Skip to content

Creating A/B Tests

  1. Go to A/B Tests in the sidebar.

  2. Click New Test.

  3. Enter a name for the test (e.g. “Landing Page Headline Test”).

  4. Choose the test type: Landing Page, Banner, Message, or Bio Page.

  5. Choose the goal: Clicks, Installs, or Custom Event. If you choose Custom Event, enter the event name.

  6. Add variants (minimum 2). For each variant:

    • Give it a name (e.g. “Control”, “Variant A”).
    • Select the content (landing page, banner, message, or bio page).
    • Set a traffic weight (percentage).
  7. If testing landing pages, select the route where the test will run.

  8. Click Create.

The test starts immediately in active status.

Each variant represents one version of your content. You need at least two: a control (your current design) and one or more challengers.

The content for each variant must already exist and be marked as A/B eligible. For example, to test two landing pages, create both landing pages first and enable “A/B eligible” on each one.

Weights are percentages that must add up to 100. Common splits:

SplitUse case
50/50Equal comparison between two variants
70/30Give more traffic to the control while testing a new design
60/20/20Three variants with the control getting more traffic

You can adjust weights after creating the test. Changes take effect immediately for new visitors.

The goal determines which metric you are optimizing for:

  • Clicks: Measures link click-through rate. Best for testing banners and bio pages.
  • Installs: Measures app install conversion. Best for testing landing pages.
  • Custom Event: Measures a specific event you track via the SDK or API. Best for deeper funnel metrics (e.g. purchases, sign-ups).

You can pause a test at any time. Pausing stops traffic splitting; all visitors see the control variant. Resume to continue the test. Data collected before pausing is preserved.

Deleting a test removes it permanently. If the test was active, traffic splitting stops immediately and all visitors see the default content.