Website Load Time Statistics 2026: Core Web Vitals and Speed

Written by Nick Galov • Edited by Jordan T. Prodanoff
• 14 min read

Speed is essential, and for years that was the end of the discussion. The numbers people quoted to prove it, a one-second delay costing 7% of conversions, the average mobile page taking 15.3 seconds, were lifted from a handful of case studies and repeated until they sounded like physical laws.

They were never universal, and most of them are now close to a decade old. Page speed is also no longer one number. It is three separate measurements covering loading, responsiveness and visual stability, each with a published threshold, measured on real visits rather than lab tests.

This page uses those measurements and the 2025 field data behind them.

What Is a Good Page Load Time?

“Under three seconds” was the old answer. The current one is more specific, because “load time” was always a blunt instrument: a page can paint its main content quickly and still feel broken if it jumps around or ignores taps.

Google defines good performance through three Core Web Vitals:

  • Largest Contentful Paint (LCP), 2.5 seconds or less. How long until the biggest piece of content in the viewport appears. This is the closest thing to the old load-time number.
  • Interaction to Next Paint (INP), 200 milliseconds or less. How quickly the page responds when someone taps or clicks.
  • Cumulative Layout Shift (CLS), 0.1 or less. How much the layout moves around while loading.

Three details matter more than the numbers themselves.

First, each is assessed at the 75th percentile of page loads, split by mobile and desktop. You are judged on your slower quarter of visits, not your average one, so a fast median hides nothing.

Second, a page passes the Core Web Vitals assessment only if it is good on all three at once. Excellent LCP does not buy forgiveness for layout shift.

Third, none of these is full-page load time, and lab and field data are different things. A lab test like Lighthouse simulates one visit on one connection. Field data, which is what Google actually uses, comes from real Chrome users on real devices and networks. They routinely disagree, and the field data is the one that counts.

Core Web Vitals at a Glance

The quick version, if you want the whole picture in one place.

The thresholds

  • LCP: good at 2.5s or less, poor above 4s
  • INP: good at 200ms or less, poor above 500ms
  • CLS: good at 0.1 or less, poor above 0.25
  • Measured at the 75th percentile, all three must pass

How the web actually performs (2025 field data)

  • Good overall Core Web Vitals: 48% of mobile sites, 56% of desktop
  • Good LCP: 62% mobile, 74% desktop
  • Good INP: 77% mobile, 97% desktop
  • Images are the LCP element on 76% of mobile pages and 85.3% of desktop pages

Page weight and pass rates (home pages)

  • Under 1 MB: 57% of mobile pages pass, 70% of desktop
  • 5 MB or larger: 30% of mobile, 38% of desktop
  • Median home page: 2.6 MB mobile, 2.9 MB desktop

What Google says about ranking

  • Core Web Vitals are used by its ranking systems
  • There is no single page experience signal
  • Good scores do not guarantee a high ranking
  • Since July 5, 2024, Search crawling uses Googlebot Smartphone

Sources and the detail behind each line are in the sections that follow.

How Fast Is the Web?

The HTTP Archive measures this annually from real Chrome user data, which makes it the best available answer to “is my site normal?”

48% of mobile websites and 56% of desktop websites achieved good overall Core Web Vitals in 2025. (HTTP Archive Web Almanac 2025) Roughly half the web fails, and it fails more often on the devices most people browse from.

Mobile has been improving steadily: 36% in 2023, 44% in 2024, 48% in 2025. Desktop went 48%, 55%, then 56%, so last year was almost flat. Mobile is catching up, but it still has not reached where desktop sat three years ago.

LCP: 62% of mobile pages are good, against 74% on desktop

Loading is where the device gap shows most clearly. Mobile also has nearly double the rate of poor experiences, 13% against 7% on desktop, which is what you would expect from slower networks and less capable hardware.

If you only have budget to fix one metric, this is usually it, and the next section explains what to point that budget at.

INP: 77% of mobile sites are good, against 97% on desktop

Responsiveness is the opposite story. Desktop is essentially a solved problem at 97%. Mobile improved from 74% in 2024 to 77% in 2025, and mobile home pages did better still at 80%.

INP is largely a JavaScript problem. Ecommerce, social and news sites execute the most script and score worst, which is a design and dependency decision rather than a hosting one.

Images are the LCP element on 76% of mobile pages

On desktop it is 85.3%. This is the single most actionable number on this page: in roughly four out of five cases, the thing holding up your largest contentful paint is an image.

That makes image optimisation the highest-leverage performance work available to most sites, which is why it leads the fixes section below.

