Only 341 of the web’s biggest home pages follow the HTML standard all the way through. That is the count from ValidateHTML, a project by independent French developer Théo Ducreux, which crawled the 5,000 most-visited domains on the Tranco list and validated the 2,656 that served a page meant for people.
The rest violated the W3C and WHATWG specifications somewhere on their home page, 87.2% of the sample. Hold sites to best-practice rules as well and only 68 sites, or 2.6%, came back with zero errors and zero warnings. Across the sample, Ducreux tallied 100,305 HTML violations and 18,863 CSS errors, according to The Register, which wrote up the findings on Monday. The figures come from an Aug. 14 crawl that the project plans to repeat with the same method.
The distribution is lopsided. The median site had 11 errors, the mean was 37.8, and the worst had 1,392. The most common violation, on 59.5% of sites, was an element placed where the spec doesn’t allow it, such as a style element inside a div. “Nobody writes that by hand,” Ducreux wrote. “It is a nesting problem created at build time, not in an editor.” In other words, front-end frameworks composing components without knowing where they’ll land.
CSS failed differently. About 64.5% of sites had at least one CSS error, but the median count was one, usually an invalid property value that the browser silently drops.
Where it hurts
Browsers are forgiving, which is why the errors pile up. “Browsers are so good at error recovery that nothing forces anyone to fix either, which is exactly why it accumulates,” the report says. Assistive technology is less tolerant. Missing alt text showed up on 20.4% of sites and landmark regions without ARIA labels on 41.6%, The Register reported, while skipped heading levels, unlabelled inputs and links with no text each appeared on more than a third. With the European Accessibility Act applying since June 2025, Ducreux calls that “a compliance surface, not a style preference.”
He doesn’t argue that invalid markup is breaking the web. “Honestly, most invalid HTML doesn’t break anything visible, browsers are built to guess what you meant and just patch it silently,” he told The Register by email. His concern is everything else that reads pages, from screen readers to software agents. “Writing to spec is what makes a page mean the same thing to all of them, not just to whoever’s rendering it in Chrome that day.”
The crawler and scoring are his own code, built on open-source validators including html-validate and Lightning CSS. “No team, no funding, no company,” he said. The Register also pointed to Shopify, whose engineers moved back toward plain, readable HTML templates this summer in part because AI models work better with it.


WebProNews is an iEntry Publication