Core Web Vitals

The three metrics, and what counts as good

• Largest Contentful Paint (LCP) — how long until the largest element in the viewport is painted, usually a hero image or headline. Good is 2.5 seconds or less; over 4 seconds is poor. It is the closest thing to a measurement of "when did the page feel like it arrived".
• Interaction to Next Paint (INP) — the latency between a tap, click or keypress and the next frame the browser paints, taken across the whole visit rather than the first interaction. Good is 200 milliseconds or less; over 500 is poor. INP replaced First Input Delay in March 2024, because almost every page passed FID trivially.
• Cumulative Layout Shift (CLS) — how much visible content jumps while the page loads, scored as a unitless number. Good is 0.1 or less; over 0.25 is poor. The usual culprits are images with no dimensions, late-loading fonts, and banners injected above content.

The scores Google uses are field data: real sessions from real Chrome users, collected in the Chrome User Experience Report over a rolling 28-day window. Lab tools like Lighthouse and the lab half of PageSpeed Insights simulate one visit on one synthetic device — invaluable for debugging, but they are not the number being assessed. A site with too little traffic may have no field data at all, in which case it simply is not assessed.

How much they matter for rankings

Less than the effort spent on them usually assumes. Core Web Vitals are one input to page experience, which Google has consistently described as a modest signal that can separate comparable results rather than promote a weak one. A fast page does not outrank a better answer. If a page gets no traffic, performance is rarely the reason — the diagnosis usually lies further upstream, in indexing, intent match, or the simple absence of anything worth ranking.

Where they reliably pay is after the click. A visitor who leaves before the page paints was never a ranking problem, and a layout that shifts under a thumb costs conversions on exactly the device where you can least afford it. Treat Core Web Vitals as conversion-rate work that also happens to be a ranking tiebreaker, and you will prioritise them correctly: fix genuinely slow pages, and stop optimising a passing score for its own sake.

The trap we hit measuring our own LCP

LCP measures what the browser records, not what a human sees — and the gap between those two is where the metric misleads. On our own homepage at AgentCeres — the AI Growth Officer at agentceres.com — the hero headline sat behind a scroll-reveal animation that only ran after hydration. PageSpeed Insights reported a 6.6-second mobile LCP with element render delay at roughly 98% of it: the page had loaded long before, and the headline was simply invisible until JavaScript caught up. Visitors with JavaScript blocked never saw it at all.

Fixing that surfaced a stranger problem. We swapped the JavaScript reveal for a pure CSS fade, and the measured LCP got worse rather than better — 8.1 seconds when we measured it live on 12 July 2026, now anchored on the small logo in the navigation bar. A fade from zero opacity runs on the compositor and never repaints the element, so Chrome emits no LCP candidate for it, and the metric quietly moved to the next thing that did paint. The fix was to give the headline no entrance animation whatsoever. The lesson generalises: anything you animate into view can stop being your LCP element, and the score you then read describes some other part of the page.