Page Weight and Performance

The 2025 Web Almanac paired page weight with field data for the first time, and the relationship is stark. For home pages:

  • Under 1 MB: 57% of mobile pages passed Core Web Vitals, and 70% of desktop pages
  • 2 to 3 MB: 53% of mobile, 63% of desktop
  • 5 MB or larger: 30% of mobile, 38% of desktop

A page over 5 MB is roughly half as likely to pass as one under 1 MB. For context, the median home page in 2025 weighed 2.6 MB on mobile and 2.9 MB on desktop, so the typical site sits in the middle band where about half of pages pass.

Two honest caveats. Page weight is not part of the Core Web Vitals calculation, so this is a correlation, not a mechanism; heavy pages tend to be heavy for reasons that also hurt performance. And this field data covers pages with enough traffic to appear in the Chrome User Experience Report, so it leans toward more popular sites.

It still gives you a usable target. Getting a home page under 1 MB roughly doubles your odds against the 5 MB crowd.

Speed, Conversions and Revenue

This is where the old statistics were weakest, so it is worth being direct about what was removed and why.

Claims that one second costs 7% of conversions, 11% of page views or 16% of customer satisfaction, and that a one-second delay costs a $100,000-a-day store $2.5 million a year, have been taken off this page. They came from individual case studies at specific companies, or from arithmetic performed on those case studies, and were then repeated as universal rules. They are not. Conversion impact depends on the site, the audience, the device and what the page is for.

The same goes for the bounce probability figures of 32%, 90%, 106% and 123% that this page used to lead with. Those came from 2017 Google research into mobile load times, measured with methods and on hardware that have both moved on.

What the evidence does support

One case study is worth keeping, with its conditions attached. A 2020 study by Deloitte Digital and fifty-five, commissioned by Google, found that a 0.1-second improvement in mobile site speed lifted conversion rates by 8.4% for retail and 10.1% for travel. (web.dev)

Before you budget against that, read the conditions. It covered 37 brands across Europe and the US over four weeks and about 30 million sessions. The improvement had to hold across every page in the user journey, on four separate metrics, before the effect appeared. And it was paid for by a company that sells the case for speed.

So: a real, measured result, from a vendor-sponsored study, under conditions most sites will not reproduce exactly. That is a reason to test speed changes on your own traffic, which is the only number that describes your site. If you are chasing checkout conversions specifically, our cart abandonment statistics show that unexpected costs outrank performance as a reason people leave.

Speed and SEO

The previous version of this page said that high bounce rates tell Google a page lacks value and directly lower its ranking, and that all else being equal the faster page ranks higher. Both are wrong, and the second is unfalsifiable in practice.

Google does not use bounce rate as a direct ranking penalty. What it actually says is narrower and more useful:

  • Core Web Vitals are used by its ranking systems.
  • There is no single page experience signal. Its core systems look at a range of signals that together add up to page experience.
  • Good scores do not guarantee top rankings. Google states plainly that strong results in Search Console or third-party tools are no promise of position, because there is more to page experience than Core Web Vitals, and far more to ranking than page experience.

The honest framing is that good page experience can contribute to success in Search alongside relevance, quality and everything else Google weighs. Speed will not rescue a page that does not deserve to rank, and it is worth fixing for users regardless.

Mobile-first indexing is finished

This page used to say Google “indexes mobile sites first”, which described a transition that has since completed. Since July 5, 2024, Google crawls sites for normal Search with Googlebot Smartphone, the last stragglers on the desktop crawler having been moved over then. A desktop crawler still exists for certain features, but it handles a minority of requests.

The consequence is blunt: if a site’s content is not accessible on a mobile device, it may not be indexable at all. Not ranked lower. Not indexed. That makes mobile rendering a correctness requirement, and it is why mobile performance gets its own treatment throughout this page. Our mobile traffic statistics cover how the device split actually looks.

How to Measure Your Site Speed

Before changing anything, find out where you stand. The important distinction is between lab tools, which simulate a visit under controlled conditions, and field data, which records what real users experienced. Lab tools are for diagnosis; field data is what Google evaluates.

Field data: what Google actually sees

  • Search Console’s Core Web Vitals report. Groups your URLs by status using real user data. The first place to look, because it reflects the assessment Google applies.
  • The Chrome User Experience Report (CrUX). The public dataset behind that report, queryable if you want to compare yourself with competitors or track trends.
  • Google PageSpeed Insights. Shows field data where available alongside a lab run, which makes it the easiest way to see both for a single URL.

Field data needs traffic. A new or quiet site may have none, in which case lab tools are all you have.

Lab tools: finding out why

  • GTmetrix. Detailed analysis with a long list of specific recommendations, and a choice of test location and browser.
  • WebPageTest. The most thorough of these. Overwhelming at first, excellent once you know which tab you need, and the best tool for understanding the order things load in.
  • Pingdom. Quick and readable, lighter on detail than the other two.
  • k6. Different job: it loads your backend rather than measuring the front end, so use it to see how the server behaves under strain.

