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

# A/B testing messages

> Write up to three versions of a message or invitation note, compare them fairly, and optionally send the winner automatically.

You can test different wording for a **Message** step or a **Connection request** note. Each person gets one version, and [Campaign analytics](/campaigns/analytics) compares the results.

## Add a version

1. Open the sequence in **Sequences** and select a message or invitation step.
2. Choose **A/B test this message**. A **Version B** is added, empty, so the step is flagged until you write it.
3. Write Version B. To add a third, choose **Add version C**. A step can have up to three versions (A, B and C).
4. To drop one, select it and choose **Remove Version B** (or C).

The step shows a badge such as **2 variants**. The final close-the-loop message of a playbook cannot be A/B tested.

## How people are split

* **Each person gets one version, fixed when they are enrolled.** The split is even and never changes partway through a sequence, so retries and re-reads always use the same version.
* **Every version is approved with the sequence.** Review shows each version, and the approval covers all of them. Changing any version's text needs a new approval.

## Reading the results

In the sequence's analytics, the step's row shows each version's sends and its acceptance rate (for invitations) or reply rate (for messages). Versions are compared only once **each has at least 30 sends**; until then you see that it is too early. A comparison is reported as a difference only when it is statistically significant (p \< 0.05).

## Send the winner automatically

The workspace **owner** can turn this on for any step with versions, in that step's row in the sequence's analytics.

1. Under **Send the winner automatically**, choose **Judge on**: **Acceptance rate** (invitations) or **Reply rate**.
2. Choose **Turn on**. The status shows **Comparing**.
3. Entryway re-checks about once an hour. A winner is recorded only when one version is significantly better than every other version (p \< 0.05, at least 30 sends each). The step then shows **Winner A**, **B** or **C**, with the numbers the decision was based on.
4. From then on, people enrolled after the decision get the winning version at that step. People enrolled earlier keep their version.
5. To undo it, choose **Revert to an even split**. New people go back to the even split and the setting turns off. **Turn off** stops comparing before a winner is found.

Turning this on or off does not need a new approval, because every version was already approved.

<Note>
  Results need real sends, so A/B comparisons start once LinkedIn sending is switched on for your
  workspace. Small campaigns may never reach 30 sends per version; in that case no winner is
  chosen.
</Note>


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