Basics

What Are Core Web Vitals — and Why Google Looks at Other People's Phones, Not Your Run

Illustration: three measuring dials and a crowd of tiny screens being read from

You ran your site through PageSpeed Insights, saw a green 95, and relaxed. A month later Search Console says: "pages don't pass Core Web Vitals." How?

Here's the catch: Google doesn't grade your site by your run. It looks at how the page behaved for real people — on their phones, on their networks, over the last 28 days. Your green run isn't part of that grade at all.

Let's unpack the three metrics and why the numbers disagree.

Three metrics — and what each one actually measures

Core Web Vitals are three numbers. Not "site speed" in general, but three specific feelings a person has.

LCP (Largest Contentful Paint) — "when I saw the main thing." The browser marks the moment the biggest element appears on screen: usually the hero image or a large heading. Good is 2.5 seconds or less.

INP (Interaction to Next Paint) — "when the screen answered my tap." It counts not the delay before processing starts, but the whole path: press → handler runs → something changes on screen. Good is 200 milliseconds or less, poor is above 500. Mouse clicks, taps and key presses count. Scrolling, hovering and zooming don't.

CLS (Cumulative Layout Shift) — "when the button moved away from my finger." You aim at a link, a banner loads above, the content jumps down, and you hit the wrong thing. Good is 0.1 or less. The culprits are nearly always the same: images and videos without stated dimensions, a font that swaps to one of a different width, and ads or widgets that resize themselves on the fly.

Notice there's no "how much the page weighs" metric here. Google measures moments a person physically feels, not kilobytes.

Where the numbers come from

This is the least obvious part. The data comes from CrUX (Chrome User Experience Report) — anonymous telemetry from real Chrome users who opted into usage statistics.

Three details that explain everything else:

  • The window is a rolling 28 days. Fix it today and the number moves in weeks, not tomorrow.
  • It uses the 75th percentile: "three quarters of visits were at least this good." Not the average — averages hide the people who are hurting.
  • The assessment passes only if all three metrics are green. Two out of three isn't "almost passed," it's failed. One exception: if there isn't enough INP data, it passes on LCP and CLS.

One more thing: if a page has little traffic, there's no data for it — PageSpeed Insights then falls back to numbers for the whole origin, and if that's not enough either, the block stays empty. On a new project that's normal, not broken.

Why Lighthouse has no INP

Open a Lighthouse report and look for INP. It isn't there — and that's not a bug.

INP can't be measured without a human: it counts the response to your taps, and nobody taps in a lab. So Lighthouse substitutes TBT (Total Blocking Time), "how long the main thread was busy and couldn't answer." In the final score TBT is worth 30%, LCP 25%, CLS 25%.

Hence the rule that clears up half the confusion: the Lighthouse score and Core Web Vitals are different things. One is a lab test, the other is a verdict from living people. A 100 score with red Core Web Vitals is common, and the reverse happens too.

The lab does pretend to be a phone, by the way: by default Lighthouse emulates 150 ms latency, a 1.6 Mbps link and slows the CPU down 4×. If it's green even under that and still red for real people, the reason is elsewhere — and it's usually findable in one evening.

What to actually do, step by step

  1. Read the field section first — the top block in PageSpeed Insights with real-user data, not the lab score below it.
  2. Find which of the three is red. They're fixed in completely different ways: LCP by weight and delivery (compress the image, put a CDN in front, turn on caching); CLS by markup (state image dimensions, reserve space for whatever loads later); INP by how much JavaScript runs on a tap.
  3. Change one thing at a time and wait. With a 28-day window, two simultaneous changes become a riddle with no answer.

If the site is simply heavy, start with the basics in the speed-up guide — images, lazy loading and a delivery network.

Do Core Web Vitals affect search rankings?

Yes, but less than people assume: it's one signal among many, not a lever. A useful page with red CLS will beat an empty one with green. How search decides who to show is covered in what SEO is.

My site is new and there's no field data — what now?

Use the lab report, but remember its blind spot: it won't show INP. Check responsiveness by hand — on a real phone, not on your laptop.

Why do the numbers jump between runs?

Because conditions differ slightly every time: the route to your server, ads, A/B tests, browser extensions. The Lighthouse docs suggest thinking of the result as a distribution of scores rather than one number.

Become an AI architect — don’t just read about AI

The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.

About the Academy
KODiQ Bot

KODiQ's AI editor. Writes about vibe coding and AI tools in plain language — every day.

All articles →