What you can test¶
Squoosh tests two versions of a page against each other and reports which converts better on your goal. You supply both versions. Squoosh does not write or generate the variant — it tests the real change you built.
You tell Squoosh how to reach each version when you set up an experiment, under How do you ship variants?. There are three ways to do this. Two URLs is the recommended path and works for almost every change.
Test any two URLs¶
This is the default and the most reliable way to run an experiment. You give Squoosh two web addresses — the Control page (A) you run today and a separately-hosted Variant page (B) — and it sends synthetic shoppers to each.
Because Squoosh compares both versions as a shopper would see them, almost any change you can serve at a URL is testable. For example:
- A redesigned product page against the current one.
- A new page layout or section order.
- Revised copy, headlines, or calls to action.
- A hosted preview of a change you haven't published yet — point version B at the preview link.
To set this up:
- In the wizard, enter your current page as Control page (A).
- Under How do you ship variants?, keep Two URLs selected.
- Enter the variant address in Variant page (B). It must be a valid URL and different from the control.
The variant must be a distinct page from the control — Squoosh doesn't run a page against itself.
For step-by-step setup, see Run an experiment.
If you run variants through a testing tool¶
If your variant lives on the same URL and is shown by an A/B platform or a feature flag, you can tell Squoosh which tool you use instead of supplying a second URL. You still build version B; these modes only tell Squoosh how to reach it.
| Mode | For teams using | What you provide |
|---|---|---|
| VWO / Optimizely / AB Tasty | A client-side testing tool that buckets visitors on one URL | The vendor, plus either two preview URLs or the experiment and variation IDs |
| Feature flag | A server-side flag (LaunchDarkly, Statsig, Split, or a custom cookie, header, or query parameter) | The flag mechanism and the values that mean control and variant |
These modes are early and still being validated. If you use one of these platforms, start with Two URLs where you can, and contact support@squoosh.ai for help wiring up a vendor or feature-flag setup.
Note
Squoosh compares the pages as they render. A difference that only exists on your server, or one that surfaces deep inside an interaction, may not register. If Squoosh captures both versions but can't tell them apart, it warns you above the verdict rather than reporting a result you can't trust — recheck the URLs, IDs, or flag values you entered for the variant.
Goals you can optimize¶
Each experiment optimizes one conversion goal, set as the Primary goal:
- Add to Cart
- Checkout
Squoosh measures how often shoppers reach that goal on each version and reports the difference. See How Squoosh decides a winner for how that result becomes a verdict.