Results fluctuate between runs even with the same tool, which is normal. Run each test a few times and look at the pattern rather than any single number.

Factors that Affect Speed the Most

Most sites fail for a short list of reasons, and the field data above points at which ones to start with.

Oversized images

Images are the LCP element on 76% of mobile pages and 85.3% of desktop pages, so this is where the largest wins usually are. Serve images no larger than they will display, use a modern format such as WebP or AVIF, give them explicit width and height so they cannot cause layout shift, and lazy-load anything below the fold.

One exception worth knowing: do not lazy-load the LCP image itself. Deferring the very element you are being measured on is a common and self-defeating mistake.

Too many plugins

Plugins and add-ons consume server resources and frequently add their own scripts and database queries to every page. Clearing out ones you no longer use is reliable housekeeping.

There is no magic number, though. Ten lightweight, well-built plugins can cost less than one that loads a framework on every request. Judge them by what they add to the page, not by how many there are.

Caching

Caching serves content that has already been generated instead of rebuilding it per request, and it remains one of the highest-value changes available. Most content management systems have good caching plugins, managed hosts often handle it at the server level, and a CDN adds another layer in front.

Compression

Text resources should be compressed in transit. Brotli generally beats gzip on compression ratio for text and is supported by every current browser; gzip remains the universal fallback. Most hosts and CDNs can enable both, and many already have.

Compression applies to text, not to images and video, which are already compressed in their own formats. Running gzip over a JPEG achieves nothing.

Use a quality content delivery network

A CDN serves static files from a location near the user, cutting both the distance data travels and the load on your origin server. It usually improves security too. Content delivery networks are widely available free at small scale, Cloudflare being the best known.

HTTP requests, bundling and modern protocols

The old advice was to cut the number of requests and combine all your CSS and JavaScript into single files. That was sound under HTTP/1.1, which could only fetch a few resources at a time.

Under HTTP/2 and HTTP/3, which nearly all hosts now support, requests are multiplexed over one connection and the penalty per request is much smaller. Request count still matters, but it is no longer the main lever. What matters more:

  • Total payload size, which the page weight data above shows tracking closely with pass rates.
  • Render-blocking resources. CSS and synchronous scripts in the head hold up the first paint regardless of how many there are.
  • Caching strategy. One giant bundle invalidates entirely when one line changes. Splitting code into sensible chunks lets returning visitors reuse most of it.
  • Prioritisation. Loading the right things first beats loading fewer things.

So: bundle deliberately rather than maximally, and split code so that pages download what they need.

JavaScript placement and main-thread work

Moving scripts above the closing body tag was the old fix. The modern equivalents are the defer and async attributes, which let the browser keep parsing HTML while a script downloads, with defer preserving execution order.

Beyond placement, the real INP work is reducing how much JavaScript runs at all: remove unused dependencies, avoid long tasks that block the main thread, and be deliberate about third-party scripts, which are often the largest single cost and the easiest to forget.

Minification and hosting

Minifying HTML, CSS and JavaScript strips comments and whitespace for a small, free saving, and most build tools do it automatically. Keep readable sources for debugging.

Underneath all of it sits your host. Server response time feeds directly into LCP, and no amount of front-end work compensates for a slow origin. Our web hosting statistics cover the market, and detailed reviews go through individual providers.

Conclusion

Website speed still matters, and it matters for the reason it always did: people leave slow sites. What has changed is that we can now say what “slow” means without inventing a number.

Good is LCP at 2.5 seconds, INP at 200 milliseconds and CLS at 0.1, all three at the 75th percentile of real visits. About half of websites manage it, fewer on mobile than desktop. Images are what usually hold up loading, JavaScript is what usually hurts responsiveness, and lighter pages pass far more often than heavy ones.

Google uses these measurements in its ranking systems and says plainly that good scores guarantee nothing, which is a more honest position than the industry’s old promise that faster always ranks higher. Fix speed because it makes the site better to use. The search benefit, where it exists, follows from that rather than the other way round. Our web design statistics put performance in the wider design context.

About Nick Galov

Author

Unaware that life beyond the internet exists, Nick is poking servers and control panels, playing with WordPress add-ons, and helping people get the hosting that suits them.

View all posts by Nick Galov →

About Jordan T. Prodanoff

Editor

A wayfarer by heart, Jordan fancies journeying into foreign lands with a camera in hand almost as much as he enjoys roving the online world. He spends his time poking at letters and pixels, trying to transmogrify them into something cool.

View all posts by Jordan T. Prodanoff →