A/B testing & rotation
Docs/Routing

A/B testing & rotation

Split a link's traffic between two or more destinations by weight, then keep the one that performs best. Starter+ (3 split-tested links on Starter, unlimited from Pro).

Set up a split

app.cloakingurl.comLinks › spring
BasicGeo targetingCloakingA/B testingSchedule & limitsProtectionQR codeHealth
A/B testing & rotationOnly clicks that no targeting rule matched reach this split.
A — long page
yourbrand.com/spring?v=a
70
B — short page
yourbrand.com/spring-b?v=b
30
Per visitor — sticky▾
+ Add variantSave A/B
Two variants weighted 70/30, sticky per visitor.
  1. Open the link's A/B testing tab and click + Add variant for each destination.
  2. Name each variant, paste its URL and set a weight.
  3. Choose the assignment and click Save A/B.

To stop the test, remove the variants and save: the link goes back to its default destination.

Weights

Weights are relative, not percentages: 70/30 and 7/3 split traffic the same way. A variant's share is its weight divided by the total. Three variants at 50/25/25 get 50%, 25% and 25%.

Per click or per visitor

Example: two landing pages, 70/30

You have a proven page (A) and want to try a shorter one (B) on a small share of traffic first.

~70% of visitorsvariant Ayourbrand.com/spring?v=a
~30% of visitorsvariant Byourbrand.com/spring-b?v=b
TipAdd a marker to each variant URL (?v=a, ?v=b, or utm_content) so your own analytics and checkout can tell which page produced each sale.

Splits together with rules

Rules are checked first. Only clicks that no rule matched reach the split. For example, with a rule country is VN → vn.yourbrand.com and a 50/50 split, visitors from Vietnam always get the Vietnamese page, and everyone else is split.

Reading the results

Variant performance on the tab shows clicks and unique visitors per variant over the last 30 days. Compare them with sign-ups or sales per variant in your own analytics (thanks to the marker above), then give the winner all the weight or make it the default destination.

Let a test run until each variant has at least a few hundred visitors before calling it.