The European Site Speed Index: 331,804 sites, measured
We loaded 331,804 live business websites between 9 and 20 September 2026. The results sit against almost all of the website-speed advice we see. The server is rarely the problem. The page is.
Three findings.
- Ten or more detected third-party tools on a page, and the average score is 65.3 out of 100. One or two tools, and it is 74.0.
- 93.7% of the sites we measured get CSS delivered that the page never uses. 70.1% request at least one file that is not there at all.
- Shops sit below other websites in every country we measured: five to seven points, across 20,407 shops and 38,609 other sites.
What follows covers what we measured, what the score means, and what it does not prove. Every figure below carries its sample. The full report, with all thirty country tables and the method in full, sits as a PDF at the end.
What we measured
We hold a register of 1,240,878 business domains. 388,387 of them have completed a scan. 331,804 returned a numeric score. This report analyses that set of 331,804. The scans ran over seven reporting days between 9 and 20 September 2026.
Each page was loaded by our own browser-scan platform, which fetches it the way a real browser does and records what happens: how long the server thinks before it answers, how many requests the page makes and how many fail, how long the page takes to become usable, whether the layout shifts while it loads, how many JavaScript errors and warnings the console reports, and how much of the CSS delivered to the browser the page actually uses. Those measurements combine into one score out of 100, where higher is better.
The score is ours, not Google's Lighthouse or PageSpeed score. Run the same page on the same day and it reproduces.
| Population | Sites | Share |
|---|---|---|
| Domains held in the register | 1,240,878 | — |
| Completed a scan | 388,387 | 31.3% of the register |
| Returned a numeric score (analysed here) | 331,804 | 100% |
| Carry a contact email address | 227,020 | 68.4% of scored sites |
| Score under 70 and carry a contact email | 72,512 | 21.9% of scored sites |
The mean is 73.1 and the median is 76. One site in three scores under 70. One in ten scores under 50.
What a page carries, and where it is hosted
Count the third-party tools a page loads. The pattern shows up straight away.
| Technologies detected on the page | Sites | Mean score | vs dataset mean |
|---|---|---|---|
| One or two | 286,883 | 74.0 | −0.9 |
| Three to five | 2,040 | 76.3 | +3.2 |
| Six to nine | 10,484 | 72.1 | −1.0 |
| Ten or more | 32,393 | 65.3 | −7.8 |
The tools that cost the most are the ones built to watch the visitor. Eighteen technologies appear on at least 1,500 measured sites and sit well below the dataset mean:
| Technology present | Sites | Mean score | vs dataset mean |
|---|---|---|---|
| Hotjar | 1,553 | 61.9 | −11.1 |
| Bing Ads | 2,035 | 62.1 | −10.9 |
| Microsoft Clarity | 4,994 | 62.4 | −10.7 |
| Google Analytics | 5,001 | 63.5 | −9.6 |
| Klaviyo | 12,711 | 64.0 | −9.1 |
| Google Tag Manager | 17,805 | 64.1 | −9.0 |
| Facebook Pixel | 18,946 | 65.0 | −8.1 |
| Klarna | 4,379 | 65.1 | −8.0 |
| Google AdSense | 31,057 | 65.1 | −7.9 |
| Google Ads Pixel | 31,982 | 65.2 | −7.9 |
| TikTok Pixel | 6,619 | 65.7 | −7.4 |
| Vimeo | 12,575 | 66.6 | −6.5 |
Rolled into categories: advertising and tracking pixels at 65.6 across 35,486 sites, marketing automation at 64.7 across 19,066, chat and support widgets at 65.0, video embeds at 67.1 across 18,491.
The other half. The web server makes very little difference:
| Server | Sites | Mean score |
|---|---|---|
| Apache | 95,989 | 74.0 |
| nginx | 56,268 | 74.1 |
| LiteSpeed | 57,151 | 72.6 |
| Cloudflare | 21,988 | 75.1 |
| Microsoft IIS | 594 | 67.5 |
Plain WordPress on plain Apache is the best-performing large group in the dataset: 27,935 sites at a mean of 76.8, above the dataset average. Nothing clever sits in that stack. It carries less than the others.
One caveat, and a real one. This is an association measured across fresh scans, not a controlled experiment. Sites that add many tools may have other things going on. The samples are large. The story is not.
What breaks, and how often
The scan also records specific faults. Below are the shares of the 331,804 measured sites where each one turned up.
| Fault found on the page | Share of sites | Sites |
|---|---|---|
| CSS delivered that the page never uses | 93.7% | 310,872 |
| At least one file requested that does not exist | 70.1% | 232,515 |
| Page takes several seconds to load | 62.9% | 208,799 |
| JavaScript errors or warnings in the console | 60.6% | 201,015 |
| Fails Google's largest-contentful-paint threshold | 49.6% | 164,622 |
| Slow time to first byte | 40.9% | 135,856 |
| Slow blog pages | 30.4% | 100,862 |
| No meta description | 28.1% | 93,210 |
| Layout visibly shifts while loading | 20.7% | 68,543 |
| Slow product or shop pages | 15.2% | 50,303 |
A page can carry several of these and still load acceptably on a fast connection. A fault-free page can still feel slow. The table shows how common each defect is across a large population, not what it costs a specific business. That cost depends on traffic, conversion rate and margin, which is why no euro figure appears against somebody else's website.
