Template tests on Shopify

A template test compares theme templates: the control renders the template the page uses today, and each variant renders an alternate template of the same resource type. Shopify only — it relies on Shopify's template system and on the Liquid attributes in the install snippet. If your snippet predates those attributes, update it first; without them a template test cannot run.

1. Create the alternate template in Shopify

In the theme editor, create a new template for the resource you are testing — product, collection, page, blog, or article — and give it a suffix (for example product.summer-layout). Design it however you want; it is a full template, not an overlay. The control needs nothing: it keeps the theme's current template, and control visitors keep a clean URL.

2. Point each variant at its template

In the wizard, each variant names the template suffix it renders. When an eligible visitor is assigned to a variant, Proa navigates the page once to Shopify's own URL for that template — the same ?view= mechanism the platform uses natively. That means:

• One extra navigation per eligible page view for variant visitors. This is the honest cost of a template test, and the wizard states it up front. If the page is highly latency-sensitive, weigh it.
• If another app has already set a view on the URL, Proa never overwrites it — and never counts that page view as control either. Coexistence over contamination.

3. Verify each variant before the draft exists

The wizard will not create the draft until every variant has been verified on your real store. Proa opens the variant's URL, reads which template Shopify actually rendered, and names the outcome: confirmed, or ineligible with the reason. This catches the classic silent failures — a typo'd suffix, a template that does not exist yet, another app's view in the way — before any visitor is split.

Two situations the checker handles explicitly:

• Testing against an unpublished theme. Paste the theme's preview link (or its id) and verification runs against that theme instead of the published one. The verification stamp names the theme it checked — remember the alternate template must exist on the published theme before the experiment starts.
• Password-protected storefront. The checker tells you the password page blocked it, and lets you confirm the variant manually instead of failing with a generic error.

4. Walk the checkout on both templates

Templates differ in more than layout. Before trusting results, place a test order through each template's buy flow — especially dynamic checkout buttons, which behave differently per template and can skip attribution. A template test where one variant attributes orders and the other leaks them is measuring the leak, not the layout.

What is fixed and what is editable

Template suffixes are editable while the experiment is a draft; once it has gone live they are part of the record. Traffic in a template test is always split by Proa itself — external split tools cannot drive it, because the assignment and the navigation have to be one decision. See Traffic split and targeting.