Core Web Vitals: why speed matters for your SEO
Learn how LCP, INP, and CLS impact your site's speed and SEO. Discover concrete examples and mobile fixes to improve user experience and search rankings.

Sur cette page6 sections
You have probably already tried to press a button that moved at the last second. An image loads, the page shifts down, and your finger opens something else. Or the menu seems unresponsive: you press a second time, and then everything triggers at once.
When we talk about a slow site, we often confuse these situations with the time it takes to display the page. However, you can have a site that appears quickly but becomes frustrating as soon as you try to use it. Core Web Vitals serve to put names and measurements on these sensations, rather than just saying "it works fine for me".
INP, or Interaction to Next Paint, focuses on the visible response after an interaction. You touch the menu: how long before you see something happen? It looks at interactions throughout the visit, not just the first click. A browser busy with heavy processing can leave your gesture waiting even if the server has already sent the entire page.
CLS, or Cumulative Layout Shift, measures unexpected layout shifts. This is the case of the button that slips away from your finger because a block takes its place. Voluntarily scrolling a page is not the same thing: here we are looking for position changes that the user did not request.
The benchmarks published by Google on web.dev are an LCP of 2.5 seconds or less, an INP of 200 milliseconds or less, and a CLS of 0.1 or less. They are evaluated at the 75th percentile, separately for mobile and desktop: for each indicator, at least 75% of measured visits must stay within the recommended threshold. So, it's not your best attempt that decides if the experience is good.
A fast page that doesn't explain your service is still a page that doesn't explain your service. If your visitor is looking for a repair in their town and finds a vague company presentation, gaining a few tenths of a second won't answer their question.
On the other hand, it's a shame to work on SEO, attract the right people, and then offer them a journey that prevents them from contacting you. Performance is primarily about not wasting that visit. This is why I prefer to start with service pages and the contact form rather than a secondary page that is easy to push to 100.
PageSpeed Insights presents a simulated test and, when enough data is available, metrics from real visits. The test helps reproduce the causes of a problem. Field data shows what happened for visitors over a period, under their own conditions.
A small site may lack field data at the page level. This is neither a perfect result nor a failure: it is unavailable information. And if the tool displays results for the entire domain, you need to know this before presenting them as the results for the form you just tested.
The Lighthouse score is also not a complete measurement of your visitors' real interactions. An automated loading test doesn't spend several minutes opening the menu and filling out your fields. You must keep both perspectives: the technical diagnosis and the manual user journey.
A cache like Varnish can speed up the response. It doesn't automatically make a multi-megabyte photo lightweight, and it doesn't stop a script from blocking the menu. Changing hosting without looking at what happens after the response is sometimes like speeding up the delivery of a package that no one can unwrap.
For LCP, we look specifically at server delay and what delays the main image or text. For INP, we examine the work required of the browser at the time of interactions. For CLS, we reserve space for images and content that arrive later, so they don't push the rest of the page around.
We can then address one cause at a time: resizing an image to a suitable size, removing an unused widget, limiting scripts loaded at the start, or planning the space occupied by a visual. Then we replay the same scenario on mobile. If you change everything simultaneously, it becomes difficult to know what actually improved the visit.
You also need to check that you haven't sacrificed what was useful just to gain points. An empty page loads fast. That doesn't make it a good business page. The goal is to make your content accessible quickly, not to delete it until you get a good grade.
The article on mobile sites complements this reading: an indicator tells you where to look, but it doesn't verify that the request actually reached your inbox.
Core Web Vitals are useful when they provide an explanation for what the visitor is experiencing. If the diagnosis finally allows us to understand why the menu isn't responding, we've gained something. If we're just collecting screenshots of green scores, we've mostly just gained screenshots.
When we talk about a slow site, we often confuse these situations with the time it takes to display the page. However, you can have a site that appears quickly but becomes frustrating as soon as you try to use it. Core Web Vitals serve to put names and measurements on these sensations, rather than just saying "it works fine for me".
