Synthetic panels
A synthetic panel is a set of people drawn from US population data, each answering as their profile suggests. Run a study against one to find confusing questions, see a rough range of answers, and try a different audience before you field the study to real people.
Synthetic answers are a test, not a finding. Every result from a panel is labeled synthetic, and none of it counts as a response, fills a quota, or reaches your report.
Making a panel
Open Panels in Prism and describe who you want, for example "100 senior citizens in the Midwest who use AI tools every week." Prism turns the description into filters you can check and change. You can also set every filter yourself.
Who's in itfiltersAge, region, income, education, work, household, politics, religion, and more. Each filter narrows the people a panel is drawn from, and the count of matching people updates as you go.
Shape the mixtarget sharesSet the share of each group, say half women and half men, and the draw is weighted to match.
Traits with no data behind themstated sharesPersonality, media habits, shopping, business roles, and fields you add yourself. You say how common each value is. Results mark these traits as assumed.
A panel can have up to 2,000 people. Changing who's on a panel draws a new version, and earlier runs keep the version they used.
Running a study against a panel
On a study's Field tab, choose Run against a panel. You see the cost before it starts. A run continues in the background, and you can pause, resume, or cancel it, or set a spending cap. Each person answers three times by default, so you can see how settled their views are.
A run can also hold interviews: each person talks with the same interviewer your study uses, and every interview gets its own summary.
Up to two panel runs can go at once in a workspace.
Reading the results
In Results, choose whose answers to read: People, one panel run, or Compare to see several side by side. Each run is its own result with its own analysis. Answers from two panels are never pooled together, and nothing is pooled with people's answers.
When people have answered the same study, Compare shows how far a panel's answers were from theirs. The panel keeps that check, so the next study on it shows how close it came last time.
Files you export from a panel run have "synthetic" in their names, and the last row says so.
Where the people come from
People are drawn from the US Census Bureau's 2023 American Community Survey (public-use microdata). Views on politics, religion and trust come from the 2024 General Social Survey, run by NORC at the University of Chicago, matched to each person by age, sex, education, race, and region.