The three metrics that matter
LCP (Largest Contentful Paint) — good at ≤ 2.5s. Usually the hero image, headline block, or main product media.
INP (Interaction to Next Paint) — good at ≤ 200ms. Measures responsiveness after clicks and taps; FID is obsolete.
CLS (Cumulative Layout Shift) — good at ≤ 0.1. Reserve space for images, ads, and fonts so the layout does not jump.
Field data vs lab data
Google evaluates Core Web Vitals primarily with real-user (CrUX / Search Console) data at the 75th percentile over ~28 days.
PageSpeed Insights and Lighthouse are synthetic labs — useful for reproducing a slow path. A Lighthouse 100 does not mean Core Web Vitals pass (lab has no field INP; ranking uses CrUX).
Lead LCP fixes with TTFB and render delay; most poor-LCP origins spend under 10% of p75 LCP time downloading the LCP image.
Highest-ROI fixes
LCP: cut document TTFB first, then preload the LCP image, avoid lazy heroes, defer non-critical CSS/JS.
INP: cut long JavaScript tasks, defer non-critical scripts, audit third-party tags.
CLS: set width/height (or aspect-ratio) on media, avoid inserting banners above existing content.