Three metrics for three moments of the visit
LCP, or Largest Contentful Paint, tracks the display of the largest content element visible during loading, often a large image or a block of text. It doesn't measure when all downloads finish. It helps to understand when a significant part of the first screen actually becomes visible.INP, or Interaction to Next Paint, focuses on the visible response after an interaction. You touch the menu: how long before you see something happen? It looks at interactions throughout the visit, not just the first click. A browser busy with heavy processing can leave your gesture waiting even if the server has already sent the entire page.
CLS, or Cumulative Layout Shift, measures unexpected layout shifts. This is the case of the button that slips away from your finger because a block takes its place. Voluntarily scrolling a page is not the same thing: here we are looking for position changes that the user did not request.
The benchmarks published by Google on web.dev are an LCP of 2.5 seconds or less, an INP of 200 milliseconds or less, and a CLS of 0.1 or less. They are evaluated at the 75th percentile, separately for mobile and desktop: for each indicator, at least 75% of measured visits must stay within the recommended threshold. So, it's not your best attempt that decides if the experience is good.
And where does Google fit in?
Yes, Core Web Vitals are part of Google's ranking systems. No, getting all green scores does not guarantee you the top spot. Google states this in its documentation: good results alone are not enough to achieve a high ranking.A fast page that doesn't explain your service is still a page that doesn't explain your service. If your visitor is looking for a repair in their town and finds a vague company presentation, gaining a few tenths of a second won't answer their question.
On the other hand, it's a shame to work on SEO, attract the right people, and then offer them a journey that prevents them from contacting you. Performance is primarily about not wasting that visit. This is why I prefer to start with service pages and the contact form rather than a secondary page that is easy to push to 100.
Why your phone and the test don't tell the same story
You know your site, you've already opened it, its files might be cached, and you're connected to Wi-Fi. The person discovering it has a different phone, other apps running, and a connection that can vary. You are not experiencing the same visit.PageSpeed Insights presents a simulated test and, when enough data is available, metrics from real visits. The test helps reproduce the causes of a problem. Field data shows what happened for visitors over a period, under their own conditions.
A small site may lack field data at the page level. This is neither a perfect result nor a failure: it is unavailable information. And if the tool displays results for the entire domain, you need to know this before presenting them as the results for the form you just tested.
The Lighthouse score is also not a complete measurement of your visitors' real interactions. An automated loading test doesn't spend several minutes opening the menu and filling out your fields. You must keep both perspectives: the technical diagnosis and the manual user journey.
The server responded, but the visitor is still waiting
Imagine a delivery. The package has arrived at the entrance, but it still needs to be opened and its contents installed. A fast server response is a good start; the browser still needs to download images, understand styles, prepare the display, and run scripts.A cache like Varnish can speed up the response. It doesn't automatically make a multi-megabyte photo lightweight, and it doesn't stop a script from blocking the menu. Changing hosting without looking at what happens after the response is sometimes like speeding up the delivery of a package that no one can unwrap.
For LCP, we look specifically at server delay and what delays the main image or text. For INP, we examine the work required of the browser at the time of interactions. For CLS, we reserve space for images and content that arrive later, so they don't push the rest of the page around.
What to fix first
I start with what prevents a person from doing something useful. A blocked menu or a moving button often deserves more attention than a technical detail with no perceptible impact on the journey.We can then address one cause at a time: resizing an image to a suitable size, removing an unused widget, limiting scripts loaded at the start, or planning the space occupied by a visual. Then we replay the same scenario on mobile. If you change everything simultaneously, it becomes difficult to know what actually improved the visit.
You also need to check that you haven't sacrificed what was useful just to gain points. An empty page loads fast. That doesn't make it a good business page. The goal is to make your content accessible quickly, not to delete it until you get a good grade.
Tracking that helps you decide
Choose a few pages that matter: a service, a project, prices if you have them, and contact. Note what seems slow, measure it, fix it, then check again under the same conditions. Field data takes longer to reflect a change, as it covers a period of visits.The article on mobile sites complements this reading: an indicator tells you where to look, but it doesn't verify that the request actually reached your inbox.
Core Web Vitals are useful when they provide an explanation for what the visitor is experiencing. If the diagnosis finally allows us to understand why the menu isn't responding, we've gained something. If we're just collecting screenshots of green scores, we've mostly just gained screenshots.




