How to speed up a website

Measure first, then work through images, fonts, scripts and caching; fix LCP, INP and CLS one by one at the end.

· 10 min readSoftware

Short answer

To speed up a website, first check in PageSpeed Insights which metric is weak in the field data; then fix images, fonts, third-party scripts, caching and a CDN in that order. Fix whatever remains for LCP, INP and CLS separately. The good thresholds are 2.5 seconds for LCP, 200 milliseconds for INP and 0.1 for CLS.

On Monday morning you paste the home page into PageSpeed Insights and see a red mobile score. The same day one person says “compress the images”, another says “delete the plugins” and a third wants to change the server. All three may be right, but measurement decides which comes first. The useful answer to how to speed up a website is not a list but an order: see where you are slow, then start with the work that has the biggest effect.

We covered what speed does for sales in why a fast site sells more; this article is only about how to do it.

1. Measure first: field data or lab data?

PageSpeed Insights shows two kinds of data for a page. The top section is field data: it is collected from real visitors, covers the trailing 28 days, and is what the assessment is based on. The section below is lab data: a single test Lighthouse runs on one device and a simulated connection. Google’s own documentation says the two can sometimes contradict each other; lab data is for debugging, field data is for seeing the real experience. To see how lab scores spread across real sites, see our measurement of 15 agency websites.

Work like this:

  1. Run it separately for your home page, your busiest category or service page and a product page. One score hides the differences between page types.
  2. In the field data, check the mobile and desktop tabs, and the “This URL” and “Origin” views. That tells you whether the problem is this page, the whole site or only mobile.
  3. Note which of the three metrics misses its threshold. Google’s good thresholds are 2.5 seconds for LCP, 200 milliseconds for INP and 0.1 for CLS; they are judged at the 75th percentile of visits, and a page passes only when all three are good.
  4. Only then open the lab section’s opportunities and diagnostics, where you can see which file or script adds weight.

A low-traffic page may show no field data at all. In that case look at the “Origin” view and treat the lab data as a guide only. We walk through reading these rows in Search Console and PageSpeed Insights in Core Web Vitals explained: how to read LCP, INP and CLS.

Seven steps for speeding up a site in order of effect: measure, images, fonts, scripts, caching and CDN, metrics, verify 1 Measure Field data and lab data 2 Images Right size, modern format 3 Fonts Few weights, own domain 4 Scripts Third-party inventory 5 Caching Long-lived files and CDN 6 Metrics LCP, INP, CLS one by one 7 Verify 28-day field data Seven steps for speeding up a site in order of effect: measure, images, fonts, scripts, caching and CDN, metrics, verify 1 Measure Field data and lab data 2 Images Right size, modern format 3 Fonts Few weights, own domain 4 Scripts Third-party inventory 5 Caching Long-lived files and CDN 6 Metrics LCP, INP, CLS one by one 7 Verify 28-day field data
Figure: Seven steps for speeding up a site in order of effect, starting with a measurement and ending with the same measurement.

2. Images: size and format

On most pages the largest element is an image, so images shape both LCP and page weight. Three jobs are enough:

  • Right size: do not send a 3000-pixel file for a photo shown at 600 pixels; use srcset to serve a size that suits the device.
  • Modern format: use WebP or AVIF. web.dev lists moving to these formats among the ways to shrink the LCP resource.
  • The first-screen image: do not lazy-load the largest image on the page. web.dev says this always hurts LCP; keep the image in the HTML with a src and give it fetchpriority="high". Put loading="lazy" on images below the fold.

One warning: web.dev’s LCP guide shows that shrinking the file alone may not fix LCP. What it takes for the image to be found in Google as well is covered in how to do image SEO. If the image is hidden until JavaScript finishes, the time you saved moves into “element render delay”. So measure again once you have shrunk the image.

3. Fonts

When a font arrives late, text either stays invisible or swaps to another font and moves the lines. A practical routine:

  1. Serve the font from your own domain where you can; the connection to a separate server disappears.
  2. Use WOFF2 and strip alphabets you do not use (subsetting). Keep the range that includes the Turkish characters (ç, ğ, ı, İ, ö, ş, ü).
  3. Use font-display: swap so text shows at once in a fallback font.
  4. Cut the number of weights; two are enough for most designs.

4. Third-party scripts

A live-chat window, a heatmap, a few ad pixels, an A/B testing tool: each looks small, but together they keep the browser’s main thread busy. That hurts INP and the page’s load time. List every script in your tag manager and every one embedded in the template, and write down who owns it and when it was last used. Remove what has no owner, merge tools that do the same job, and load the rest with defer or async. For keeping tags tidy, see our Google Tag Manager setup guide.

