Split URL Testing for Links: Judge Winners by Conversions, Not Clicks

Split URL testing sends the people who click one link to two or more destination URLs at a ratio you set (say half to each) and compares what each destination produces. The split happens in the redirect, in the hop between the click and the landing page, so the visitor never sees a choice or a second link. If you want to know what goes on in that hop, our guide to how link redirects work covers it.

Most teams judge these tests on the wrong number. They watch clicks, but every variant shares the same inbound link, so clicks mostly tell you who landed where, plus whatever preview fetchers and scanners wandered in. Conversions after the landing settle the test.

If you run campaign links and want to know which page earns the signups or sales, you can pick one metric before launch, tie conversions back to the variant each visitor saw, and know when a split-traffic A/B test has run long enough to trust.

Clicks Make a Terrible Tiebreaker

On a split link, every variant shares one inbound link. Same email, same post, same QR code, same audience. The redirect deals visitors out at whatever ratio you set, so the click counts on each side should come out close to that ratio almost every time. That's how a split works. When the counts match, you've confirmed the coin flip works, and you've learned nothing about which page is better.

The differences that do show up in click counts are mostly noise. Paste a link into Slack or WhatsApp and a preview fetcher requests it to build the little card with the title and image. Send it in a newsletter and the recipient's email security scanner follows it before any human sees the message, sometimes more than once. Crawlers and scrapers pick it up from wherever it gets posted. Each of those requests goes through the same redirect and gets assigned a variant like anyone else, but they arrive in bursts. A scanner might fire a dozen requests in a minute, and a few of those bursts can easily land more heavily on one side.

That unevenness is enough to change the answer. Say, as an example, each variant gets 400 human clicks, and a couple of scanner bursts add 35 bot hits to variant A and 8 to variant B. A now leads 435 to 408 and looks like the winner, even though the humans split evenly. On a close test, a handful of machines picked your landing page. The guide to bot click traffic goes further into where these requests come from and how much they can distort your counts.

So stop treating clicks as a scoreboard. When both arms get the same audience, the click tells you only that someone or something arrived. The useful signal comes after the landing page loads, when a real visitor either signs up, books the demo, buys, or leaves. That's the number the test should be decided on.

Does Your Split URL Testing Reward the Right Outcome?

Pick the metric before you pick the variants. One metric, written down, matched to what the campaign is for. A product launch email wants purchases. A free-trial push wants signups. A sales-led campaign wants demos booked, or better, qualified leads, because a demo booked by a student researching a class assignment counts the same as one booked by a buyer with budget if all you track is the booking.

You can still look at the other numbers. They just don't get a vote.

Next comes the plumbing that ties a conversion back to the variant someone saw. When the redirect sends a visitor to a destination, the visit carries _actu_clid, a first-party click ID that belongs to that one click on that one variant. When the visitor converts, your server makes a server-to-server call to Acturity with that click ID and the order value, and the conversion lands on the right side of the test. The A/B report then shows statistical significance and confidence intervals per variant, so you're comparing outcomes and can see how sure the comparison is.

The older way to do this is a browser pixel and a cookie, and it leaks. Ad blockers strip the pixel, browsers cap or delete cookies, and visitors decline the consent banner. That alone makes your totals look smaller than they are. The worse problem for a split test is that the leak isn't even. Say one variant is a lean page with a native signup form and the other sends people through a checkout on a different domain. The second one loses its cookie more often, so it looks worse than it really is, and you crown the wrong winner with total confidence. Our guide to cookieless conversion tracking goes through the mechanics. In short, a first-party click ID reported from your own server doesn't depend on the browser cooperating, so both variants get measured the same way.

Clicks still have a job, just a smaller one. Check them in Acturity's link analytics to confirm the split ratio is holding and nothing odd is hitting one arm.

And once the test is running, the metric stays put. If purchases come out flat and you start checking whether one variant "won" on newsletter signups instead, you're no longer testing anything. Run enough comparisons and one of them will look like a winner by chance alone. If the metric you picked turns out to be the wrong one, write down a better one and start a new test.

Set the Sample Size Before You Peek

A split test needs a stopping rule before it needs a winner. Write these down before the first visitor lands, and don't edit them once traffic starts.

• Pick a minimum number of conversions per variant. Count conversions, not clicks or visitors. There's no universal number, but a handful per side tells you nothing, and a subtle difference needs far more than an obvious one. If your link won't reach your target in a reasonable time, test a bigger change.
• Set the duration in whole weeks. Monday traffic doesn't behave like Saturday traffic. A demo booking page might convert steadily on Tuesday and barely at all on the weekend. Run at least one full week, ideally two, so every variant sees the same mix of days.
• Change one thing. A new headline, new layout and new offer in one variant gives you a win you can't explain and can't repeat.
• Don't stop when one side pulls ahead. One side will pull ahead. Early on, a few conversions decide the lead, and small numbers swing hard.

Peeking is how teams end up with false winners. Say you check results every morning for two weeks. Each check is another chance for random noise to cross the line and look significant. If you stop the first time it does, you've picked the luckiest moment instead of the real answer, and the odds of that happening climb with every look.

Acturity's A/B tests report statistical significance and confidence intervals per variant, which helps. Read them on the date you set, not on the day the chart looks exciting.

Three Tests Worth Stealing

None of these come from customer data. They're made-up scenarios, and in each one the click count points one way while the conversions point the other.

The newsletter link. A software team sends its weekly newsletter with one link, go.example.com/q3-report, split evenly between a long-form landing page and a short one with a single form near the top. The team expects the short page to win because there's less to read, and it picks completed signups as the metric before the email goes out. Clicks come back nearly level, and the small lead on one side is the kind of gap a company email scanner can create by itself. Signups go the other way. Readers who'd just finished the newsletter wanted the detail before handing over an email address, and the long page gave it to them. Match the page length to how warm the visitor already is. Our post on short links in email marketing covers the tracking side if you're setting this up.

The LinkedIn post. A B2B team splits a post link between its pricing page and a demo booking page. The demo page looks like the obvious winner, since it asks for the meeting directly. The metric is qualified leads, as marked by sales, and the team writes that down before the post goes live. The demo page gets more bookings, but a lot of them are students and people who are just curious. Pricing visitors who reached out had already seen the number and were still interested, so more of them pass qualification. If you'd counted bookings, you'd have shipped the wrong page. Count the conversion your sales team actually cares about, and decide that before launch.

The printed flyer. A shop puts a QR code on a flyer and splits it between a product page and a discount offer for the same item. The bet is that the product page sells on its own and the discount just gives margin away. The metric is purchases. Scans are close, which makes sense, because everyone saw the same flyer. Purchases favour the discount, since people who scan a flyer at home are often browsing and need a reason to buy now. And because the QR code holds a short link, you can point every flyer at the winning page after the test ends, without reprinting anything.

Fewer Tests, Bigger Answers

A team that launches a new split every week usually ends up with a pile of results nobody trusts. Button colour, a reworded headline, one fewer form field. Each test is too small to reach a real answer, so the winners are mostly noise, and the backlog fills up with changes that never needed making.

Run fewer tests and aim them at choices that change something: which page a whole campaign sends people to, whether a discount earns back its margin, whether a cold audience books demos faster than it reads pricing. Pick the metric, fix the sample size, then wait it out.

Acturity splits the traffic at the redirect and ties each variant to the conversions it produces, with significance reported per variant. If you want to combine a split with country or device targeting, our guide to link routing rules for geo and device walks through the setup.