A redirect test, also called a split URL test, is an A/B test where each variation lives at its own URL and the testing tool sends a share of visitors from the original page to the alternative one.
Use it when the challenger is a different page altogether, such as a full redesign, a new template, or a page on another platform.
For small edits to one page, a standard A/B test is simpler and cleaner.
The trade-off is that a redirect adds a page load and some tracking risk.
In return, you can test changes that a visual editor can't build.
Below is how the method works, when it is the wrong choice, and what real split URL tests in our database found.
Split URL testing: how a redirect test works
In a normal client-side A/B test, everyone loads the same URL and a script changes the page in the browser for visitors in the variant.
In a split URL test, the control and variant are two separately built pages, for example /pricing and /pricing-new.
VWO's help center defines it as "the technique to test multiple variations of your website hosted on different URLs".
Optimizely calls the same thing a redirect experiment and says it fits when you want to compare "two pre-built landing pages against each other".
The flow looks like this:
- A visitor requests the original URL.
- The testing tool buckets them into control or variant.
- Control visitors stay put. Variant visitors are redirected to the variant URL.
- The tool records which bucket each visitor got, so conversions on later pages are credited to the right arm.
Split URL test vs A/B test vs multivariate test
| Standard A/B test | Split URL (redirect) test | Multivariate test | |
|---|---|---|---|
| What changes | Elements on one page | The whole page or flow, at a different URL | Several elements in combination |
| How the variant is built | Visual editor or code injected on the page | A separate page your team or platform builds | Editor or code, one version per combination |
| Best for | Headlines, buttons, sections | Redesigns, new templates, new platforms, different funnels | Learning which element combination works |
| Traffic needed | Moderate | Moderate (still two arms) | High (many arms) |
| Main risks | Flicker | Redirect delay, tracking loss, SEO setup | Underpowered cells |
VWO's rule of thumb matches the table: split URL testing "is preferred when you wish to completely redesign each page of your website, while A/B testing is preferred when you want to test minor changes." Testing several element combinations at once is a multivariate test instead.
When a redirect test is the right choice
- Full redesigns. Hubstaff ran its homepage redesign as a split URL test for 3 months on 259k visitors and got +49% visitor-to-trial conversion.
- A separate experience for one segment. Hype tested a new mobile homepage for new visitors on BluTV as a split URL test over 19 days and 89,290 visitors, reporting +42% trial sign-ups.
- Two whole sites. Shirtinator sent Android tablet visitors 50/50 to either the phone-oriented mobile site or the new HTML5 desktop site, and the desktop site won with +71.81% revenue and +32% orders.
- Same page, different build. Oliver Kenyon compared a client's landing page rebuilt natively in Shopify against the exact same page in a landing page builder: same copy, images and offer. The native build won with +9.9% conversion rate and +20.9% revenue per visitor. A visual editor can't express a change like that, because the difference is the code underneath.
When it is the wrong choice
- Small edits. Changing one headline or button doesn't need a second page. You add a redirect, a second page to maintain, and more chances for tracking errors, with nothing gained.
- Low traffic. A redirect test is still a two-arm test, and redesigns often move metrics in both directions. If you can't reach a sensible sample, a redesign test will likely end inconclusive.
- When you need to know why it won. A redesign changes many things at once. You learn which page wins, not which change did the work. Follow up with iterative tests on the winner.
- Pages where speed is everything. A client-side redirect means the variant group loads two pages. If your tool supports it, a server-side or edge split avoids the extra hop.
Small changes can still be run this way, and sometimes are.
BliVakker.no removed the Facebook login button from its checkout login page as a split URL test on about 8,000 visitors and saw +3% conversions.
With that sample, a 3% difference is hard to tell apart from noise, and the source gives no significance figure.
The redirect added nothing that an on-page test couldn't have done.
Running a split URL test step by step
Most of the work is in the setup.
Get the URLs, tracking and SEO signals right before any traffic flows.
1. Build the variant page and keep it out of the way
The variant URL should exist only for the test.
VWO notes that the control URL should be the one present on the site, so visitors land on control and the tool moves some of them to the variant.
Don't link to the variant from navigation, ads or emails, or you will get visitors who skipped bucketing.
2. Map the URLs, including patterns
If the test covers many pages, such as every product page on a new template, use a URL pattern.
VWO's guide to split URL tests across multiple pages covers wildcard matching.
Check that query strings like UTM parameters carry through the redirect, or your channel reports will break for the variant arm.
3. Set the SEO signals Google asks for
Google's guidance on minimizing A/B testing impact in Search covers redirect tests directly:
- "Use a 302 (temporary) redirect, not a 301 (permanent) redirect."
- "Use the rel=\"canonical\" link attribute on all of your alternate URLs to indicate that the original URL is the preferred version."
- Don't show Googlebot different URLs than humans see. Google calls that cloaking.
- When the test ends, ship the winner and "remove all elements of the test as soon as possible."
Most testing tools use a JavaScript redirect rather than a server redirect.
VWO says its split URL tests work this way.
More detail is on our A/B testing and SEO page.
4. Make sure tracking survives the redirect
The bucketing decision happens on the original page and conversions happen later, so the tool has to carry the assignment across.
Optimizely does this with a redirect cookie, but its docs note that the feature "does not support redirects from one cookie domain to another".
They also say that if the destination page runs a different Optimizely snippet, that snippet must have the Google Analytics integration enabled too.
Two practical checks:
- Install the testing snippet and your analytics on both URLs, and confirm conversions fire from each arm.
- Watch the referrer. A redirect can make the original page look like the referrer for variant visitors, which skews source reports. Optimizely's cookie stores the pre-redirect referrer for this reason.
5. Run an A/A check and watch the split
Redirects are a classic source of uneven traffic.
Variant visitors who leave during the extra load never get counted, so the variant arm comes up short.
That is a sample ratio mismatch, and it biases the result because the visitors who drop out are often the least patient ones.
A short A/A test (control vs. an identical copy at a new URL) shows whether your redirect setup splits traffic evenly before you test a real change.
6. Pick metrics that go past the click
Redesigns often move one metric up and another down, so set guardrail metrics before launch.
Hubstaff's redesign raised visitor-to-trial conversion by 49%, but trial-to-paid fell 7%, which the team attributed to factors outside the homepage.
TheHOTH's stripped-down homepage with only a sign-up form lifted sign-ups from 1.39% to 13.13% over 30 days and close to 3,000 visitors.
The write-up also notes concerns about lead quality, and most of that traffic was already brand-aware.
Convert Better's booking page redesign for High Voltage SEO ran as a 30-day split URL test and reported +119% time-slot clicks.
The metric was clicks on calendar slots, so check booked and attended calls before treating that as a revenue lift.
7. Expect some redesigns to lose
A new page is not a safe bet just because it looks better.
Clean Commit rebuilt a hockey landing page to match a better-performing page for another sport.
After 29 days the redesign showed -38.5% conversions, with an 86.9% probability that the original was better.
Testing the redesign against the original is the reason to run a split URL test in the first place, instead of just launching the new page.
@oliverkenyon
Clean CommitPre-launch checklist
- Variant URL not linked anywhere outside the test.
- 302 or JavaScript redirect, rel=canonical on the variant pointing to the original.
- Query strings preserved through the redirect.
- Testing snippet and analytics on both pages, same cookie domain.
- Traffic split checked (A/A run or SRM check in the first days).
- Primary metric plus at least one downstream guardrail.
- Run length planned in advance, in full weeks.
- Full QA of the variant on every device you include.
You can browse more redesign and full-page results in the test database.