Testing tab reference
You'll find what every part of a campaign's Testing tab means, from the holdback card to each A/B test card's status, versions table, and actions.
The campaign's Testing tab is where you measure the campaign and manage its A/B tests. It was called Split Tests; old links redirect. Use this page to look up what a card, tag, or button means. To set a test up, see run an A/B test on a page or audience.
The holdback card
The top card measures the whole campaign against your page as it is. The holdback needs Scale or Pro.
| Part | What it shows |
|---|---|
| Holdback toggle | Turns the holdback on or off. Off, the field dims but stays editable. |
| How many visitors should be held back? | 1% to 100%. A Recommended tag marks the suggested share. |
| Recommendation | "We recommend {n}% for about {n} visitors a day", or "We'll suggest a share once this campaign has a week of traffic." |
| Split bar | Personalized and Held back shares, with visitors a day when traffic is known. Hidden when the holdback is off. |
| Campaign vs holdback | Personalized, Held back, and Difference rates, a verdict sentence, the Likely range bar from Campaign worse to Campaign better, and Difference by week. |
| Status tag | Collecting evidence, Winner, Loser, Inconclusive, Can't reach a verdict, Paused (health issue), Not decision-grade, Paused · campaign off, or Not measurable. |
Without traffic, the card says "No visitors reached this campaign in" the last few days, with Check the snippet. When the holdback is off, it shows how many visitors you'd hold back and Turn on holdback. See compare personalization with a holdback.
The A/B tests section
Under the holdback, A/B tests lists one card per Original page or audience test, with tests that need you first. Open campaign editor opens the editor, where you build versions. With no tests, the section points you to A/B test the original page and A/B test this audience in the editor.
Test cards
Each card has a kind (Page test or Audience test · {audience}), a title, a meta line (goal, week, next check, split, missed checks), a status tag, and a coloured band with the next step.
| Status tag | What's happening | Next step in the band |
|---|---|---|
| Setup | Versions are being built. | Start test in the footer |
| Starts on Publish / Restarts on Publish | The test starts or restarts when you publish. Visitors keep seeing what's published until then. | Cancel start / Cancel restart |
| Collecting evidence | Running, no verdict yet. | None |
| Winner found | A version beat Control and another is still collecting. | End test, or wait |
| Ready to end | Every version has a verdict, or the test reached week 26. Tests never end on their own. | End test |
| Paused (health issue) | Visitors weren't split as configured, or more than 5% saw more than one version. Verdicts are paused; the test keeps running. | Debug on site |
| Paused · publish unconfirmed | RightMessage couldn't confirm when your last publish went live, so results are on hold. Counts show "—". Not a health problem. | None |
| Can't reach a verdict | More than four weekly checks were missed. | Restart with changes or End test |
| Paused · campaign off | The campaign is off. Turning it back on starts a new measurement period without reshuffling visitors. | Turn campaign on |
| Ends on Publish | Ending goes live when you publish. The test keeps running until then. | Undo end |
| Finished | The test ended. Its results are frozen. | None |
| Deployed · {n} days | An Original page winner is shown until your website has the changes. | Undeploy |
A version in a test that ended without a verdict reads No verdict, and the card says "Ended before a verdict. No version was proven better than Control." A finished test never reads Collecting evidence.
The versions table
| Column | Meaning |
|---|---|
| Version | The version's name and share of visitors. Control is "Your page as it is" on a page test, or "This audience's changes" on an audience test. |
| Visitors | Visitors counted in this version. |
| Conversions | Visitors who converted on the goal. |
| Rate | Conversions divided by visitors. |
| Difference | Relative difference from Control. Coloured only once the version has a verdict. |
| Likely range vs Control | A bar of the differences that still fit the results, with the observed difference as a dot, from worse to better. "Not enough data yet" before the first check. |
| Result | Baseline, Collecting evidence, Winner, Did worse, No clear winner, or No verdict. |
The card footer
| Button | What it opens |
|---|---|
| Weekly trend | "Difference from Control by week", with the Observed difference line and Likely range band. Descriptive only. See read results and test with low traffic. |
| How verdicts work | The rules the weekly checks use (below). |
| Send to your analytics | Each version's GA4 experiment string, PostHog feature flag, Google Tag Manager event and page event, plus Preview this version on your site. Before the test starts: "Starts sending when the test starts." |
How verdicts work
The panel says:
- "Results are checked every Friday. Each week's visitors get 7 days to convert before they count."
- "The earliest verdict comes at the week 2 check, once every version has at least 20 visitors who converted and 20 who didn't."
- "A version wins when its whole likely range is better than Control, and loses when it's all worse. Checking every week doesn't raise the chance of a false winner."
- "With more than one challenger, the allowed error is split between them, so each one needs a little more evidence."
- "A version with no verdict by week 26 is called inconclusive. Tests never end on their own."
The full rules, including the boundaries, are in how verdicts work.
Restart with changes
Restart with changes unlocks the versions in your draft. Visitors keep getting the running test until you publish, which starts a new measurement period with a fresh split. Even small edits restart the results. The dialog notes "The current results stay in this test's history as Run {n} · {start} – {end}." Restarting needs Pro.
End test
End test asks which version to keep and, on an Original page test, how visitors get it. Each version shows its verdict, such as "Beat Control at the week 7 check" or "Not proven better than Control yet". Confirm with End test and show {version} or End test and keep Control. Ending doesn't need Pro and goes live on Publish. See end a test and deploy the winner.
Draft reverts
Starting, restarting, and ending are draft changes. Until you publish:
- Cancel start removes a pending start.
- Cancel restart keeps the running test as it is.
- Undo end keeps the test running.
Each puts the test back the way it's published.
Plans
The holdback is on Scale and Pro. Starting or restarting a test needs Pro. Without Pro, a setup card says "Starting tests needs Pro. You can still set up, view and end them." Ending, deploying, and undeploying don't need Pro, and a running test keeps running if you move from Pro to Scale. Grow doesn't include campaigns, so campaign tests stop showing on your site.