Test up to five versions of a form at once
Experiments now take a control and up to four challengers, each with its own traffic share, its own results card, and its own lift against the control — then promote whichever one wins.
An A/B test used to mean exactly two forms. If you had three headline ideas, you ran two tests back to back and hoped nothing about your traffic changed in between. Now you can put them all in the same test.
One control, up to four challengers
New experiment still starts with one challenger. Click Add challenger to add more, up to four, for five arms in total. Each challenger can be a duplicate of the control that you then edit, or an existing published form, and you can mix the two in the same test.
With more than one challenger they're lettered Variant B, Variant C and so on, everywhere they appear: the experiment page, the Forms list pills, and the declare-winner dialog. A test with a single challenger still just says Variant.
Set the split per arm
Each arm gets its own percentage, from 1% to 99%, adding up to 100%. Split evenly resets them, and adding or removing a challenger does the same. The split stays editable while the test is a draft and locks once it starts, so data you've already collected keeps its meaning.
Every arm is served to the control form's audience, so you're comparing forms, not visitors. A visitor still sees the same arm on every visit.
Results for every arm
The experiment page shows every arm side by side, control first. Each challenger gets a lift against the control ("+20.0% vs control") and a verdict: Beats control, Trails control, Not significant yet, or Not enough data yet. The arm with the best opt-in rate wears a Leading badge once every arm has traffic.
Declare winner lists every arm with its numbers and preselects the leader, so picking the winner of a five-way test takes the same one click as a two-way test.
Starting a test is one click
Challenger forms stay off while the test is a draft, so a duplicate you're still editing never shows up as a second popup next to your control. Start test turns every arm on, publishes each one's latest saved version, and starts splitting traffic. If something would block the start, such as an archived form or a form in test mode, a banner tells you what to fix first.
Challengers stay live only while their test runs. When it ends, the control keeps serving as normal.
Your existing tests carry over
Running and finished two-arm experiments moved to the new model unchanged. They keep their results, history and winners, and their pages now use the new results layout.
How many arms should you run? Each extra arm splits your traffic further, so every arm gets fewer visitors and the test takes longer to reach a verdict. Start with one challenger unless you have plenty of traffic and genuinely different ideas to compare.
See Experiments for the full guide.