How do I A/B test a link?

Create a rotator short link on urlik.xyz, add two or more destinations and set the split to 50/50. Share the one link; visitors are divided between the variants automatically and the click stats show each variant separately.

How long should an A/B test run?

Long enough to cover full weeks and gather a few hundred conversions per variant. Weekday and weekend traffic behave differently, and stopping early — the moment one variant pulls ahead — is how noise gets mistaken for a result.

Should I judge an A/B test on clicks?

No. Clicks tell you the split worked, not which variant pays. Tag each destination with its own UTM so your analytics can attribute signups or revenue to a variant, and decide on that.

Can I send mobile and desktop visitors to different pages?

Yes — that is device routing rather than a test. One short link can send mobile traffic to an app store and desktop traffic to the web app, which is useful for a QR code or a link you can only print once.

A/B Testing Links: Split Traffic, Find What Converts

Everyone has an opinion about which headline wins. The opinion is usually wrong, and it is always cheap. An A/B test replaces the argument with a number — and at the link level you can run one in about two minutes, without touching your website, without a developer, and without an analytics rollout.

What you can actually test with a link

Link-level testing means one short link that splits traffic between two or more destinations. The visitor clicks once; the link decides where they land. That lets you answer a specific class of question:

Which landing page converts better — the long explainer or the short one with the form above the fold.
Which offer pulls harder — free shipping or 10% off.
Which product to lead with in the same newsletter slot.
Which pricing layout gets more signups — three tiers or two.

What it does not test is anything on the same page: button colours, headline copy above the fold, form length. That needs a page-level testing tool. Link splitting is for testing whole destinations against each other — which is the bigger, likelier win anyway.

Set it up

On urlik.xyz a rotator link takes a minute: create your short link, add the destinations you want to compare, and set the split. Share one link everywhere; the traffic divides itself.

Two things to get right at setup:

Weighting. 50/50 is the default and the right choice for a real test. Unequal splits (90/10) are for cautiously introducing a risky variant, not for measuring — a lopsided test needs far more traffic to reach the same confidence.
Tag the variants. Give each destination its own UTM tag so your analytics can tell them apart after the click. Without this you will see the split at the link and a shapeless blob everywhere else. Our UTM guide covers the naming.

Measure the thing that pays

Here is where most link tests go wrong: they measure clicks. Clicks are not the outcome. If variant A gets 20% more clicks and converts half as well, A lost.

The link tells you the traffic split; your analytics or your database tells you what happened after. Decide the metric before you start — signups, purchases, revenue per visitor — and make sure you can attribute it to a variant. Then let the tracker show you clicks, sources and devices per variant, and match those against conversions on your side. There is more on reading the click side in how to track link clicks.

Run it long enough to mean something

The failure mode of A/B testing is not bad tooling, it is impatience. Some rules that keep tests honest:

One variable at a time. Change the offer or the layout, not both. Two changes and a win tells you nothing about why.
Run full weeks. Tuesday behaves nothing like Saturday. A test that starts Monday and ends Thursday measures the working week, not your audience.
Do not peek and stop. Early leads are noise. Checking hourly and stopping the moment a variant is ahead is how you turn randomness into a decision you will regret.
Respect small numbers. 12 conversions against 9 is not a result, it is a coin. If the difference is under about 20% on a few hundred conversions, treat it as a tie and go test something bigger.
Ship the winner, then re-test. Yesterday's winner is today's control. Audiences move.

A worked example

You are sending a newsletter to 20,000 people and you want to know whether the discount or the free-shipping angle sells better. You build one rotator link, 50/50, pointing at two landing pages that are identical except for the offer, each tagged utm_content=discount and utm_content=shipping. The newsletter carries one link.

A week later the link reports 1,840 clicks, near enough evenly split. Your store reports 61 orders from the discount page and 88 from free shipping — a real gap on a real sample. Free shipping wins, you make it the default, and next month you test it against a bundle. That is the loop. Each round is cheap and each round compounds.

Beyond testing: rules that route

The same mechanism does things that are not tests at all. Route mobile visitors to the app store and desktop visitors to the web app. Send one campaign's traffic to a region-specific page. Here the split is not an experiment — it is just sending each visitor somewhere sensible, from one link you can print.

The short version

Build one rotator link, split it evenly, tag each variant, decide your metric up front, run full weeks, ignore early leads, and judge on conversions rather than clicks. The tooling is the easy part — the discipline is what produces answers you can spend money on.