Reading the verdict¶
The experiment results page shows the outcome and the evidence behind it. The standard view has a verdict, Why it won and the evidence. Experiments run at High rigor or above in Two URLs mode report a measured lift and show Village conversion lift instead; read the section below if that label appears. Runs at Standard or Medium rigor report an estimated lift in the standard view.
Your experiment compares two pages: Control page (A), your current page, and Variant page (B), the version you supplied. The results page reports which one converted better on your goal.
Conversion-evidence results¶
This result mode applies automatically when you launch at High rigor or above in Two URLs mode. Standard and Medium, the vendor and feature-flag variant modes and demo workspaces use the standard estimated verdict. Existing experiments keep the result mode they started with.
When the page says Village conversion lift, the numbers come from recorded synthetic-shopper conversions for the declared event and covered audience. Shopper feature feedback explains experience without changing conversion lift. These are not measurements of human-customer conversions.
| Field | How to read it |
|---|---|
| Control conversion rate and Variant conversion rate | Estimated rates for the covered audience, based on accepted paired shopper observations and the audience mix. |
| Relative lift | How much the variant rate changed relative to the control rate. It is unavailable if no control shoppers converted. |
| Absolute rate change | The variant rate minus the control rate, in percentage points. |
| Sampling intervals | Uncertainty from the synthetic sample under the experiment's assumptions. An interval is not a probability that a version wins or a measure of accuracy against human customers. |
| Evidence coverage | How much of the target audience and planned paired sample the accepted evidence covers. |
| Executed pairs | Planned, attempted, accepted, excluded, unavailable and cancelled pairs, plus recorded conversion totals. |
| What shoppers experienced | Recorded reactions relative to the shopper's expectations. Missing feedback does not turn a recorded conversion into a failure or change the lift. |
This view can report Variant converts more shoppers, Control converts more shoppers or No clear conversion winner. It does not display the standard 80% confidence meter. Raw conversion totals can differ from the audience-weighted rates; read the coverage alongside the rates.
The 95% interval on the absolute rate change is estimated from how often the paired shoppers disagreed (one converting, the other not). At High rigor it is at most ±20 percentage points and typically about ±12; it narrows at higher levels and when shoppers agree more. A winner also needs the disagreeing pairs to lean one way more than chance would explain, so a handful of one-sided pairs never calls a result. No clear conversion winner means the difference did not clear that bar, not that the two pages perform the same. A method line under the headline restates the setup, Measured conversions · High rigor, followed by weighted to GA4 traffic when a fresh Google Analytics traffic snapshot weighted the rates, or panel-weighted when the rates were weighted evenly across your calibrated shoppers because no such snapshot was available.
If the page reports Insufficient conversion evidence, Village result unavailable or Village result withdrawn, do not treat that as zero lift or a completed winner. Read the accompanying reason. An unavailable result is not replaced by an expert reviewer's score. If the record cannot be verified, contact support with the experiment ID.
The sections below describe the standard results view.
The verdict band¶
The band at the top states the outcome in one line, with the lift and the confidence beside it.
The verdict band: the outcome, the lift, the confidence, and the change that was tested.
The headline is one of three outcomes:
| Headline | What it means | What to do |
|---|---|---|
| Variant B wins | Variant page (B) beat the control on your goal. | Ship version B. |
| Keep the original | Variant page (B) did worse than the control. | Keep your current page live. |
| Too close to call | Neither page separated from the other. | Re-test with a sharper change. |
Keep the original is a real result, not a failure. Knowing a change would cost you conversions is worth the test. A small chip next to the experiment name restates the outcome.
Beside the headline, the band shows two numbers:
- Lift — how much version B changed your conversion rate versus the control, as a signed percent, with a 95% CI range under it.
- Confidence — how sure Squoosh is that the result is real, on a meter marked at the 80% bar Squoosh uses to call a winner.
The lift Squoosh shows is the measured difference between your two shopper groups — the variant's conversion rate versus the control's — not a discounted estimate. For how to read both numbers, see Reading lift and confidence. For how the two combine into the outcome, see How Squoosh decides a winner.
The band's action buttons let you watch the shopper replays or re-run the test. There is no ship button — see You decide.
Note
A higher raw number in the table doesn't always win. When version B's conversion rate ran higher but the change didn't earn that lift, the verdict still says Keep the original and explains why. The headline reflects Squoosh's read, not the raw rate alone.
Why it won¶
Directly under the verdict, the Why it won section explains the result the way an experienced conversion analyst would — read the outcome, then read why. Every completed experiment gets one; an inconclusive run reads What we learned instead and never invents a winner that isn't there. It has three parts:
- Squoosh's read — a short, plain-language summary of what separated the two versions: what changed, what shoppers noticed, and why it moved (or didn't move) the decision.
- Issues they hit — the friction shoppers ran into, grouped by area (for example Trust & Security, Navigation, Product), so you can see what to address next.
- In their words — a rail of quote cards showing individual shoppers in their own words. Each card carries the shopper's name and persona, a verbatim quote, and whether they Added to cart or Did not convert. Squoosh surfaces the shoppers who decided differently between the two pages, because those decision-flippers explain the result. Click any card to open a paired replay — the same shopper's session on Control page (A) and Variant page (B) side by side, so you can watch where the two versions sent them in different directions.
The evidence¶
Below "Why it won," the evidence section — How we know — the proof — shows the numbers behind the verdict.
Result detail¶
A table compares the two pages side by side. Each row is one version — Control and B · {your variant name} — and shows that version's conversion rate, its change versus the control, and a result marker. The winning row is enlarged, and the Confidence is restated below it.
The result markers are short chips:
| Chip | Meaning |
|---|---|
| Winner | This version won the experiment. |
| Flat | No meaningful difference from the control. |
| Lost | This version converted lower than the control. |
| Control | The baseline that version B is measured against. |
Lift by segment¶
A bar per audience segment shows each segment's change versus the control. A change can help one segment and hurt another, so the bars can point in opposite directions. They blend back to the single headline lift — use them to see who a change worked for, not as separate verdicts.
You decide¶
Squoosh never applies a change to your site. There is no ship or deploy button. The verdict is a recommendation backed by evidence; rolling out version B is your call.