If you came here for the steps to install Google Optimize before running A/B testing on Shopify, the honest answer is that there are none any more. Google retired Optimize on 30 September 2023, the containers stopped serving, and every install guide written before that date — including the one that used to sit at this URL — describes a product that no longer exists.
That does not remove the underlying job. A/B testing on Shopify is still how you find out whether a change helps or quietly costs you money, and the tooling that replaced Optimize is in some ways better suited to a store than Optimize ever was.
Why the old instructions cannot be salvaged
Two things changed, not one.
Optimize itself is gone. There is no container, no anti-flicker snippet worth keeping, and no Optimize reporting inside Analytics. If your theme still carries the old snippet, it is dead weight in the head of every page — find optimize.js or the anti-flicker style block and delete it.
The place the snippet used to go has also gone. Shopify retired the Additional Scripts field and checkout.liquid for checkout customisations, with the last accounts migrated by August 2026. Anything that injected a testing script at checkout has to move to a Checkout UI extension or a pixel. If a guide tells you to paste a tag into Additional Scripts, it was written for a Shopify that no longer exists either.
Rollouts: A/B testing on Shopify, natively
Shopify now ships a native experiment feature, Rollouts. It splits live traffic between your published theme and a variant and reports conversion rate for each.
This is the right tool for one specific question: does this version of the theme convert better than that one. It is a whole-theme split, so it suits a redesign, a new template, or a restructured product page — changes big enough to justify duplicating a theme. It is not built for testing a single button colour, and trying to use it that way means maintaining a near-identical theme copy for the sake of one element.
Because it runs inside Shopify, there is no flicker problem, no third-party script in the critical path, and no consent-mode wiring to get wrong. That alone makes it worth exhausting before you reach for anything else.
When you need more than a theme split
For element-level tests, offer tests and price tests, you need an app. Three are worth knowing:
- Intelligems — the one to reach for when the variable is money. Price testing, shipping-threshold testing and offer testing are its reason to exist, and those are usually the highest-leverage tests a store can run.
- Shoplift — built specifically for Shopify theme and section tests, so it sits closer to how a store is actually assembled than a generic tool does.
- Optimizely — the enterprise option. Powerful, priced accordingly, and generally overkill unless you are already running a structured experimentation programme.
Whatever you pick, check how it renders the variant. A tool that swaps content client-side after the page has painted reintroduces exactly the flicker that Optimize needed an anti-flicker snippet to hide, and that snippet cost you measurable load time.
Measuring A/B testing on Shopify after Optimize
Optimize’s real convenience was that it read your Analytics goals and reported significance for you. Nothing inherited that directly.
In practice you now measure a test in one of two places. The testing app reports on its own conversion tracking, which is fine when the metric is a purchase and the app is reading Shopify’s order data rather than guessing from a pixel. Or you push the variant assignment into GA4 as an event parameter and build the comparison in an exploration.
The GA4 route is more work and has a catch worth knowing before you commit to it: standard GA4 retains user-level and event-level data for 2 or 14 months depending on your setting, and that retention governs explorations specifically. A test you want to re-examine a year later needs its results recorded somewhere else, or a BigQuery export, which is free to switch on.
Note also that a GA4 key event and a Google Ads conversion are not the same object. If your test’s success metric is defined in Ads, do not read it from GA4 and expect the numbers to line up.
What is actually worth A/B testing on Shopify
Most test programmes die because the first six tests were all cosmetic and none of them moved anything. Order the queue by how much of the decision each element carries:
- Price, shipping threshold and offer structure. Nothing else on this list has the same effect size. Free shipping over a threshold versus a lower threshold versus none is a real test with a real answer.
- The product page above the fold. Image treatment, the first line of copy, where the price and delivery promise sit, whether reviews appear before the scroll.
- Cart and checkout entry. Slide-out cart versus cart page, whether the upsell fires before or after the customer has committed.
- Collection page density and filtering. How many products a visitor has to look at before they can narrow the set.
- Trust and risk reversal. Returns policy visibility, guarantee wording, payment badges — worth testing, but expect small effects.
Button colours are last for a reason. They can win, but the size of the win rarely justifies the weeks of traffic spent finding it.
How long to run it
Decide the sample size before you start, not when the numbers look good. A store doing a few hundred orders a month cannot resolve a 3% lift in a fortnight, and stopping the moment a variant goes green is how you ship changes that were noise.
Two practical rules. Run in whole weeks, because Tuesday traffic does not behave like Saturday traffic and a test that starts mid-week and ends mid-week weights one day twice. And do not run a test through a promotional period unless the promotion is the thing you are testing — a sale changes who is on the site, and the answer you get will not hold afterwards.
If the honest answer is that you do not have the traffic to test, that is useful information. It means the next improvement should come from a decision made on judgement and reasoning, and your effort is better spent on traffic than on experimentation infrastructure.
Mistakes that make the result meaningless
Running two tests on the same page at once, so neither result is attributable. Changing the variant mid-test. Letting a test run with unbalanced traffic because the split was configured per-session rather than per-visitor and returning customers keep flipping. Measuring add-to-cart when the thing you care about is revenue per visitor — an upsell that lifts add-to-cart and cuts completion has failed, and add-to-cart will not tell you.
And the quiet one: forgetting the test is live. Ship the winner, archive the loser, and remove the tooling from the pages that no longer need it.
Where this fits
Testing tells you which version wins. It does not tell you what to try (a Shopify heatmap is usually the quickest way to find a hypothesis worth testing), and it will not rescue a page with a structural problem — if the product page is slow, the test is measuring patience, not preference.
If you are not sure which part of the journey is losing people, start by finding the hole rather than testing at random: conversion funnel optimization covers how to locate it. For the tooling side, website optimization tools lists what is worth installing. And if the store itself is the constraint, Shopify SEO services and conversion rate optimization are where we pick it up.
253|
0
