"The site is slow" is not a diagnosis. It can mean the page takes four seconds to show anything, or that it shows quickly and then ignores your first tap, or that the text jumps down just as you were about to read it. Those are three different problems with three different fixes.
Google gave them names and thresholds, and the set is called Core Web Vitals. They are a useful way to stop arguing about whether a site feels slow and start fixing the part that is.
The three numbers
Largest Contentful Paint (LCP) is loading: how long until the biggest thing in the first screen, usually a hero image or a headline, is on screen. Good is 2.5 seconds or less.
Interaction to Next Paint (INP) is responsiveness: how long the page takes to react after a tap or a click. Good is 200 milliseconds or less. INP became a Core Web Vital in 2024 and replaced the older First Input Delay, so advice that still talks about FID is out of date.
Cumulative Layout Shift (CLS) is visual stability: how much the page jumps around while it loads. Good is 0.1 or less. It is a score, not a time.
A page counts as passing when it meets all three at the 75th percentile of real visits, measured separately for phones and desktops. In other words, the slowest quarter of your visitors sets the grade, not your fast office connection.
Measure first, and measure real visitors
PageSpeed Insights shows two kinds of data and people mix them up all the time. The field data is what real Chrome users experienced over the last 28 days. It only appears when the page has enough traffic. The lab data is a single simulated test run on a throttled device. It is good for finding causes and poor for judging the result.
Search Console's Core Web Vitals report groups your pages by problem, which tells you whether you have one slow template or a hundred separate issues. Usually it is the first. A site with four page templates has four things to fix, not four hundred.
Test on a phone profile. Most of your visitors are on one, and a page that is fine on a laptop can miss the LCP target by a wide margin on a mid-range handset on mobile data.
If LCP is poor
Find out what the largest element is. The PageSpeed report names it. In most cases it is an image, and the fixes are dull and effective. Serve it at the size it is shown, not at the size the camera produced. Use a modern format such as WebP or AVIF. Do not lazy-load it, because lazy loading is for images below the fold and it delays the one image the visitor is waiting for. Tell the browser early that it matters, with a preload or a priority hint.
If the largest element is text, the culprit is usually the font. Host the fonts yourself, load only the weights you use, and make sure the text is visible while the font arrives.
If neither helps, look at the server. A slow first response caps everything that follows. Caching, a CDN and a lighter page template are the usual remedies.
If INP is poor
Slow responses almost always mean the browser is busy running JavaScript when the visitor taps. The cure is less script, not faster script. Remove the widgets nobody uses, and be suspicious of third-party tags: chat bubbles, tag managers, heat-map recorders, social buttons. Each looks small and together they can dominate the page.
Anything you keep should load late, after the page is usable, and long tasks should be split so the browser can answer a tap in between. This is also the area where an animated, script-heavy design can get you into trouble, so test animations on a mid-range phone, not on the designer's laptop.
If CLS is poor
Layout shift has a short list of causes. Images and videos without width and height attributes, so the browser cannot reserve space. Banners and embeds that appear after the text and push it down, and cookie banners are a classic offender when they are inserted into the page flow instead of overlaid. Fonts that swap and change the size of the text. Adverts of unknown height.
Reserve the space in the markup, overlay anything that appears late instead of inserting it, and the score usually drops below 0.1 within one release.
Does it matter for rankings?
Google describes these metrics as part of its page experience signals, and it is clear that a fast page with thin content will not outrank a slower page that answers the question better. So do not expect a perfect score to lift a mediocre page.
The better argument is commercial. People abandon pages that feel slow, and the cost lands on enquiries and sales before it ever shows up in a ranking report. Fix the worst template first, re-measure after 28 days of field data, and then move to the next one.