# Synthetic testing with AI personas

Canonical page: https://www.questionpunk.com/support/synthetic-testing

Create AI-generated respondent personas and run them through your survey before launch.

Time: 10-20 minutes

Level: Intermediate

Audience: Researchers, Survey designers, QA teams

## Steps

### 1. Start a synthetic test

In the survey editor Build tab, click the **Test with AI** button to run a fast synthetic test. You can also manage personas from the [Audiences](https://www.questionpunk.com/support/audiences) page and run them against any survey.

![Start a synthetic test](https://www.questionpunk.com/support/synthetic_testing_tab.png)

### 2. Create personas

Create personas in three ways: **AI-generated** (describe the persona and AI fills in details), **from a template** (pre-built demographic profiles), or **manual** (specify every detail yourself). Personas are saved to your [Audiences](https://www.questionpunk.com/support/audiences) library for reuse across surveys.

### 3. Configure and run

Select which personas to run, set the number of responses, and start the test. Optionally enable **quota sampling** to enforce demographic quotas across your persona pool. You can also run tests through different AI models to compare output quality.

### 4. Review synthetic responses

Synthetic responses appear in the Results tab with a "synthetic" label. Compare how different personas answered to identify question issues, biased wording, or logic problems.

## Details

Synthetic testing is like a dress rehearsal for your survey. AI personas simulate real respondents, helping you catch issues before launch.

Personas can be created with AI assistance: just describe "a 35-year-old marketing manager who is skeptical of new tools" and the AI generates a complete persona profile.

Pre-built persona templates cover common demographics and archetypes, making it easy to test with a diverse set of simulated respondents.

Personas are stored in your Audiences library and can be reused across multiple surveys, saving setup time for recurring studies.

Quota sampling lets you enforce demographic targets (e.g., 50% female, 30% age 25-34) across your synthetic respondent pool for representative results.

Multi-model testing lets you run the same personas through different AI models to compare response quality and behavior. The **multi-model comparison** view in the Report tab shows side-by-side results across models.

**Fidelity reports** compare synthetic responses against real human responses to measure how well 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.

## Related articles

- [Create your first survey](https://www.questionpunk.com/support/create-your-first-survey.md)
- [Design for better responses](https://www.questionpunk.com/support/survey-design-best-practices.md)
- [Fraud detection and response quality](https://www.questionpunk.com/support/fraud-detection.md)
- [Audiences and respondent management](https://www.questionpunk.com/support/audiences.md)