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

# Website tests

> AI users operate a real browser on your live site

In a website test, an AI persona drives a **real browser** on your live, publicly accessible website, working toward a task you define. You get a step-by-step action log with a screenshot and reasoning for every step, plus a full session recording.

## Configuration

| Setting            | Description                                                                                                             |
| ------------------ | ----------------------------------------------------------------------------------------------------------------------- |
| **Website URL**    | The page where the AI user starts. The site must be publicly reachable.                                                 |
| **Task**           | What the user should accomplish — e.g. "Find a two-bedroom apartment in Austin under \$2,000 and start an application." |
| **Viewport**       | Desktop sizes 1920×1080, 1536×864, 1366×768 (default), 1280×720, 1024×768 — or 500×1080 to simulate mobile.             |
| **Test sections**  | The dimensions the run is evaluated on. Defaults: *Initial Impressions, User Action & Behavior, Recommendations.*       |
| **Personas**       | One or more [personas](/ai-tests/personas), each with an execution count.                                               |
| **Extended limit** | Raises the per-run cap from 50 to 100 browser steps for longer journeys (may cost up to 2 credits).                     |

## While it runs

* **Live view** — watch the AI user navigate in real time.
* Each step logs the action (navigate, click, type, scroll), the element involved, a screenshot, and the AI's reasoning for taking it.

## What you get

* **Task completion status** with a confidence score — and if the AI gave up, the abandonment reason.
* **Step-by-step replay**: the full click path with screenshots, plus a session video recording.
* **Findings per test section**, and an aggregated summary with key takeaways and recommendations linked back to the runs that produced them.
* Optional **accessibility analysis** — WCAG-based checks with an impact breakdown and recommendations.

## Tips

* **Write tasks with a verifiable end state.** "Add the cheapest annual plan to the cart" gives a clean success/failure signal; "explore the pricing page" doesn't.
* **Start narrow.** Test the critical path (sign-up, checkout, key search) before broad "look around" tests.
* **Match the viewport to your traffic.** If most visitors are on phones, run the mobile viewport — layouts fail differently at 500px.
* **Staging environments** work as long as they're publicly reachable (no VPN or login walls the AI can't pass).