- Google judges three things: how fast the largest element appears (LCP), how the page reacts to gestures (INP) and how stable it stays (CLS).
- Only real visits count towards the assessment; the laboratory score out of a hundred is a working tool, not a verdict.
- Speed is one signal among others: it helps a useful page, it never replaces relevant content.
Why speed matters
Most of your visitors do not discover your website sitting in front of a large screen on the office broadband. They discover it on a phone, in the street, in a queue, on a connection that comes and goes. In those conditions every heavy element has a price: an oversized photo, one more script, a font that lags behind. The visitor does not know why the page is dragging; they only know that it is, and they go back to choose the next result.
Slowness also has a less visible cost: it spoils the impression you leave. A page that jumps just as someone tries to tap a button, a menu that does not respond, an image that turns up after the text: all of this tells the visitor that the house is not well kept. These are details, but details are how people judge a craftsman, a practice or a shop before walking through the door. Google, for its part, has chosen to measure exactly these details and to take them into account among other signals.
Google's three measures, in plain words
Google groups these measures under the name Core Web Vitals, and Google Search Console has a report of the same name. There are three of them, and each answers a simple question that every visitor asks without realising it: does it appear, does it respond, does it stay put? The acronyms are a little forbidding, but the ideas behind them are common sense. Here is what each one observes.
LCP: does it appear?
LCP (Largest Contentful Paint) measures how long it takes to display the largest element visible on screen when the page opens. On most websites this is the large photograph at the top, or the main heading when there is no image. The choice of that element is no accident: it is what gives the visitor the feeling that the page has arrived. Text that appears quickly beneath an image that does not come makes a half-loaded page, and that is the moment LCP captures.
INP: does it respond?
INP (Interaction to Next Paint) measures responsiveness. Each time the visitor taps a button, opens a menu or types in a field, the page must show that it has understood: INP times the gap between that gesture and the next frame displayed on screen. It keeps one of the slowest responses of the visit, not the best. This measure replaced an older one, FID, in March 2024; FID only looked at the very first gesture and let through pages that froze afterwards.
CLS: does it stay put?
CLS (Cumulative Layout Shift) measures visual stability. It looks at the unexpected movements of content during the visit: a paragraph that drops because a banner has slipped in above it, a button that slides away just as the finger arrives. It is not a duration but a score without a unit, which rises as the jumps grow larger. It is the most tangible of the three indicators: everyone has tapped the wrong link because the page moved at the last moment.
The thresholds to aim for
For each of these measures, Google publishes thresholds that sort a page into three zones: good, needs improvement, poor. For LCP, a page is considered good if its largest element appears within 2.5 seconds, and poor beyond 4 seconds. For INP, the limit of good is 200 milliseconds (a fifth of a second), and poor begins beyond 500 milliseconds. For CLS, a score of 0.1 or less is good, and anything above 0.25 is poor.
These thresholds do not apply to the average visit, a point that is often overlooked. Google takes the value below which three quarters of visits fall, what statisticians call the 75th percentile. In other words, for a page to be judged good, at least three visits out of four must be in the green zone. The assessment is also made separately for mobile and for desktop: a website can be excellent at the office and mediocre in your customers' pockets.
Measuring your own website
There are two kinds of measurement, and confusing them is the first source of misunderstanding. Field data comes from real visitors: Google gathers it in the Chrome User Experience Report (CrUX), from visits made with that browser. Laboratory data comes from a simulation: the Lighthouse tool loads your page on a deliberately modest phone and connection, then calculates a score out of a hundred. The first tells you what your customers experience; the second tells you why, and is the one you work with.
The simplest tool for seeing both is PageSpeed Insights, free and published by Google. At the top it shows the Core Web Vitals assessment based on real visits, when it has enough of them; below, the laboratory diagnosis and its score. Google Search Console, which every website owner should have set up, also offers a Core Web Vitals report that sorts your pages into groups. A small website, or a page with few visits, may not have enough visits for field data: the tool then says so plainly, and you have to make do with the laboratory.
