Every few months a client sends me a screenshot of a red PageSpeed score and asks if the site is broken. Usually it isn't. But the score is pointing at something real: pages that feel slow, jump around, or ignore a tap for half a second.
Core Web Vitals are Google's attempt to put numbers on those feelings. Here is how I think about them, and the fixes I reach for first.
The three numbers that matter
- Largest Contentful Paint (LCP): how long until the biggest thing on screen, usually a hero image or headline, is visible. Aim for under 2.5 seconds.
- Interaction to Next Paint (INP): how quickly the page responds after someone taps or clicks. Under 200 milliseconds feels instant.
- Cumulative Layout Shift (CLS): how much the layout jumps while it loads. Keep it under 0.1.
Fixing a slow LCP
Nine times out of ten, a slow LCP is a large hero image that the browser discovers too late. Serve it in a modern format like WebP or AVIF, size it for the screen it is shown on, and never lazy-load it. If it is a background image set in CSS, consider moving it into an img tag so the browser can find it sooner.
Fonts are the other common culprit. Self-host them, preload the one or two weights you use above the fold, and use font-display: swap so text appears immediately.
Fixing a sluggish INP
INP is almost always about JavaScript doing too much on the main thread. Look for long tasks in the Performance panel. Defer scripts that aren't needed for the first interaction, break big loops into smaller chunks, and be suspicious of third-party widgets. A single chat plugin can cost more than your entire theme.
Fixing layout shift
Give every image and video explicit width and height attributes, or an aspect-ratio in CSS. Reserve space for ads and embeds before they load. And never insert banners above existing content after the page has rendered.
Measure on a real mid-range phone, not your laptop. That is where your visitors are.
Where to start
Run the page through PageSpeed Insights, but trust the field data at the top over the lab score below it. Fix the biggest problem first, deploy, and wait a few weeks for the real-user numbers to catch up. Small, steady improvements beat one heroic rewrite every time.