How to test whether a synthetic panel works in healthcare
Everyone shows you the studies that worked. Here is the test design, the public benchmark, and what happened when we pointed it at ourselves.
The report that kept changing
A reviewer told us the same study produced two different numbers. He was right, and the problem underneath was much bigger than the one he had found.
How do you justify synthetic respondents to your client?
The question a research director asks is not "do they sound real". It is "how do you know they are who you said they were?" Here is our answer, the number behind it, and the case where we got it wrong first.
We pre-registered a prediction in our own favor. The data said no. We published it.
A pre-registered study of 2,600 synthetic respondents across eight markets, sealed by hash before we saw a single data point. It showed that fixing a persona's value profile shifts how it reasons, in the predicted direction in all eight markets, and it falsified what we would have loved to claim: that the same layer is what makes markets differ. We publish both.
Synthetic Respondents vs Real Participants: When to Use Which (An Honest Guide)
The question is not whether synthetic respondents are real people. It is whether they fit the decision in front of you. Here is when each one earns its place, and when using the wrong one costs you.
How to Test Your Discussion Guide Before Fieldwork (Without Burning Your Recruit Budget)
Pilot interviews cost almost as much as the study, so most teams skip them and walk into fieldwork with an untested guide. Here is how to pressure-test the guide, the framing and the segments against synthetic respondents first.
Favorite Brand Is Not Refrigerator Brand
Ask someone their favourite juice and you learn who they want to be. Ask what is in their fridge and you learn who they are. Across Spain, the US and the UK, synthetic respondents knew the difference.
When Synthetic Respondents Stop Choosing
What eight markets revealed when we stopped asking them to pick — and started asking them to decide.
Beyond Preferences: How Synthetic Respondents Reveal Decision Processes Under Uncertainty
When a question has an obvious answer, synthetic respondents converge. But when two trusted authorities disagree and both admit they could be wrong, respondents stop choosing — and start designing a process.
You Don't Need More Data — You Need Better Questions
Most teams don't suffer from lack of data. They suffer from starting in the wrong place. And better data won't fix a bad question.
The Problem With AI That Always Sounds Right
Fluency is not validity. The most dangerous failure in AI-driven research isn't being wrong — it's being wrong while sounding right, because confident prose is exactly what stops anyone from checking. Here's what we do about it.
The Surprising Boundary Between Psychology and Culture
In June 2026 we back-tested synthetic respondents against World Values Survey populations in the US and GB. The variance came out narrower than in real populations, and one finding redrew how we separate psychology from culture.
When Consumers Contradict Themselves — That's the Insight
Most research tries to remove inconsistency. Clean the data, align responses, find the pattern. But inconsistency is not noise. It's the signal.
Why Most Qualitative Research Feels Right — and Still Misses the Point
A good interview feels convincing. The respondent is articulate. The story flows. The insights sound clear. And that's exactly the problem.
Synthetic Research Is Not About Speed — It's About Simulating Decisions
Most tools compete on speed. But speed is not the real shift. The real shift is being able to explore how a decision might form, in conversation, before you take the question to real people.
Consumers Don't Know Why They Buy — They Just Explain It Later
Ask someone why they bought something and you'll get an answer. It will sound reasonable. Coherent. Even confident. And it will probably be wrong.
Decision Simulation: A Cross-Market Study on Sustainability (2026)
652 synthetic respondents. 3 markets. Under 2 hours. What this study set out to test — and what the data actually showed.
The Biggest Risk in Research Isn't Speed — It's Starting Wrong
Most research problems don't fail in the field. They fail before the first question is even asked — because the starting point is wrong.
When Synthetic and Human Responses Align — and When They Don't
The question isn't whether you can trust synthetic respondents. It's when different sources tell the same story — and when they don't.
Why Most AI Evaluation Fails at Conversation
Most systems are good at generating answers. Very few are good at sustaining reasoning. And almost none can tell you when a conversation actually holds up.
Persona engine
QualiSynth synthetic respondents are powered by StrataSynth. The StrataSynth blog covers persona construction, segment modelling, and the methodology behind synthetic conversation.
Go to StrataSynth Blog