Variant visitors hit a 404 or broken page
The variant URL isn't reachable. Confirm the page is published in Shopify and loads in a browser, and check for a typo in the Variant URL (the most common cause). Cooee only performs the redirect — it doesn't create or host the page, so the variant must stay live for the whole test.
The variant isn't redirecting
Usually one of these — the first two are expected:
The visitor is in the control arm. Only the variant's share of traffic is redirected; the rest stay on the origin as the comparison group.
The visitor isn't on the origin page. The test only runs there — a visitor is entered only when they land on the origin.
The Origin URL doesn't match. It must be the page's canonical path, entered exactly, with no wildcard. A collection-nested product URL (
/collections/…/products/shoe) still matches its canonical/products/shoe, and a trailing slash is tolerated — but a different path won't match.
The Variant URL is rejected
The variant must be an absolute URL that starts with https://. A relative path or an http:// URL won't be accepted.
Search engines are indexing the variant URL
For an experiment that runs more than a few weeks, add a rel="canonical" tag on the variant page pointing back to the origin URL. This tells Google the origin is the preferred version and avoids keyword cannibalisation. If you don't want the variant indexed at all, add a noindex tag to it.
External analytics attribute sessions to the variant URL
Because assignment is sticky, a variant visitor's later sessions start at the variant URL, so tools like Google Analytics attribute those sessions to that path. This is expected — read external analytics alongside Cooee's per-arm data, which already splits by arm.
QA sessions are showing up in results
Bots and automated tools are excluded, but a real-browser QA session can be assigned. Test in a private/incognito window and clear the site's data between checks, since assignments are sticky.
Two tests targeting the same page
Run only one Split URL test per origin page. If two experiments name the same origin, that's a misconfiguration — only the first one takes effect on that page.