Blog
A/B testing with short links: splitting traffic at the redirect
One link, two destinations, weighted. How the split is decided, what the variant report shows, and the three ways a redirect-level test misleads you.
By Wasim Idrishi · Updated · 4 min read
An A/B split on a short link sends some visitors to one destination and the rest to another, from a single link you can print, post or read aloud. The choice is made inside the redirect, in the same millisecond as everything else, and the visitor sees no interstitial and no flicker.
It is the cheapest way to compare two landing pages honestly, and the easiest to misread.
How the split is decided
Each variant has a URL and an integer weight. The weights are summed, the visitor is placed in a bucket, and the bucket falls into whichever variant's cumulative range contains it.
The bucket comes from the visitor hash — the same value the analytics pipeline uses to count unique visitors, built from a rotating secret, the IP address and the user agent, and never stored as an IP. Two consequences follow, and both matter:
The split is deterministic within a day. The same person, same network, same browser, gets the same variant on every click for 24 hours. Refreshing does not reshuffle the experiment.
It is not deterministic across days. The salt rotates every 24 hours, so tomorrow's hash is a different number and the same person may land in the other variant. This is the price of not following people around, and it is a real limitation, not a bug — see below for when it disqualifies the technique.
| Variant | Weight | Share of traffic | Label in reports |
|---|---|---|---|
| Long-form page | 3 | 75 per cent | ab:A |
| Short page with one button | 1 | 25 per cent | ab:B |
Setting one up
Open the A/B testing section of the link editor
Add between two and five destinations with weights. Start at 50/50 unless you have a reason not to; a skewed split needs proportionally more traffic to reach the same confidence.
Give each variant its own utm_content value
This is the step that makes the test useful. The redirect counts clicks per variant; only the destination can tell you what happened next, and it needs a way to tell the two apart.
Check the rules above it
Country and device rules are evaluated before the split. If a rule already catches the traffic you wanted to test, the test never sees it.
Leave it alone
Changing weights mid-test re-buckets everyone and mixes two experiments into one dataset. Decide the split, then decide the end date, then do not touch it.
The three ways it misleads
Calling it too early. A hundred clicks a side feels like a lot on a dashboard and is nowhere near enough to distinguish a 2 per cent difference from noise. Decide the sample size before you start, and if you cannot get it, run the test on a bigger difference instead of a smaller sample.
Ignoring the traffic mix. A split link in a bio and the same link on a poster produce different audiences. If the campaign changes channel halfway through, the second half of the test is a different experiment.
Testing two pages that differ in six ways. A redirect test tells you which URL performed better, not why. If the two pages differ in headline, layout, price and photograph, a win teaches you nothing you can apply anywhere else.
When to use something else
If the two experiences must stay consistent for a returning visitor — a signup flow, a pricing page someone will come back to, anything where seeing a different version tomorrow would be confusing — a redirect split with daily re-bucketing is the wrong instrument. Use a testing tool on the destination that holds assignment in a cookie, and accept the cookie banner that comes with it.
If you only need to compare two things and have no Business plan, two separate short links work. You publish both, in the same place if you can, and compare the counts by hand. It is less elegant and it measures the same thing.
What it is genuinely good for
Comparing two landing pages behind a print campaign, where you cannot change the printed material and cannot run a script on someone else's page. Comparing a video and a text page for the same offer. Deciding whether an app store link or a web page performs better from the same QR code.
In each of those the alternative is not a better test — it is an opinion.
Questions people ask
Will the same person always see the same variant?
Within a day, from the same network and browser, yes. The bucket is derived from the visitor hash, and that hash is rebuilt every 24 hours from a rotating salt, so the same person tomorrow may land in the other variant. It is the direct consequence of not tracking people across days, and it means a redirect split is right for comparing landing pages and wrong for anything where a changed experience would confuse a returning user.
How many variants can a link have?
Two to five, each with an integer weight. Weights do not have to add up to 100 — a 3 and a 1 is a 75/25 split, and it is often easier to reason about than percentages.
Do geo or device rules interact with the split?
They take priority. Country rules are checked first, then device rules, and only traffic that matches neither reaches the A/B split. A test running alongside a rule that captures all iOS traffic is a test on desktop and Android users, whether or not you meant it to be.
Where do I see the results?
Under A/B variants in the link's statistics, labelled ab:A, ab:B and so on in the order you listed the destinations. It counts clicks per variant, not conversions — what happened after the redirect has to come from the destination's own analytics, joined on a UTM value you set per variant.
Which plan includes it?
A/B splitting is a Business plan feature, alongside geo and device targeting. Below that, running two separate short links and comparing them by hand gets you most of the way, at the cost of having to publish two links.
Keep going
Try it with your own link
Paste a long link and get a 10-character gc.mw link with a QR code. No account needed for 30 days; free forever with one.