How to know
what this is worth.

What exactly is a synthetic panel?

A synthetic panel is a set of individually generated customer profiles — age, location, household, income, buying style, price sensitivity, priorities and concerns — built from the audience you describe.

Each profile then evaluates your concept on its own, in a separate simulated interview. Nobody sees anyone else's answer, so the panel does not converge on one opinion the way a single AI summary would.

Where do the participants come from?

They are model-generated from your audience specification and your segment plan. They are not real people, and they are not drawn from any customer database or scraped profiles.

If you define segments yourself, personas are allocated strictly to those segments and to the quotas you set. If you don't, the panel builder clusters the audience and must state its rationale — which is then labelled as model-derived in the report.

How many participants are actually simulated?

Every study simulates a real sample of respondents and reports the panel size it represents alongside it — for example, "20 simulated respondents representing a 250-person panel".

The sample adequacy check in the Credibility tab scores this honestly, and headline figures carry a margin at 95% confidence rather than a false-precision single number.

Which numbers are measured, and which are interpreted?

Measured figures are counted in code from stored participant responses: purchase intent, sentiment distribution, mean enthusiasm, spread, segment shares and per-segment intent. The model is never asked what these are.

Interpreted findings are the analyst layer: themes, objections, opportunities, price read, who loves it and who doesn't. Every figure in the report is badged Measured or Interpreted so you always know which you are reading.

Can I verify a number myself?

Yes. Open the Participants tab and count. Every measured percentage is a straight tally of stored answers, so anyone can reproduce it from the same study.

Are the quotes real?

Quotes are matched word-for-word against stored participant answers. Each verified quote links to the participant who said it. Anything that doesn't match is flagged "do not cite" rather than quietly kept.

What does the credibility band mean?

Each study runs a set of checks — sample adequacy, smallest segment cell size, quota adherence, response completeness, consensus versus spread, quote grounding and segment stability. Each reports pass, caution or fail with a plain-English reason.

They roll up to a band: Indicative, Directional or Weak signal. It is deliberately a band, not a score out of 100, because a two-decimal confidence figure would be exactly the false precision we are trying to avoid.

Is a study repeatable?

The measured layer is reproducible from the stored responses of that run. A fresh run generates fresh personas, so expect movement in the interpretation. Using saved segment definitions keeps the breakdown comparable across studies.

Every completed study stores an audit trail: model, pipeline version, stage timings, questions asked, audience spec, and planned versus realised panel composition.

No real people are involved, so there is no personal data to protect

Participants are model-generated profiles built from the audience description you write. No customer lists, no panel providers, no scraped profiles, no personal data of any real individual is used to create them.

That removes the usual research headaches: no consent forms, no incentive payments, no respondent PII sitting in a spreadsheet, and no personal data to delete when someone asks.

What about GDPR?

Because the panel contains no personal data of identifiable people, running a study is not processing of personal data — there is no lawful basis to establish, no privacy notice to serve to respondents, and no data subject rights attach to the participants.

Anything you paste into a concept is your own material. Do not include real customers' personal data in a concept, question or audience spec; nothing in the pipeline requires it.

Who can see my studies?

Studies, panels and results are scoped to your account and enforced at the database level, not just in the interface. Nothing is shared with other accounts unless you export or share it yourself.

Your concepts and questions are sent to the model to run the simulation, and are not used to train it.

Any names or quotes are invented

Participant names, ages and locations are fictional constructions used to make the sample readable. Any resemblance to a real person is coincidental, and quotes are the words of a simulated respondent — never a real customer.

It is not a substitute for fieldwork

Synthetic research is decision-support before you spend money — a way to pressure-test a concept, sharpen the questions and find the obvious objections early. Major, expensive or irreversible decisions still warrant real customer validation.

It is not a market forecast

Purchase intent here is intent within a simulated panel, not a demand projection. Treat the direction and the relative differences between concepts, segments and price points as the signal — not the absolute number.

It is not free of model bias

Participants inherit whatever the underlying model believes about the audience you described. Under-represented or niche groups are the most likely to be flattened into stereotypes. Vague audience specs make this worse; specific ones make it better.

It is not evidence of what people will do

Stated intent and actual behaviour diverge for real respondents too. Synthetic respondents have no budget, no habits and nothing at stake, so the gap is wider. Use it to rank options, not to promise outcomes.

Small segments are not readable

A segment with a handful of participants is a quotation source, not a statistic. The cell-size check flags these, and the report shows ranges rather than point estimates where the sample is thin.