What is Squoosh?¶

Squoosh tests two versions of a web page by sending a panel of AI synthetic shoppers to each one and reporting which converts better. Synthetic shoppers are agents that browse your pages the way real customers do, then decide whether to buy. You give Squoosh your current page and a new version; Squoosh runs the panel against both and returns a verdict, with the reasons behind it, in about 20 minutes for a Standard run instead of the weeks a live A/B test takes.

Squoosh is for e-commerce teams iterating on product pages, landing pages, and checkout flows who want to know whether a change helps before they ship it.

How it works¶

An experiment runs in four stages:

1. You supply two versions. Your current page is the Control page (A), tagged A · Control. Your new version is the Variant page (B), tagged B · Variant. You provide each as a URL — this is the Two URLs option, the default and verified way to ship a variant.
2. A panel of synthetic shoppers visits both. Squoosh builds the panel from your real traffic, so the mix of devices, sources, and behavior matches the visitors you get. Each shopper makes a decision on each version against a conversion goal such as Add to Cart or Checkout.
3. Squoosh measures and reviews the result. It compares how shoppers decided on the two versions and how confident that difference is, and an expert reviewer reads both pages to explain what changed and why it moved shoppers.
4. Squoosh returns a verdict. The result page leads with one of three outcomes — Variant B wins, Keep the original, or Too close to call — followed by a plain-language summary of what drove it and the supporting evidence.

A run can end early once a winner is clear, so you often get a result sooner than the full panel would take. For how the verdict, lift, and confidence are decided, see How Squoosh decides a winner and Reading lift and confidence.

What you can test¶

You can test any page you can reach by URL — a live store, a staging environment, or a Shopify theme. The change is yours to define: a new layout, different copy, a reordered checkout, a swapped hero image. Squoosh compares the two versions you give it; it does not assume what changed.

If your team already manages variants in a third-party testing tool, Squoosh can pull a version from VWO, Optimizely, AB Tasty, or a feature flag in LaunchDarkly, Statsig, or Split. These paths are early and still in validation. For a reliable result, supply your variant as a second URL.

What Squoosh does not do¶

• Squoosh does not write your variant. You design and build version B yourself, then point Squoosh at it. Squoosh compares your two versions; it does not generate, edit, or suggest the change.
• Squoosh does not deploy changes. There is no ship or deploy button. The verdict is a recommendation backed by evidence — rolling out the variant is your decision.

Get started¶

1. Add a property and connect your analytics. See Install Squoosh.
2. Calibrate your synthetic shoppers so they match your real traffic. See Create synthetic shoppers.
3. Once your shoppers are calibrated, open the Experiments tab and click New experiment to launch a test.