qualitative work we do, as well as offer opportunities for user testing for organisations that have very limited research and insight budgets.

In our next blog on synthetic research, we’ll share top use cases and how to mitigate risks. If you’d like to discuss this with our team, get in touch at hello@thepsc.co.uk.