Frequently asked questions¶
Short answers to the questions customers ask most. Each one links to a fuller page.
Squoosh basics¶
What is Squoosh?¶
Squoosh runs AI synthetic shoppers against two versions of a web page and tells you which one converts better. An A/B test reports the selected synthetic sample's outcome and supporting evidence; duration depends on sample size, page behavior and capacity. See What is Squoosh?.
Do I provide the variant, or does Squoosh make it?¶
You provide it. Squoosh does not generate or write the variant — you build version B, and Squoosh tests your real change. In the wizard you give Squoosh the Variant page (B) so its shoppers and reviewer land on the right version. See How you ship variants.
What can I test?¶
Any page you can reach by URL, compared as two versions: a Control page (A) and a Variant page (B). The proven path is Two URLs — one address per version. If your variant lives on a single URL behind a testing tool (Convert, VWO, Optimizely, AB Tasty) or a feature flag (LaunchDarkly, Statsig, Split, or your own cookie, header or query parameter), those modes are supported too — Convert, VWO, Optimizely and feature flags are verified, AB Tasty is still in beta; start with How you ship variants.
How long does a test take?¶
An A/B test runs at the Rigor level you choose; higher rigor runs more paired shoppers, which takes longer and uses more credits. Standard and Medium finish in minutes; the largest levels can take hours or days. The launch footer shows Priority instead of a fixed time estimate. Results arrive as shoppers finish. See Experiment lifecycle and stopping.
What is a Conversion Report?¶
A Conversion Report examines a single page for friction in the path to conversion, rather than comparing two versions. Instead of a large statistical sample, it uses a named panel of 10 synthetic shoppers spread across the archetype styles you choose (Budget hunter, Deep researcher, Impulse buyer, Brand loyalist, Skeptic, Gift buyer). You can watch any individual shopper from the panel. There is no winner, lift, or confidence — the output is a ranked leak report with per-shopper evidence. See Conversion Reports.
Credits and limits¶
What is a credit? What is a test?¶
A credit is the unit you spend to run a test. A test is one experiment run — an A/B test or a Conversion Report. The sidebar's N tests left estimate uses the default 30-credit A/B price and your remaining monthly test allowance. Larger samples cost more. Crawls share the credit balance without using a monthly test slot. See Credits and tests.
How much does a test cost?¶
The wizard shows the exact cost before you launch and your balance after. An A/B test's cost rises with its rigor level. A Conversion Report costs 10 credits for Standard or 30 for Deep. See Credits and tests.
How do I start a Site Crawler capture?¶
Open Site Crawler in the property sidebar, click New crawl, select Default Property or a Variant URL, and set the page and time limits. Click Review crawl, then Start crawl. The default reservation is 200 credits; unused credits return after the crawl ends. See Site Crawler.
Do crawls replace the browser sessions in experiments?¶
For the higher rigor levels, yes. With Site Crawler turned on in Settings → Features, A/B tests at High rigor and above run their shoppers on the saved captures of both pages instead of live browser sessions; Standard and Medium still use live sessions. See Site Crawler.
How do I get more credits or raise my limits?¶
Contact the Squoosh team at support@squoosh.ai to top up credits or change your plan — self-serve billing isn't available yet. Plans cap how many tests you can run each month (5 by default) and how many properties you can have (1 by default); hitting a cap prompts Upgrade your plan. See Credits and billing.
Reading results¶
Why does Squoosh call a winner at 80% confidence?¶
The standard verdict uses an 80% confidence bar to balance early decisions and uncertainty. Runs at High rigor or above in Two URLs mode report a measured lift, labeled Village conversion lift, with sampling intervals instead of this confidence meter. See How Squoosh decides a winner.
What does the lift number mean?¶
The lift is the measured difference in conversion between your two shopper groups — how much more (or less) often the variant's shoppers converted than the original's, as a relative percent. It's the shoppers' actual result, shown on the same scale as the per-arm rates in the evidence table; Squoosh does not discount or scale it down. See Reading lift and confidence.
What's the difference between estimated and measured lift?¶
The rigor level sets the method. At Standard or Medium the lift is estimated: Squoosh's AI reviewer compares the two pages and grounds its read in what the shoppers did, so treat it as a directional read rather than a statistically conclusive result. At High and above in Two URLs mode the lift is measured: Squoosh counts verified conversions on each page across paired synthetic shoppers and reports the difference with a 95% sampling interval. See How Squoosh decides a winner.
Can the original win?¶
Yes. Keep the original is a real, first-class outcome — a change that costs you conversions is worth knowing about before you ship it. There is no ship or deploy button; rolling out the variant is always your call. See How Squoosh decides a winner.
Why does the verdict disagree with the table?¶
In the standard view, the headline can reflect the reviewer's read while the table shows raw shopper rates. The measured view that High rigor and above produce does not use that reviewer override; its reported rates account for the covered audience mix. See Reading lift and confidence.
Synthetic shoppers and accuracy¶
How are synthetic shoppers calibrated?¶
Squoosh reads your real traffic mix — device, traffic source, and geography — from your connected analytics source and builds a shopper pool whose distribution matches it. The Shopper match score shows how closely the synthetic pool reproduces your recorded traffic, reported as "N% across K calibrated attributes" so it only ever counts the attributes actually matched to your data. Behavioral traits stay modeled, not measured. See Synthetic shoppers.
Do I need Shopify?¶
No. Connect Shopify or Google Analytics to calibrate your shoppers to your real traffic — either source works. Without a connected source, Squoosh can still generate a pool from a general e-commerce default, but the match score is unavailable until you connect one. See Connect Shopify or Connect Google Analytics.
How do I keep synthetic traffic out of my analytics?¶
Squoosh marks its synthetic shoppers so you can exclude them from your analytics, and you control how that traffic is handled in Settings → Analytics & Privacy. See Keep synthetic traffic out of your analytics.
When does a run stop on its own?¶
A/B experiments finalize once their shopper sessions have finished or otherwise ended. They do not currently stop when an interim confidence score crosses a threshold. You can stop an active run yourself; doing so does not produce a completed winner. See Experiment lifecycle and stopping.