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

# A/B tests

> Split visitors between two versions of a page and measure which one converts.

An experiment compares two saved versions of one page. Visitors are split by weight, each counts one visit and at most one conversion, and you end the test when you have enough data. Developer plan and above. No credits.

Manage experiments at [app.getsite.ai/experiments](https://app.getsite.ai/experiments) or from a page's **Settings → A/B Tests** in the editor.

## Before you start

Make the alternative version. Open the page in the editor, change what you want to test, and **Save**. The page must be published.

## Create an experiment

1. **Page**: the site and page.
2. **Name**.
3. **Variants**: pick the version for **Control (A)** and for **Variant (B)**. They must differ. Drag the slider to set Control's share of traffic, 10% to 90% in 5% steps.
4. **Goal**, one of:

| Goal | Counts when |
| - | - |
| **Click** | A visitor clicks the element matching a CSS selector. The crosshairs button next to the field opens the Control (A) version so you can click the element instead of typing a selector |
| **Form submission** | Any form on the page is submitted |
| **Pageview** | The visitor later loads a path on the same site, such as `/thank-you` |
| **Custom** | Your JavaScript calls `window.__gsAbConvert()` |

The experiment is saved as a **Draft**. Nothing changes on the live page until you press **Start experiment**. Name, variants and goal cannot be changed afterwards in the app.

## While it runs

Each visitor is assigned a variant and keeps it for 30 days through a cookie. The page is served uncached. Only one experiment can run per page.

The results page shows visits, conversions and conversion rate per variant, and a **Confidence** figure. It reads **Needs 100+ visits per variant** until both variants have 100 visits. After that it reads **Statistically significant** at 95% or more and **Getting close...** between 80% and 95%. The test does not stop by itself.

**Pause** stops serving variants; **Resume** continues. A running experiment cannot be deleted. Pause it, then delete it to discard the test without changing the live page.

## End it

**End & pick winner** marks the variant with the higher conversion rate as the winner (Control on a tie) and publishes its version as the live page. This happens whether the experiment is Running or Paused.

A completed experiment cannot be edited or restarted; it can only be deleted.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.