Split traffic between destinations and read the result

Weights that add up, and a split that stays put

The A/B tab holds a list of variants, each with a destination URL and a weight. The form refuses to save until the weights sum to exactly one hundred per cent, which sounds pedantic until you have debugged a test where ten per cent of traffic vanished into a rounding gap. Two variants at fifty and fifty is the default; uneven splits like ninety and ten are equally valid when you are testing something risky on a slice of traffic.

Variant assignment is deterministic per visitor rather than random per request. The same person following the same link twice sees the same destination, which matters because a visitor bounced between two landing pages produces both a bad experience and a meaningless result. The assignment derives from the same daily visitor hash used for unique counting, so nothing extra is stored and no cookie is set to make it work.

Results reported without overclaiming

The analytics screen gains an A/B widget for links that have a split running: clicks and unique visitors per variant, the actual share each variant received against the share you configured, and the change against the previous period.

Alongside the numbers sits a confidence interval computed with a normal approximation, and a caption that states the threshold at which the difference becomes statistically meaningful. Most link tools show two bars and let the reader assume the taller one won; the point of the caption is to stop a hundred clicks with a six per cent gap from being reported to a client as a victory.

The variant label travels into the click event itself, so every other breakdown can be filtered by it. You can ask whether variant B outperformed variant A specifically on mobile, or only in one country, without setting up a second experiment.

What is worth testing on a link

Landing page variants are the standard case: two versions of a page, one link, real traffic instead of an opinion. Pricing and offer pages benefit most, because the difference between framings is usually larger than any headline tweak.

Beyond that, splits are useful for comparing destinations that are not variants of each other at all — a product page against a comparison article, a signup form against a demo booking, a store listing against a web checkout. Affiliate marketers use them to compare two offers on the same traffic, which is a fairer test than running them on consecutive weeks.

Because the split lives on the link rather than in the page, you can test destinations you do not control and pages that have no room for a testing script.

• ✓Two landing pages on the same campaign traffic
• ✓A percentage rollout of a new page before switching fully
• ✓Two offers from different advertisers on identical traffic
• ✓Web checkout against an app store listing on mobile campaigns

Ending a test is a one-field change

When a variant wins, set the winning URL as the base destination and disable the split. Every link shared until then keeps working — the short URL never changed — and the historical click data stays attached to it, including the variant labels from the test period.

Splits combine with the rest of the link. Targeting rules take precedence, so a market with a dedicated destination is excluded from the experiment automatically. Campaign parameters are appended to whichever variant is chosen, so both sides of the test appear correctly attributed in your own analytics as well as ours.