&w=3840&q=80)
Despite the heightened focus on accessibility, the web didn’t get more accessible this year. In fact, data shows the opposite – it got less accessible.
WebAIM has spent eight years evaluating the home pages of the top one million websites. Its 2026 report, published in March based on a February 2026 scan, shows something that hasn’t happened in the seven previous editions: a clear reversal. After six straight years of small, steady improvements, accessibility on the world’s most visited home pages got measurably worse.
This comes at an awkward time. It’s now just over a year since the European Accessibility Act (EAA) came into effect across the EU on 28 June 2025, and failure to comply now carries real financial and legal consequences. Many organisations spent the run-up to that date scrambling to get their digital products in order. This data suggests that progress hasn’t held.
The headline numbers
The share of home pages with at least one detected WCAG 2 failure climbed from 94.8% to 95.9%, breaking a six-year streak of small year-over-year improvements. Because WebAIM’s tool only catches automatically detectable failures, the true rate of full conformance is certainly lower than the 4.1% of pages that passed.
Average detected errors per page rose 10.1%, from 51 to 56.1. Across the million-page sample, that adds up to over 56 million distinct accessibility errors.
The most common issues haven’t changed in seven years: low contrast text (83.9% of pages, up sharply from 79.1%), missing alt text on images (53.1%, though this metric actually improved), missing form labels (51%), empty links (46.3%), and empty buttons (30.6%). These six categories account for 96% of all detected errors. Fixing just them would meaningfully improve the state of the web.
Why this happened: complexity, ARIA & vibe coding
The clearest driver in the data is complexity. The average home page grew 22.5% in element count in a single year, reaching 1,437 elements. Pages are not just getting bigger, but also more technologically layered, leaning more heavily on third-party frameworks, libraries and components.
ARIA usage tells a similar story, and it’s the most striking finding in the whole report. ARIA attribute use jumped 27% year over year, now averaging over 133 attributes per page, six times what it was in 2019. You’d expect that to help. It didn’t. Pages using ARIA averaged 59.1 errors, compared to 42 on pages without ARIA at all.
To put it plainly: wrong ARIA is worse than no ARIA. Misapplied ARIA roles, states and properties tell screen readers and other assistive technology that a page works one way when it actually behaves differently, which is a worse outcome for a disabled user than a plain, unenhanced page that simply behaves as expected.
WebAIM’s own conclusion points to the likely cause: increased reliance on third-party code and “automated or AI-assisted coding practices”, what’s now commonly called vibe coding. Pages are being built faster, with more generated boilerplate, and accessibility fundamentals aren’t keeping up.
That diagnosis lines up with a separate, complementary benchmark worth knowing about: AIMAC, the AI Model Accessibility Checker from the GAAD Foundation, built in partnership with ServiceNow.
AIMAC takes a different angle than WebAIM. Rather than scanning live websites, it prompts dozens of current AI models to generate web pages from scratch, with no accessibility instructions given, and then audits the resulting code with axe-core.
The results are exactly what you’d expect given the WebAIM findings: wildly inconsistent outcomes between models, with some generating reasonably clean HTML by default and others producing pages with serious, structural accessibility violations as a matter of course.
If AI is going to write more and more of the web’s front-end code, and current trends suggest that it will, the accessibility habits baked into these models matter just as much as the habits of the developers using them.
The smaller signals worth watching
A few other details in the report are easy to miss but matter in practice. Skipped heading levels, which break the navigation structure screen reader users rely on, were present on 41.8% of pages, up from 39%. One out of ten “skip to content” links was broken. And ecommerce platforms and several popular JavaScript libraries and ad networks correlated with notably higher error counts, reinforcing that the tools and dependencies a team chooses have real accessibility consequences, often invisible to the people choosing them.
On the positive side, missing alt text and missing document language both continued to improve, and sites built on platforms like Astro, Next.js and React, along with CMS platforms like Adobe Experience Manager and Squarespace, had meaningfully fewer errors than average. Good defaults in the right tools still help.
What this means for you
The pattern across both WebAIM and AIMAC’s data is the same: speed and complexity are increasing faster than accessibility discipline is keeping up, and “the AI wrote it” or “the framework handles it” is not a valid substitute for testing.
We help teams catch exactly this kind of gap, through accessibility audits, WCAG 2.2 and EAA-focused remediation, and developer training that builds these habits in rather than bolting them on at the end.
If you’re not sure where your own site stands a year into EAA enforcement, that’s worth finding out before a regulator, a lawsuit, or a lost tender does it for you. Learn more about our accessibility services or get in touch with one of our certified accessibility specialists.