5. Caching and a CDN

The first part of LCP is how long the server takes to send the first byte (TTFB). According to web.dev, a high TTFB can make a 2.5-second LCP hard or even impossible; the causes it lists include chains of redirects, a server far from the visitor and content that cannot be served from cache. What to do:

  • Remove needless redirect chains (go from http to https and from www to non-www in one step).
  • Cache files that do not change, such as images, CSS, JavaScript and fonts, for a long time; change the file name when the file changes so visitors do not stay on an old copy.
  • Serve pages and files through a CDN close to the visitor. CDN servers sit nearer to users, which shortens the distance each file travels.

6. LCP, INP and CLS: the usual cause and first fix for each

If a metric is still red after the five steps above, take it on its own. The table below summarises the usual causes and the first fix from web.dev’s guides.

Metric Good threshold Usual cause First fix
LCP 2.5 seconds or less Slow first byte, a main image found or loaded late, JavaScript that hides the main content Lower TTFB; make the main image discoverable in the HTML and prioritise it; delay the script that hides content
INP 200 milliseconds or less Long tasks holding the main thread (input delay), heavy event handlers, late painting Break up long tasks, remove unneeded scripts, make the visible change first and leave the rest for later
CLS 0.1 or less Images and iframes without dimensions, late ads and embeds, animations, web fonts Set width and height or aspect-ratio, reserve space for ads and embeds, reduce font shifts

For LCP, web.dev splits the time into four parts: time to first byte, resource load delay, resource load duration and element render delay. On a well-optimised page the first byte and the load duration each take about 40%, and the two delay parts stay under 10%. The “LCP breakdown” in a Lighthouse report shows which part has swollen; start the fix there.

For INP, the time has three phases: input delay, the event handlers running, and the screen updating. Find a slow interaction in the field data first, then record that interaction in the Performance panel of Chrome DevTools and see which phase is long. Lab tools cannot measure INP directly; they give only an approximate indicator.

For CLS, PageSpeed Insights shows two values: the CLS in the field section covers the whole life of the page, while the lab CLS reflects only the load. If they are close, the problem arises during loading, and Lighthouse lists images without dimensions and the elements that shifted. If the field value is much larger, the shift starts after loading, for example in content that loads late as you scroll; the “Layout Shifts” track in the DevTools Performance panel shows it.

Common mistakes

  • Making the lab score the target. Decisions follow field data; the lab score only shows where to look.
  • Looking at one page. The home page can be fast while the product template is slow.
  • Lazy-loading the main image. It is a common setting and it breaks LCP directly.
  • Adding a plugin. Every plugin installed for speed brings one more script of its own.
  • Expecting results at once. Field data covers the last 28 days, so the effect appears gradually.

Common questions

How do you speed up a website, and what should you check first?

Look at the field data in PageSpeed Insights and find which metric (LCP, INP, CLS) and which device (mobile, desktop) misses its threshold. Then work through images, fonts, third-party scripts and caching in that order, and solve what remains metric by metric.

Why do field data and lab data differ in PageSpeed Insights?

Field data reflects the last 28 days of real visitors; lab data is one simulated load on a single device. Google’s documentation notes that the two can sometimes contradict each other. Use lab data to look for causes and field data to decide.

When do the results of a speed fix show up?

Field data covers the last 28 days, so the result shows gradually over a few weeks. In the lab you can see a change straight away; it shows you are heading the right way but does not replace the field assessment.

If you would rather hand the fixes to an agency or a development team, our web and software service covers this work. To keep your rankings when you rebuild a site, read keeping SEO through a site rebuild, and for the overall shape of the site see what a company website needs. A short definition of the metrics is in the glossary.

One step you can take today: enter your busiest page into PageSpeed Insights. Then, on a sheet of paper, write down the first metric in the field data that misses its threshold and, if it is LCP, the longest part of its breakdown. Our Core Web Vitals checker gives a quick view of the three metrics. Fix only that one line and re-measure after 28 days.

Sources

This article is general information, not legal, financial or technical advice. Platforms, their rules and the law change over time, so some details may have changed since 11 November 2026 or contain small errors. Before you make a decision for your business, check the current official sources or write to us.

Tell us about
your project.

Fill in the form or write to us directly, and we will agree on a day to talk.

Your first meeting is with Deniz Sezer, founder of Detmore.

First meeting free · Reply within one working day · Written proposal in 2 to 3 days · NDA ready

What do you need? pick as many as you like

Name, email and phone are required.

Contact

We reply within one working day.