> ## Documentation Index
> Fetch the complete documentation index at: https://docs.buildbetter.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Persona Tests

> Run structured research studies against a persona or panel — collect scores, sentiment, verdicts, and aggregate metrics

Persona Tests are structured research runs. You show a concept, message, or set of options to a persona (or a whole panel) and every respondent returns a scored, tagged, cited reaction. Aggregate results roll up into a read on how the idea landed across the audience.

## Study Types

<CardGroup cols={2}>
  <Card title="Qualitative" icon="comment-dots">
    Open-ended reactions. Each respondent writes a verdict, summary, and reasoning in their own voice. Best for messaging, positioning, and concept feedback.
  </Card>

  <Card title="Single-Choice" icon="list-check">
    Respondents pick one option from a set you provide. Returns a vote breakdown plus per-respondent reasoning for each choice. Best for A/B testing headlines, designs, or value props.
  </Card>
</CardGroup>

## Launching a Test Run

<Steps>
  <Step title="Write the prompt">
    This is what respondents will react to — a headline, a product concept, a value prop, a chunk of landing page copy, anything.
  </Step>

  <Step title="Pick the study type">
    Qualitative for open reactions, or single-choice if you want respondents to pick an option.
  </Step>

  <Step title="Add options (single-choice only)">
    Add 2 or more options. Each option can have a label and an optional image.
  </Step>

  <Step title="Attach prompt images (optional)">
    Add up to 4 images to show alongside the prompt — useful for design mocks, screenshots, or marketing assets.
  </Step>

  <Step title="Pick the audience">
    Either select people directly, select a list of persona IDs, or choose an entire saved panel. Replica sets inside a panel all participate.
  </Step>

  <Step title="Launch">
    Review the estimated credit cost, confirm you have enough credits, then launch. Runs go through `queued → running → completed` (or `failed`). Large panels run asynchronously — you can leave the page and come back.
  </Step>
</Steps>

<Info>
  Persona test runs show estimated credit cost before launch. Fully failed runs are not charged, and partially completed runs are billed only for successful completions.
</Info>

## What Each Respondent Returns

For every respondent, the test produces:

| Field                | What It Means                                                                                  |
| -------------------- | ---------------------------------------------------------------------------------------------- |
| **Overall score**    | 0–100, higher is more positive. Color-coded in the UI (green 75+, yellow 50–74, red under 50). |
| **Sentiment**        | Positive, neutral, or negative.                                                                |
| **Verdict**          | One-line bottom-line reaction.                                                                 |
| **Summary**          | First-person reaction in the persona's voice.                                                  |
| **Reasoning**        | Why the persona reacted that way, grounded in their profile and signals.                       |
| **Tags**             | Topic/theme labels the persona reacted to — used for aggregate top-themes rollup.              |
| **Selected option**  | The option they picked (single-choice only).                                                   |
| **Dimension scores** | Clarity, relevance, trust, and likelihood to act — each 0–100.                                 |
| **Cited signals**    | Signal IDs the respondent referenced when reacting.                                            |

## Aggregate Results

At the top of every completed run you'll see:

* **Response totals** — completed, failed, and running counts
* **Average scores** — overall plus each dimension (clarity, relevance, trust, likelihood to act)
* **Sentiment distribution** — positive / neutral / negative counts
* **Top tags** — the themes that came up most across respondents
* **Option breakdown** (single-choice only) — vote count and percentage per option, with the winner highlighted

Below the aggregate you get a sortable, filterable respondent table showing each persona's score, sentiment, verdict, and expandable reasoning.

## Saving and Sharing Results

Every completed run has two export paths:

* **Save to Knowledge** — creates a structured Knowledge page with the prompt, aggregate metrics, top respondents, and breakdown by type. Keeps the finding inside BuildBetter next to other research.
* **Export to PDF** — produces a formatted report with the same content for external sharing.
* **Download marketing assets** — for marketing variant projects, download project assets and winning variants.

## Marketing Variant Projects

Marketing variant projects let you test creative ideas against saved panels or personas:

1. Start from a marketing project or directly from a persona panel.
2. Choose the target audience from saved panels or personas.
3. Upload assets or text ideas.
4. Generate AI image variations.
5. Compare results and review synthetic persona feedback.
6. Download project assets and winning variants.

Before larger generation and evaluation runs, BuildBetter shows projected usage estimates and asks for confirmation.

<Tip>
  Run the same prompt against two different panels — e.g., prospects vs. power users — and compare the aggregate cards side by side. It's one of the fastest ways to see where messaging resonates differently by segment.
</Tip>

## Limits

* Up to **100 replicas** per panel member
* Up to **4 images** per test prompt
* Single-choice tests require **2 or more options**
* No limit on qualitative study prompt length