Split URL Testing
Split URL testing is an experiment method that hosts each version of a page on its own URL and divides incoming traffic between them with a redirect. The control stays on the original address and the variant sits on a separate one, so visitors see different pages rather than different elements on the same page. It is used when versions differ too much to swap in place, such as a full redesign or a rebuilt checkout flow.
How Split URL Testing Works
The split happens at the entry point, before the visitor sees the page.
- Build the variant at its own address. The control stays live. The variant is published separately, often as
/pricing-v2, a staging subdirectory, or an entirely different template. - Set the traffic split. Most tests run 50/50. Uneven splits (90/10) limit exposure when the variant carries risk.
- Redirect a share of visitors. Traffic hitting the control URL is randomly assigned. Assigned visitors are redirected to the variant address. The redirect runs server-side, at the edge, or through a testing tool’s JavaScript.
- Make assignment sticky. A cookie or stored identifier keeps each visitor on the same version across sessions. Without it, a returning visitor can land on both versions and pollute the result.
- Measure at the same conversion event. Both URLs must fire identical tracking, or the comparison measures the tagging rather than the pages.
Split URL Testing vs A/B Testing
The dividing line is where the variant lives. Everything else follows from that.
| Split URL Testing | A/B Testing | |
|---|---|---|
| Variant location | A separate URL | The same URL |
| Delivery | Redirect | Element swap on the page |
| Suited to | Redesigns, new flows, different templates | Headlines, CTAs, images, copy |
| Setup cost | Higher (a full second page) | Lower |
| SEO handling | Required (canonical and redirect rules) | Usually none |
| Result tells you | Which whole experience wins | Which single change wins |
Split URL testing is a form of A/B testing, not a rival method. The statistics are identical: random assignment, a pre-set sample size, significance at a fixed endpoint. Only delivery differs. The tradeoff is interpretation. A split URL test on a redesigned page tells you the new page won, not which of the 30 changes caused it.
When to Use Split URL Testing
Reach for it when the variant cannot be produced by editing the control.
- Full page redesigns. A new layout, template, or design system.
- Restructured flows. One-page checkout against a multi-step checkout, or a new signup sequence.
- Different tech stacks. A React rebuild tested against a legacy template.
- Backend or pricing logic changes. Anything a front-end script cannot alter safely.
- Heavy variants. Large DOM changes applied client-side cause visible flicker. A separate page loads clean.
Do not use it for small changes. Testing a button color across two URLs adds redirect latency and SEO overhead for a change a standard A/B test handles in place.
SEO Rules for Redirect Split Tests
Google publishes explicit guidance for this in its Search Central documentation on website testing. Four rules matter.
Use a 302, never a 301. A temporary redirect tells Google to keep the original URL indexed. A 301 signals a permanent move and can swap the variant into the index in place of the page you are testing.
Point the canonical at the original. Add rel="canonical on every variant URL, referencing the control:
Google recommends canonical over noindex here. With noindex, Google may still pick a variant as canonical, which can drop the original page from the index as a duplicate.
Do not cloak. Googlebot must be treated like any other visitor and enter the same random split. Serving the control to crawlers and the variant to users is cloaking.
Clean up when the test ends. Remove the redirect, retire the losing URL, and update the site to the winner. Tests left running for months read as permanent duplicate content.
How to Track Split URL Tests
Separate URLs make measurement simpler than same-page tests, because analytics already splits the two by page path. Two conditions still need checking.
Attribution must survive the redirect. Campaign parameters on the inbound link have to pass through to the variant URL, or paid and email traffic arrives stripped of its source and lands in direct. Confirm the redirect forwards the full query string.
Traffic sources need to be comparable. If the variant receives a different mix of channels than the control, the test measures the mix, not the page. Segment results by source and device before declaring a winner.
Common Split URL Testing Mistakes
- Redirect latency skewing the result. The variant loads slower purely from the extra hop, and loses on speed rather than design. Run the redirect server-side or at the edge, and consider redirecting both arms so each pays the same cost.
- No sticky assignment. Visitors see different versions on repeat visits, mixing the groups and flattening any real difference.
- Unequal page speed. A new page built on a heavier framework confounds the comparison. Check Core Web Vitals on both before launching.
- Losing the query string. Redirects that drop parameters break campaign attribution for every tagged link pointing at the control.
- Reading one number. A redesign that lifts signups but cuts trial-to-paid conversion is not a win. Track a secondary metric downstream.
Frequently Asked Questions
What is split URL testing?
Split URL testing is an experiment that hosts two or more versions of a page on separate URLs and divides traffic between them using a redirect. The control keeps its original address while each variant lives at its own. It suits changes too large to apply to a single page, such as a redesign or a rebuilt checkout.
What is the difference between a split URL test and an A/B test?
Location. In an A/B test both versions are served from the same URL, with elements swapped on the page. In a split URL test each version has its own URL and visitors are redirected. The statistical method is the same for both. Split URL suits whole-page changes; standard A/B tests suit single-element changes.
Do redirect split tests hurt SEO?
Not when set up to Google’s stated rules. Use a 302 rather than a 301, add rel="canonical on variant URLs pointing back to the original, avoid cloaking Googlebot, and remove the test once it concludes. Problems come from permanent redirects, missing canonicals, and tests left running indefinitely.
How do you split test URLs without a testing tool?
Handle the split at the server or CDN edge. Assign each new visitor a random bucket, store it in a cookie so the assignment sticks, and issue a 302 to the variant for the assigned share. Preserve the full query string on the redirect so campaign parameters survive, and send a variant identifier into your analytics.
To keep each variant separable in GA4, generate tagged links for both arms with the free A/B test URL generator at linkutm.