Three numbers, each measuring something a visitor actually feels. That is the useful part. They are not abstract scores, they map onto complaints you have already heard from your own users.
LCP, or how long until the page looks loaded
Largest Contentful Paint measures when the biggest visible element finishes rendering, usually a hero image or a headline. Under 2.5 seconds passes.
What breaks it, almost always: an oversized hero image. A 4MB photograph scaled down in the browser still downloads 4MB. Compressing images is the highest-return fix on most sites. On this one it took the homepage from 29.5MB to 1.2MB.
INP, or how long until the page responds to you
Interaction to Next Paint replaced First Input Delay in 2024. It measures the lag between a tap and something visibly happening. Under 200 milliseconds passes.
What breaks it: JavaScript. Chat widgets, analytics stacked on analytics, carousel libraries, cookie banners that load before anything else. Each one occupies the main thread, and while it is busy your buttons do nothing.
CLS, or how much the page jumps around
Cumulative Layout Shift measures content moving after it appears. Under 0.1 passes. Everyone has experienced this one: you go to tap a link, something loads above it, and you tap the wrong thing.
What breaks it: images and embeds without declared dimensions, web fonts swapping in at a different size, and anything injected at the top of the page after load.
- 2.5s
- LCP — loading
- 200ms
- INP — responsiveness
- 0.1
- CLS — visual stability
Where to measure
- PageSpeed Insights. Free, and shows both lab data and real Chrome user data. The real data is what Google ranks on.
- Search Console's Core Web Vitals report. Groups your URLs by status, which is how you find the failing template rather than the one failing page.
- Lighthouse in Chrome DevTools. Fast to iterate against, but it is a simulation. Do not celebrate a lab score.
Lab scores and field data disagree constantly. Field data is what counts, and it lags by 28 days, so fixes take a month to show. Do not rebuild twice because week one looked flat.
What a passing site actually scores
So this is not theory: mexhomesanmiguel.com, a hand-coded real estate site I built and optimize, measured 95 performance, 96 accessibility, 100 best practices, and 100 SEO on PageSpeed Insights in August 2026. The screenshot and the build details are in the case study. Nothing exotic got it there — the five fixes below, applied before launch instead of after.
The fixes worth doing first
- 01Compress and resize every image. Serve WebP. Set explicit width and height.
- 02Remove scripts you are not using. Most sites carry at least one dead tag.
- 03Self-host fonts and preload them, or accept the system stack.
- 04Reserve space for anything that loads late: embeds, ads, chat widgets.
- 05Stop shipping JavaScript for things CSS already does. Most animation qualifies.
Those five cover the majority of failing sites. Past that you are into architecture, which is a rebuild conversation rather than a fix list — see WordPress or custom-coded. If your site was built by someone else and the scores will not move, that is what the monthly SEO service exists for.
