A/B testing is available on Standard and Pro plans.
Overview
A/B testing lets you split traffic between 2–4 published prompt versions with configurable weights. Each new conversation is randomly assigned a version based on the weights, and you can track which version converts better.Start an A/B test
1
Create multiple versions
Write and publish at least 2 prompt versions. You can have up to 4 variants in a single test.
2
Start A/B test
Click Start A/B test next to the Publish button. This opens the A/B test configuration modal.
3
Configure weights
Assign a percentage weight to each variant. Weights must be integers that sum to 100.Use the Auto-balance button to distribute weights evenly.
4
Launch
Click Start test. Any referenced drafts are automatically published. New conversations are now split across your variants.
How traffic splitting works
When a new chat session starts, a database trigger randomly selects a prompt version based on the configured weights. The selected version is recorded on the session, so the entire conversation uses the same prompt.Monitor results
Navigate to Prompt Insights to view per-version analytics:- Conversations — how many conversations each version had
- Appointments — bookings attributed to each version
- Conversion rate — appointments ÷ conversations
- Timeline charts — conversion rate and cumulative appointments over time
Adjust the split
Click Adjust Split to change weights while the test is running. This only affects new conversations — existing sessions keep their assigned version.End the test
1
Click End A/B test
Opens the winner selection modal.
2
Pick a winner
Select the best-performing version. This version becomes the sole active prompt.
3
Confirm
The A/B split is cleared. All new conversations use the winning version.
Publishing a single version while an A/B test is active automatically ends the test and makes the
newly published version active.