Test your survey with synthetic AI personas
Create AI-generated respondent personas and run them through your survey before launch.
Synthetic testing creates AI personas with specific demographics, attitudes, and backgrounds, then runs them through your survey so you can check question flow, logic, and quality before real respondents see it.
Before you start
Open the survey in Build. Synthetic test runs show an estimated cost before you start. Build from your data also uses AI credits; prepare enough written responses or notes to support distinct personas, and use only material you are permitted to use. Existing interview questions keep their saved model selection.
Prepare a test with AI personas (classic editor)
A navigation overview using actual English-interface demo screenshots, generated English narration, and English/Japanese captions. This is a narrated screenshot sequence, not a continuous recording. Examples are fictional. This walkthrough shows the classic editor. Your account may show different controls; follow the written steps for your editor.
Download videoRead the video transcript
In Build, click Test with AI. Use the adjacent AI test runs history control to return to earlier runs. Choose the number of respondents, persona source, and model, then check the estimated credit cost. Review the configuration and estimated cost before starting a run, or cancel to return without spending credits. When a test finishes, filter Results by Synthetic responses to verify logic and wording. Synthetic answers test your setup and are not human research evidence.
Steps
Open a synthetic test from Build
In the survey editor, stay in Build. Open the menu beside Preview and choose Test with AI respondents, or click the separate Test with AI button if shown. Choose how many respondents to run. Then choose a respondent source: auto-generated diverse personas, an audience, custom targeting (describe who you want), panel-based personas, or digital twins of real respondents. You can also manage reusable personas on the AI Personas page and run them against any survey.English interface · Fictional demo questionnaire. This screen does not show participant findings. Classic editor shown.
Create personas
Create reusable personas with AI Generate, from a template, or Manual on AI Personas. To use existing evidence, find From your data and click Build from your data. Choose a survey, paste call notes or support tickets, or upload text, Markdown, or CSV notes. Select 3–6 personas and click Build personas. Review the source quotes shown under each persona before using it.Set options and run the test
Choose a Respondent source, such as auto-generated diverse personas or a saved list. Set the Number of respondents to 3, 5, 10, or a custom number up to 500. Then start the test. Under Advanced, turn on Diverse responses to increase variation in answers. Set Target marginals to guide answer shares for multiple-choice questions. You can also choose different AI models from the model picker and compare a pilot run. Before you start, the dialog shows an estimated credit cost or "Free". It updates as you change the model, respondent count, and run options. While the run is in progress, a progress banner shows the estimated time.English interface · Two views of the test setup before a run. Model options and credit estimates can change; no generated responses are shown. Classic editor shown.
Review synthetic responses in Results
Synthetic responses appear in Results with a "synthetic" label. Review how different personas answered. Use this to spot question issues, biased wording, or logic problems.English interface · Synthetic filter navigation overview. No AI test has been run in this demo. Classic editor shown.
Details and limitations
Synthetic testing is a dress rehearsal for your survey. AI personas simulate respondents so you can catch issues before launch.
From your data builds personas from the selected survey’s newest 150 real responses and AI interviews, from notes, or from both. Each persona keeps source quotes that are supplied when it answers. Review those quotes and the resulting profile: grounding a simulation in your data does not turn its new answers into human evidence or a representative sample.
Pre-built persona templates cover common demographics and archetypes. This makes it easier to test with a varied set of simulated respondents.
Personas are stored on the AI Personas page and can be reused across multiple surveys. This saves setup time for recurring studies.
Target marginals let you set the answer percentages you expect on multiple-choice questions, for example from census figures. Diverse responses samples each persona's answers from estimated probabilities so runs do not all answer alike.
Use the model picker to compare responses from different models. In Build, open the AI test runs history control beside Test with AI to inspect past runs and credit usage.
Fidelity reports compare synthetic responses against real human responses to measure how closely AI personas mirror actual respondent behavior. The Human vs. Synthetic comparison view highlights where synthetic and real distributions align or diverge.
Run history tracks all synthetic test runs, so you can compare results across iterations as you refine your survey. Synthetic answers are simulations for testing. Do not present them as evidence from real participants.
Check the result
You should see responses marked "synthetic" in Results and be able to compare how personas moved through the survey.
If something goes wrong
If test controls are missing, open the menu beside Preview in Build. Past runs are under AI test runs in that menu, or beside the separate Test with AI button when shown. If data-based persona creation reports too little evidence, add more written responses or longer notes; numeric answers alone may not provide usable source text. If it reports insufficient credits, check your balance before trying again. Review any notice about unmatched quotes being left out.



