Core Web Vitals: what LCP, INP, and CLS measure

A page loads. The hero photo appears. A user taps a button and waits. An ad slams in from above and the text jumps — the user taps the wrong link. Every chapter in this pillar — the event loop, the render pipeline, V8, hydration — produces exactly this kind of experience. Three field metrics measure it, and Google ranks pages by them.

The three metrics — one question each.

Each Core Web Vital answers one question a real user has the moment they land on a page:

• LCP — Largest Contentful Paint — did the main content show up fast? Good: ≤2.5 s.
• INP — Interaction to Next Paint — does the page respond when I tap or type? Good: ≤200 ms.
• CLS — Cumulative Layout Shift — did things stay put, or did a button jump out from under my finger? Good: ≤0.1 (unitless score).

Each threshold is measured at the 75th percentile of real users. That matters: optimising only for fast devices hides the slow quarter of your audience, and that quarter is exactly what the score reflects.

The shop metaphor.

Walking into a physical shop is the clearest frame for these three metrics. LCP is how long before you can see what the shop sells — if the shelves are still being stocked, you wait at the door. INP is whether the clerk answers right away when you ask a question — a good shop responds immediately, a bad one makes you repeat yourself. CLS is whether the floor stays put — imagine reaching for an item on a shelf and the shelf sliding sideways just as your hand arrives, so you grab the wrong thing. A good page is a good shop.

Where the numbers come from — and why it matters.

Core Web Vitals as Google scores them come from real Chrome users in the field, not from a developer’s test run. The public dataset is called CrUX — the Chrome User Experience Report. It aggregates visits from opted-in Chrome users, and the field p75 from CrUX is what Search ranking actually uses. A great Lighthouse score on your laptop and poor field vitals are entirely compatible — and the field number is the one that counts.

The practical consequence: every performance fix in this pillar ultimately lives or dies on whether it moves the field p75 for real users. Lab tools (Lighthouse, DevTools) are for debugging and catching regressions; CrUX is the verdict.