One question comes up the moment you choose how to build a website: which technology produces the fastest site? The answer exists, it is public, and it is measured every month across millions of real sites. It is also less flattering to custom builds than you might hope.
What Google measures, exactly
Google grades three things on every page, using real visits from Chrome users. They are called the Core Web Vitals.
- LCP: how long the largest visible element takes to appear. It must stay under 2.5 seconds.
- INP: how fast the page reacts to a click or a keystroke. Under 200 milliseconds.
- CLS: how stable the layout stays while loading. Under 0.1.
A site passes when all three grades are good for 75% of its visitors. That is demanding: as of May 2026, only 55.9% of web origins manage it. That number is the benchmark that matters, because it turns an abstract percentage into a relative position.
The April 2026 ranking
HTTP Archive publishes the data by technology. Share of sites passing all three Core Web Vitals, followed by median page weight.
| Platform | Sites passing all three CWV | Median page weight |
|---|---|---|
| Duda | 85% | 1.78 MB |
| Wix | 80% | 2.67 MB |
| Shopify | 79% | 3.77 MB |
| Astro | 67% | 1.65 MB |
| Drupal | 64% | 2.39 MB |
| Joomla | 58% | 2.65 MB |
| WordPress | 49% | 2.76 MB |
| Web overall | 55.9% | n/a |
Astro is the stack I build on. It comes fourth. Better to say so upfront than to let the reader find out: three self-service platforms do better, and that is no accident.
WordPress sits below the web average
49%, against 55.9% for the web as a whole. WordPress, which powers more than 40% of all sites, does worse than the average of what it powers. That is the most striking result in the dataset.
The software is not the cause. A bare WordPress install with a plain theme is fast. What weighs is what piles up on top: a plugin installed in 2018 and never removed, a page builder, two SEO modules doing the same job. Each adds its own stylesheet and script. After a few years the page loads forty files before showing its first word, and the site owner has no idea.
The trend softens the picture: WordPress was at 46.3% in November 2025 and is at 49% in April 2026. It is repairing itself, slowly, as hosting and themes improve.
Why hosted platforms win
Duda, Wix and Shopify take the top three places, and the reason is simple: they decide for you. You cannot install a plugin that breaks rendering, pick slow hosting, or load four typefaces. The range of possible mistakes is closed off.
That is a constraint, and it is exactly what produces the result. A custom site inherits the decisions of whoever built it, for better and for worse. The freedom that lets you do better also lets you do worse.
Page weight is not the culprit you think
Shopify serves the heaviest pages in the dataset, 3.77 MB, and still passes at 79%. Astro serves the lightest, 1.65 MB, and passes at 67%. If weight were the deciding factor, the ranking would be reversed.
What matters more is the number of requests and when they are sent. Forty 40 KB files cost more than two 800 KB files: every request carries a fixed cost, and the browser cannot ask for everything at once. A tracking script placed before the content delays the entire render for 15 KB.
What a custom build actually buys
Not speed. The data is clear on that, and claiming otherwise would be easy to check.
What it buys is control over what destroys speed. On a site I build there is no plugin to disable, because there is none. No theme to slim down, because there is no theme. When a page slows, the cause is identifiable and fixable, and it does not depend on a third-party vendor's release schedule.
The other difference is the ceiling. A hosted platform takes you to a good level with no effort, then leaves you there. That is an excellent deal if that level is enough for you, and it is more often than people admit.
What to do about your own site
Before switching tools, measure. Most slow sites do not have a technology problem, they have an accumulation problem, and it gets solved without rebuilding anything.
- Test your site on PageSpeed Insights, and read the field data rather than the lab score: it is the first section, the one about your actual visitors.
- List your plugins and remove the ones whose purpose you no longer remember. It is the single highest-yield move.
- Compress your images and give them dimensions. An image without dimensions triggers a layout shift, so a poor CLS.
- Count your third-party scripts. Every analytics tool, every chat widget, every banner is paid for in milliseconds.
If your site passes all three thresholds after that clean-up, the platform question is settled. If nothing moves, then it genuinely arises.
Sources
The figures come from the Core Web Vitals Technology Report by HTTP Archive, April 2026 snapshot, which combines Chrome UX Report field measurements with HTTP Archive crawling. The web average is the May 2026 figure. This data is public and refreshed monthly: you can check it yourself.
If you want me to look at yours, performance optimisation is a service in its own right, quoted after measurement rather than before.