Where the good and the bad pages are
Site speed is not evenly spread across Europe. Germany averages 80.0. Italy averages 71.6. Countries with fewer than 150 measured sites are left out. A sample that small will not move a national figure.
| Country | Sites | Mean | Median | Share under 70 |
|---|---|---|---|---|
| Germany | 32,317 | 80.0 | 84 | 18.7% |
| Finland | 2,608 | 79.3 | 83 | 18.7% |
| Austria | 3,488 | 79.0 | 83 | 20.2% |
| Denmark | 3,752 | 78.7 | 82 | 21.2% |
| Belgium | 3,913 | 77.7 | 81 | 21.7% |
| France | 14,618 | 75.9 | 77 | 25.1% |
| Poland | 5,283 | 75.6 | 79 | 29.4% |
| United Kingdom | 7,320 | 75.3 | 78 | 29.3% |
| Ireland | 562 | 73.2 | 74 | 35.9% |
| Spain | 7,183 | 73.6 | 77 | 34.0% |
| Greece | 1,842 | 72.3 | 74 | 38.3% |
| Italy | 6,492 | 71.6 | 74 | 39.2% |
| Portugal | 1,120 | 71.5 | 76 | 34.9% |
Ireland sits towards the bottom of that table at 73.2, on 562 measured sites. The sample is small. Say it here and the number cannot firm up in the retelling.
A caveat on the country figures. 209,768 of the scored sites, 63.2%, could not be placed in a country with confidence. They score 70.8 between them, four points below the sites we could place. They get their own row in the report, not a guess at their location.
Shops score lower than other sites, in every country we measured
Ecommerce sites sit below non-commerce WordPress sites in every market in the dataset, by five to seven points. Same direction every time.
| Country | Shops | Mean | Other WordPress | Mean | Gap |
|---|---|---|---|---|---|
| Italy | 1,864 | 67.7 | 3,503 | 74.1 | −6.4 |
| Spain | 933 | 69.9 | 1,404 | 76.4 | −6.5 |
| Poland | 1,463 | 72.5 | 2,627 | 79.3 | −6.8 |
| France | 1,195 | 72.5 | 2,436 | 78.0 | −5.5 |
| United Kingdom | 1,665 | 72.6 | 3,049 | 78.9 | −6.3 |
| Ireland | 200 | 72.3 | 214 | 78.0 | −5.7 |
| Netherlands | 1,717 | 74.9 | 2,684 | 80.0 | −5.1 |
| Germany | 2,752 | 76.9 | 8,226 | 81.5 | −4.6 |
The gap is not a sign of bad building. A shop loads a payment stack, a checkout script, an app for reviews, an app for email, an app for returns and three tracking tags before a product photograph appears. Each one is defensible on its own. Together they cost points.
Markets outside the European Union
The register reaches past Europe. The spread out there is wider than anything inside it.
| TLD | Sites | Mean score | Share under 70 |
|---|---|---|---|
| .vn (Vietnam) | 3,459 | 57.6 | 72.3% |
| .au (Australia) | 4,026 | 62.1 | 62.7% |
| .jp (Japan) | 3,238 | 62.3 | 64.1% |
| .za (South Africa) | 1,115 | 65.0 | 56.4% |
| .br (Brazil) | 4,506 | 68.2 | 47.7% |
| .com (no country in the extension) | 128,670 | 71.2 | 39.9% |
Australia: 4,026 sites at a mean of 62.1, with 62.7% scoring under 70. Germany averages 80.0 with 18.7% under 70. Australia is an English-speaking, high-income market, and the worst-built developed market in this dataset.
What this does not prove
- No trend. Each domain was measured once. Seven days in September 2026 is a snapshot, and nothing here says whether a market is getting faster or slower. That needs the same sites measured again.
- No causation. The link between tool count and score is measured, not proven. It holds across 331,804 sites.
- One page per site. We scored the page a visitor lands on, not the whole site. A fast homepage over a slow checkout will look better here than the business is.
- Our score, not Google's. Different composite, different weights. It ranks nobody's engineering.
What we would do with this
Four things, in this order. Three of them cost nothing:
- Find out what the page loads and never uses. 93.7% of the sites we measured carry at least some of it.
- Fix the requests that fail. 70.1% of sites ask for files that are not there, and the repair costs nothing.
- Clear the console errors before adding another tool. 60.6% of sites throw them on the page a visitor lands on.
- Audit the tag stack before blaming the host. Sites with ten or more tools average 65.3, and the server underneath barely moves that number.
If you want your own site's number, we will run the same scan against it by hand and send back what it found: the requests that fail, the CSS that is never used, the console errors, the timings, and a score out of 100, with the pages it was measured on named so you can check it. No charge, no obligation.
Ask for the free teardown, or take the full report below.
The European Site Speed Index, September 2026 (PDF)
Every table in full: all thirty countries, the technology breakdowns, the fault rates, the method and the limits. Download the report.
Method in one line: 331,804 sites from a register of 1,240,878 domains, loaded by our own browser-scan platform between 9 and 20 September 2026, scored out of 100. Percentages are shares of the measured set. Individual sites are not named in this work. The measurements are ours, and a site's own numbers go to its owner on request.