Run an A/B test on a page or an audience
You'll build versions of a page or of one audience's changes in the campaign editor, check how long a verdict might take, then start the test and publish.
Try two to four versions of a page, or of what one audience sees, and let the weekly checks tell you whether a version beats Control. You'll build the versions in the campaign editor, pick the goal that decides, check the time-to-verdict estimate, and start the test with your next publish.
For example, WorkForge (opens in a new tab) could test a trust-first homepage headline against its usual one for everyone who isn't in an audience, and test two agency headlines inside its agency audience.
Choose what to test
| If you'd like to test… | Test | Control is |
|---|---|---|
| The page visitors see when no audience matches | The Original page | Your page as it is. You add up to three variants. |
| What one audience sees | That audience | The audience's own changes. You add up to three challengers. |
You can test the Original page and several audiences in one campaign. A visitor who matches an audience gets that audience's changes on the elements it changes, and the Original page test covers the rest of the page. Campaigns don't need audiences: a campaign with none is a plain page A/B test. Each version is a whole alternative, not a combination of separate element tests.
1. Add the test in the editor
- Open the campaign and choose Edit on page.
- In the left panel, open the actions menu for Original page and choose A/B test the original page, or open an audience's menu and choose A/B test this audience.
- The row gets an A/B test badge, with its versions nested underneath. Choose Add variant from the same menu to add more, up to four versions including Control.
If a winner is deployed on the Original page, A/B test the original page is unavailable: "Undeploy first so a new test compares against your updated website." See end a test and deploy the winner.
2. Build each version
Choose a version under the row, then click the elements you want to change and write the new content, just as you change page content. Control stays unedited on an Original page test. You can rename versions.
The Matrix shows each version as its own column, such as "Original page · Control" and "Original page · Variant A". Empty cells show the original, and inherited version cells follow A. See edit copy in the Matrix.
Check each version in Preview. Original page versions are under the Original page test group, and audience versions are nested under their audience.
Ask yourself whether each version is different enough to matter. Small wording tweaks need a lot of traffic to show a difference.
3. Choose the goal
The campaign's goal decides the verdict. On the campaign's Conversion Goals tab, turn on Tracking for the goal you care about, and choose a Primary goal if you track several. A test can't start without a goal: "Choose a campaign goal before starting this test." See goals and conversion tracking.
4. Check the split and the estimate
Open the test with Review test in the editor, or on the campaign's Testing tab. The setup card shows:
- Split: each version's share. It starts even, and you can set any share from 1% to 99% per version.
- Time to a verdict: "About N weeks", if one version is about 20% better.
- Smallest difference it can spot: the smallest difference reliably spotted within 26 weeks.
Both come from the last 28 days of traffic. Low traffic never blocks a test: if a verdict would probably take more than 26 weeks, the estimate says "Probably over 26 weeks" and you can still run the test and choose a version yourself. See read results and test with low traffic.
5. Start the test
Starting or restarting a test needs Pro. Without it, the card says "Starting tests needs Pro. You can still set up, view and end them."
- Choose Start test.
- Read the dialog. It opens with "Starting is a draft change. Visitors keep seeing Control until you publish." It repeats the estimate and checks whether tested elements above the fold are protected with
rmcloak. - Choose Start test to confirm. The card now reads Starts on Publish.
Start is disabled until the test has a goal, at least two versions, and no challenger that's identical to Control. Changed your mind? Cancel start puts the draft back.
To stop visitors seeing the original before their version, add the class rmcloak to tested elements on your website. See prevent a flash of original content.
6. Publish
Review and publish the project. The test starts when the publish goes live. If the campaign has a holdback, starting a test also restarts the campaign's own measurement against your page as it is, and the publish dialog lists it under Restarts measurement.
If you publish a draft that starts a test without Pro, publishing fails with "Starting an A/B test needs the Pro plan. Remove the test or upgrade to Pro."
Check that it's working
After publishing, the test card reads Collecting evidence, with the week of the test ("Week 3 of up to 26", for example) and the next check date. Send to your analytics lists each version's string and Preview this version on your site, which opens your site showing that version.
The earliest verdict comes at the week 2 check, about three weeks after the test starts, and many tests take longer. Next, learn what the Testing tab shows.