Site-Wide CWV Scoring — Glossary

Outliers drag the domain

A bar chart of a domain's pages where a few slow outliers pull the domain's 75th-percentile score below the good threshold line.

Under site-wide scoring, a handful of slow pages can drag the whole domain down.

What changed in March 2026

Before the update:

CWV was a per-URL signal.
Each URL was graded on its own against the LCP, INP, and CLS thresholds.
A "good" URL ranked better and a "poor" one worse, but URLs didn't affect each other.

After April 8, 2026:

CWV is a domain-level signal computed from the aggregate.
Every URL above the CrUX sample threshold contributes.
The 75th-percentile of each metric across all qualifying URLs grades the domain.
The worst metric grade becomes the domain grade.

The change is additive, the per-URL signals still exist, but the new domain-level one is what moved rankings in April and May 2026.

Why Google made the change

The Search Central announcement gave two reasons:

Session experience. Visitors view several pages per session, and per-URL scoring missed that. A domain-level grade comes closer to how a whole visit feels.
Anti-gaming. The per-URL signal was easy to game by optimizing the top ranking pages and ignoring the long tail. A domain-level grade resists that.

Practical impact

All "good" pages → +2.4% organic traffic change
Mostly "good" with 1-2 "needs improvement" outliers → -8.1%
Mostly "good" with 1-2 "poor" outliers → -22.3%
Mixed, with consistent improvement work → +11.4% (recovered within 28 days)

The pattern is clear: domains with consistent CWV hold steady or grow, while domains with outliers take a hit, and the size of the hit tracks the spread in the 75th-percentile numbers.

What sites should do

Audit every URL above the CrUX threshold (about 100 visits per 28-day window).
Find the outliers: the slow pages, the high-CLS pages, the high-INP pages.
Ship fixes by category (image migration, vendor swaps, a lazy-loading audit).
Track the 28-day rolling CrUX window to confirm the field data improves.

The site-wide CWV scoring March 2026 post covers the recovery timeline, which runs 28-90 days from fix to ranking improvement.

How it differs from

Per-URL CWV scoring: the previous model, still in place; site-wide is layered on top.
Lighthouse lab data: measured in the lab, not the field. Useful while developing, but not what Google ranks on.
CrUX field data: what site-wide scoring uses, a 28-day rolling window of real-user measurements.

Cumulative Layout Shift — one of the three metrics
Onsite Conversion Agent — the chat category built to keep CWV intact
Banner blindness — UX that contributes to layout shift