Flawed markup is regarded in many projects as a cosmetic problem: as long as the page looks as desired in the browser, validation errors remain ignored. A study of high-reach websites counters this attitude with an uncomfortable number. 87 per cent of the top pages examined violate standards for HTML and CSS or requirements for accessibility.
The scale is remarkable because these are not improvised projects, but websites with high reach, professional maintenance and corresponding budgets. Standards compliance is therefore evidently not a resource problem, but one of prioritisation.
Why browsers forgive the errors – and other systems do not
Browsers are extremely fault-tolerant. HTML parsers repair incompletely closed elements, ignore unknown attributes and still render something sensible even when the structure of a document is technically incorrect. This leniency has grown historically and has made the web robust – but it conceals the fact that something is wrong beneath the surface.
Other consumers of web content do not have this leniency to the same extent. Screen readers – software that reads out screen content or outputs it in Braille – rely on the semantic structure of a document: heading levels, landmarks such as navigation and main content, correctly associated form labels, alternative texts for images. If this structure is missing or inconsistent, the result is not a nicely rendered stopgap, but an unusable page.
The new factor: machine reading
An aspect that hardly featured in older debates about validation is now gaining weight: the evaluability of websites by AI systems. Language models, assistants and automated agents access web content in order to answer questions, summarise information or carry out tasks. In doing so, they do not process the visual layout, but the underlying structure.
A document whose hierarchy cannot be discerned, whose tables do not identify header rows and whose central content lies in structureless container elements is harder to interpret. What is the main statement on this page, what is navigation, what is advertising, which price belongs to which product? Good markup performs precisely this assignment – and precisely this is missing when structure exists only visually.
What is remarkable here is the overlap: the measures that make a page usable for screen readers are largely the same ones that make it accessible for machine evaluation. Semantics, clear hierarchy, described controls, meaningful text alternatives. Accessibility was never merely a question of consideration towards a minority – it is a question of the interpretability of content, regardless of who or what reads it.
How errors get into projects
The typical causes are rarely ignorance in the narrower sense. More often they are effects of the division of labour and project dynamics:
- Component libraries and frameworks that generate generic markup and leave semantics to the user.
- Third-party code for consent banners, tracking, chat widgets or advertising, which lies outside one's own quality assurance but ends up in the delivered document.
- Editorial maintenance without guidelines: headings are chosen by font size instead of by hierarchy, alternative texts remain empty or repeat the file name.
- Missing checks in the delivery chain: validation and accessibility tests run once at acceptance, then never again.
- Changes built in later that arise under time pressure and dilute the originally clean structure.
What can be anchored in a relaunch
A relaunch is the cheapest moment to establish standards compliance not as final acceptance, but as an ongoing property of the website. Three levels in particular are practicable.
First, the components. If the semantics are once correctly embedded in reusable building blocks – buttons as buttons, form fields with associated labels, navigation areas as landmarks – correctness is inherited by every page that uses these building blocks. This is considerably more economical than subsequently correcting hundreds of individual pages.
Second, automation. HTML validation and automated accessibility checks can be integrated into the build and deployment chain so that regressions are noticed before they go live. Automated tests do not find everything – questions such as the meaningfulness of an alternative text or the comprehensibility of an error message remain human tasks. But they reliably catch the mechanically checkable errors, and these account for a considerable part of the problem.
Third, the editorial team. A content management system can either facilitate errors or prevent them. If heading levels are offered logically in the editor instead of as a format selection, if alternative texts are mandatory fields with an explanation, and if the system does not permit arbitrary layout nesting, quality arises without additional training. In TYPO3, such guard rails can be mapped via configured content elements and restricted rich-text configurations.
The economic framework
Beyond the technical argument, there is regulatory pressure: accessibility is a legal requirement for a growing circle of providers, not a voluntary additional service.
Anyone who has to meet standards anyway should not do the work twice – once for the compliance check and once for findability in AI-powered searches and assistants.
And there is a visibility aspect. If users increasingly obtain information via assistants, it is no longer solely the ranking in a results list that determines whether an offering is noticed, but also whether a system can correctly capture and summarise the content. A page that is difficult for machines to read loses reach at this point without it being immediately noticeable in classic analyses.
Classification for web and software projects
The finding that almost nine out of ten large websites are faulty absolves no one – it merely shows that validation has fallen out of the process almost everywhere. Anyone who anchors standards and accessibility in the next relaunch as an automated, permanently checked requirement rather than as an acceptance point simultaneously gains legal certainty, better usability and better machine evaluability. That is the same effort with three effects – and therefore one of the few items in the project budget that pays off multiple times.
Sources
- Errors in HTML, CSS and accessibility – 87 percent of top sites faulty — heise developer News