Improving Core Web Vitals: Find and Fix Four Brakes

Large photos, too many third-party scripts, late-loading fonts and jumping content: four causes of slow pages, how to measure them yourself and how to monitor the values automatically.

If you want to improve your Core Web Vitals, do not start with some plugin, but with the question of why the page loads so slowly in the first place. Four causes are worth checking first, because you can find and fix them without a computer science degree: photos that are too large, too many third-party scripts, fonts that load late and content that jumps while loading.

How much each of these slows you down only a measurement of your own site can show. That is why each point says how to recognise it and what you can do.

What are Core Web Vitals anyway?

Core Web Vitals are three measurements that describe how fast and stable your page feels to real users. LCP measures when the largest element becomes visible. INP shows how quickly the page reacts to clicks and input. CLS records whether things jump around while loading. An LCP of up to 2.5 seconds, an INP of up to 200 milliseconds and a CLS of up to 0.1 count as good, measured at the 75th percentile of page views (web.dev, accessed 25 September 2026).

And the ranking? Whether faster values change your position on Google, nobody can promise you. What Google itself writes about user-friendliness is in its documentation on page experience (Google Search Central, accessed 25 September 2026). All that is certain: speed does not replace good content. Your visitors do not know the abbreviations anyway, but they notice whether a page loads sluggishly.

Why do huge photos make a website slow?

Because the browser has to download every image completely before it can show it. Someone uploads a photo straight from the phone camera or the SLR, 5000 pixels wide and 8 megabytes in size, and builds it in as the cover image. On a mobile network at the bus stop in Puchheim, that takes what feels like forever.

A worked example, not a measurement from a client project: say a baker in the district uses a bread photo of 8 megabytes as the cover image. That is 64 megabits. At an assumed 10 megabits per second, the transfer alone takes about 6.4 seconds. Brought to display size and compressed, say to 200 kilobytes, it would be 0.16 seconds on paper. How small the file really gets for your picture, you only see by trying. Measure before and after with the same tool.

The solution, in brief:

  • Bring photos to display size before uploading. For a cover image, 1920 pixels of width is often enough; for phones, supply a smaller variant via srcset.
  • Use a modern format. According to Google, lossy WebP images are 25 to 34 percent smaller than comparable JPEGs at the same image quality (Google for Developers, accessed 25 September 2026).
  • Set width and height in the code and use loading="lazy" only for images further down, not for the cover image.

How many tracking scripts are too many?

As many as you do not really need. Quite a lot piles up over the years: Google Analytics, plus the Facebook pixel, a chat widget, a review badge, a cookie banner tool, perhaps an old newsletter pop-up. Each of these scripts loads code from a third-party server, can block the page from building and costs processing time on your visitors' phones.

Count for yourself. Open your page in Chrome, press F12, switch to the "Network" tab and reload. Sort by domain: everything that does not come from your own address comes from a third-party service. Ask yourself honestly about each entry: do I really need this, and if so, does it have to load immediately as the page builds, or is later enough?

Why does the font flicker while loading?

Because the actual typeface is only loaded afterwards. The page loads, for a moment you see text in a default font, then everything switches. That is called FOUT, and it looks restless. If the font comes from Google Fonts, meaning from a third-party server, the browser waits for it.

Two things help. First, put the font files on your own server instead of fetching them from Google. A side effect: loading the font then sends no request with your visitors' IP address to Google. Second, set font-display: swap in the code so that readable text appears immediately. And if you are honest, one or two font styles are usually enough. Not seven.

What is layout shift, and why is it so annoying?

Layout shift means content moves while loading. You want to tap a button, at the last moment an image loads above it, everything slides down, and you hit the ad instead of the button. That is exactly what CLS measures. A value of up to 0.1 is good.

Typical triggers are images without a fixed size, ad blocks that only claim their space late, and late-loading fonts with a different height. The solution is usually unspectacular: you reserve the space from the start, with height and width for every image and fixed containers for dynamic content. Nothing wild, just care when building. Our glossary explains the three measurements one by one under Core Web Vitals.

What does the optimisation actually bring?

That depends on the starting point, and a serious answer only comes after a measurement. If a page is already built decently, you may gain a few tenths of a second. If there is an 8-megabyte image in it as in the worked example, the difference is considerably larger on paper. Whether more enquiries come in afterwards, nobody can promise beforehand. You only see that if you record the same figures before and after, for example the enquiries through the contact form per month.

Automating website upkeep: how do you keep an eye on the values?

You can automate this part of website upkeep by having the measurement run regularly by itself and report only when a value rises above the threshold. That does not replace looking at the cause, but makes sure it happens in good time.

The values rarely get worse in one go. Usually a new photo is added, a widget, a script for a campaign, and nobody notices. Here is how to go about it:

  1. In Google Search Console, open the report on the three measurements. It shows which groups of pages break the thresholds. With few visitors it may stay empty, because there is not enough user data.
  2. Pick three to five pages that matter, such as the homepage, a services page and contact.
  3. Have these pages measured every week through the PageSpeed Insights interface. For frequent automatic queries, Google recommends your own API key (Google for Developers, accessed 25 September 2026). On the same page, Google announces that it will no longer deliver the real-user data from the Chrome User Experience Report through this interface, and recommends the CrUX API for that.
  4. Decide when a message is sent, for example when LCP rises above 2.5 seconds or CLS above 0.1, and who it goes to.
  5. Have images shrunk automatically on upload and converted to WebP or AVIF. Then load time does not depend on whether someone in the office remembers it.

Say someone at Muster Joinery uploads a new cover image straight from the camera. In the next weekly measurement the homepage LCP rises above the threshold, and a message goes to the owner before visitors notice. How we build such processes is described on our page on AI and automation for businesses. We are only just building this offer, and there is not yet a business where it already runs permanently with us. Every process starts as a prototype at a fixed price from €1,500 with four weeks of testing in your business, followed by a monthly flat fee, cancellable monthly.

Frequently asked questions

How can I test my Core Web Vitals myself?

For free through Google's PageSpeed Insights. Enter the address, choose the mobile view, read off the values. Look at the field data at the top (real users), not only at the lab test. If the field data is missing, the lab test remains as a guide.

Is a caching plugin enough to make the website faster?

It helps but does not solve the core problem. A cache delivers the page faster, but it does not make an 8 MB image smaller and does not remove a superfluous tracking script. Remove the causes first, then put a cache on top.

Do I have to rent a more expensive server for fast load times?

Not necessarily. First check how long the server takes for the first response; in the network tab this shows as "Waiting for server response". If that value is small and the page is slow anyway, the problem lies in how the page is built. Optimise first, then see whether the server is the bottleneck at all.

At bytebrise we build websites for small businesses on the rental plan bytebrise Sichtbar: €490 one-off to start, then €149 a month, hosting and maintenance included, cancellable at the end of any calendar month with no notice period. If you want to know where your site is slowing down, write to us through the contact form.

Researched and written with AI assistance, approved for publication by .

How does your website stack up?

Free 60-second check: meta data, content, images, schema.org and GDPR basics. Instant, server-side result.

Check my website free