- Choose the right pages
Measure the home page, but also the two or three pages that matter most: a service page, a product page, the contact page. They do not always have the same faults.
- Start with mobile
In PageSpeed Insights, stay on the mobile tab first: that is often where your customers are, and where faults show most clearly.
- Read the field before the score
Check whether real visits pass the assessment before worrying about the figure out of a hundred. An average score with green field data is better than the reverse.
- Find the element at fault
The laboratory diagnosis usually names the element chosen for LCP and the heaviest resources. Write them down: they are your first leads.
- Measure several times
A laboratory measurement varies from one run to the next. Run it two or three times before drawing conclusions, and always compare under the same conditions.
What slows down a small business website
On the websites of shops, practices and craftspeople that we examine, the causes of slowness look very much alike. They rarely come from a deep flaw, and almost always from accumulation: a tool added here, a photo uploaded as it is there, a theme that carries a host of functions in order to use a few. Each addition seems harmless; together they are not. Here are the most frequent.
- Images that are too heavy, uploaded straight from the camera in older formats, when the WebP and AVIF formats are generally much lighter at equal quality.
- Images without declared dimensions (the
widthandheightattributes): the browser does not know how much room to keep for them, and the text jumps when they arrive. - Heavy web fonts, loaded in full and in several weights, for a handful of headings.
- Third-party scripts: live chat window, embedded map, advertising pixels, audience measurement. Each is useful, but each runs on your visitor's phone.
- Carousels, and themes or site builders loaded with unnecessary code, which weigh down every page, even those that do not need it.
- Banners and adverts inserted without reserved space, which push the content down just as the visitor is reading or tapping.
- Too much JavaScript to run at once: while the browser is busy, it cannot respond to gestures, and INP suffers.
What we do to speed up a website
Speeding up a website is rarely a matter of magic; it is a matter of discipline, applied from the design stage rather than afterwards. On the websites we build, every image is delivered in WebP, at the size at which it will be displayed, with its width and height declared. Fonts are hosted with the website itself rather than by a third party, and split into slices: the browser downloads only those the page needs. Scripts that are not essential to the first display are deferred, and no audience measurement script is loaded before the visitor has consented to it.
We also reserve the place of every element before it arrives: an image, a panel, a window each have their space planned in the layout, and nothing disturbs the text you are reading. The Japanese call this deliberately kept space ma; where speed is concerned, it is what protects CLS. This work is part of what we mean by web design: a beautiful website that loads badly is not a finished website.
Our own website, measured without indulgence
We applied the same rigour to our own house, and the measurement first proved us wrong. The mobile PageSpeed Insights score of our website was recorded at 32. The main cause was a 3D ceramic piece, superb on a computer but costly to compute on a phone: on small screens, an image of the same piece now appears first. We then found that the website's Japanese fonts were requested before the first display, and we deferred them. The score rose to around 96 in laboratory measurement: this promises nothing for another website, but shows that two well-aimed decisions are worth more than a host of small adjustments.
What speed will not do
It must be said honestly: speed is not a magic wand for search rankings. Google presents page experience, which includes Core Web Vitals, as one signal among others, and makes clear that more relevant content can rank above a faster page. A very fast website that does not answer the visitor's question will stay behind a slower one that answers it well. Speed mostly separates pages of comparable quality, and keeps the visitors that the content has managed to attract.
This calls for a sense of proportion. Chasing a laboratory score of 100 at the cost of an impoverished website makes no sense; aiming for green on real visits while keeping what makes you distinctive makes a great deal of sense. Before speeding up a page, make sure it says clearly what you do, for whom and where: that is the subject of our article on the title, the description and the H1, and the heart of search engine optimisation. And if you are preparing a new website, take the opportunity to lay these foundations from the start without losing what the old one had earned, as we explain in redesigning your website without losing your search rankings.
A fast website does not replace a useful page; it simply keeps that page from losing its readers along the way.
What you can do this week
Tick as you go: nothing is saved, it is simply a reminder.
Founder and director of Digital Air Media, a digital creation house opened in Paris in 2017. He writes here what the house applies to its own pages before offering it to others.