Core Web Vitals 2026: LCP, CLS & INP Explained
What LCP, CLS, and INP mean, how they're measured, what scores you need to rank, and the fastest ways to fix each metric.
Three measurements, three thresholds. Below them is green, above is red, and in between is "needs improvement" — which in practice means red.
LCP — how long until something important appears
Target: under 2.5 seconds.
The time until the largest element in the first viewport renders. It's almost always an image or a large heading.
What breaks it, by frequency:
- an uncompressed hero image in an old format
- an image lazy-loaded when it should be prioritised
- a slow server first response
- render-blocking fonts
What helps: modern formats (WebP, AVIF), explicit dimensions, fetchpriority="high" on the hero image, and no loading="lazy" on it.
INP — how quickly it responds to touch
Target: under 200 milliseconds.
It replaced FID and it's stricter: it measures every interaction, not just the first. A button that feels "sticky" when pressed is bad INP.
The cause is almost always JavaScript blocking the main thread. What helps: fewer third-party scripts, code split into chunks, heavy work moved off the main thread.
Tracking scripts are suspect number one. A tag manager with 15 active tags can push you over the threshold on its own.
CLS — how much the layout jumps
Target: under 0.1.
Elements that move after the page looks loaded. The classic causes:
- images without
widthandheight - ads and embeds inserted into the flow
- fonts that change line height on load
- banners appearing above content
The fix is the same everywhere: reserve the space in advance.
How to measure properly
PageSpeed Insights gives you two data sets. The top one, "field data", comes from real users and is what counts for ranking. The bottom one, "lab data", is a simulated test — useful for debugging, not for judging.
If you don't have enough visitors for field data, you've also learned something about where the real problem is.
How much it matters for ranking
Directly, a little. It's a tiebreaker between otherwise equal pages. Indirectly, a lot: a page that takes 5 seconds loses visitors before they read anything.
How many hours a week do you lose to repetitive work?
In 20 minutes on Zoom we work out which of them can run on their own. You leave with the answer, whether we work together or not.
Free call, no obligations. You leave with a concrete list of problems and automation opportunities, whether or not you choose to work with us.