Web accessibility statistics 2026: what the data actually shows

The headline number, and the part that matters more

95.9% of the top one million home pages had detected WCAG 2 failures, up from 94.8% in 2025. In total, 56,114,377 distinct errors were detected across the sample, an average of 56.1 per home page.

The percentage is not the interesting part. It has sat above 94% every year the study has run, and a number that high stops carrying information. The rise reversed six consecutive years of small improvements, which is the finding rather than the headline number.

Home pages also got more complex. Average page element counts rose 22.5% in a single year, which is the likeliest mechanism: more elements means more places for a failure to occur, and the failure rate rose alongside the complexity rather than independently of it.

That is worth sitting with, because it is not a story about anyone getting lazier. Home pages in 2026 carry more third-party embeds, more consent tooling, more chat widgets and more marketing components than they did a year earlier. Every one of those is markup a site owner did not write and often cannot edit. A team can improve everything under its own control and still watch its error count rise, because the growth is happening in the parts of the page it does not own.

What actually fails, and why that is good news

Six failure types accounted for 96% of all detected errors. Low contrast text appeared on 83.9% of home pages, missing alternative text on 53.1%, and missing form input labels on 52.8%.

The useful part for a site owner is that the most common failures are also the cheapest to fix. Take the three largest in turn.

Low contrast text is a colour value. WCAG 2 at level AA asks for a contrast ratio of at least 4.5:1 between normal-size text and its background, and 3:1 for large text. Browser developer tools report the ratio for any element you inspect, so this is checkable in seconds and fixable by changing a hex code. It appears on 83.9% of home pages mostly because light grey on white reads as tasteful in a design tool and unreadable on a phone in daylight.

Missing alternative text is a sentence somebody writes. The fix is not automatic: an image that carries meaning needs text describing what it conveys, and an image that is purely decorative needs an empty alt attribute so a screen reader skips it. Both are a judgement, which is exactly why a machine can tell you the attribute is absent but cannot tell you whether the one that is there is any good.

Missing form labels are an attribute. A placeholder is not a label. It disappears the moment someone types, it is often too low-contrast to read, and it is not reliably announced. A visible label bound to the input is the fix, and it makes the form easier for everyone rather than only for people using assistive technology.

None of these require an architectural change, a replatform, or a specialist retainer. Together with three others they account for 96% of what automated testing detects. That reframes the problem. The common failures are not hard engineering problems. They are decisions nobody made, on pages nobody checked.

Does the platform matter?

Somewhat, and not in the direction a web studio would find convenient. By content management system, WordPress home pages (252,302 sampled) averaged 52.8 detected errors, Drupal (18,222) 41.2, Wix (3,183) 33.3 and Squarespace (2,669) 33.0.

Squarespace and Wix scored better than WordPress. The likely reason is the same constraint that frustrates people who outgrow those platforms: a hosted builder decides your markup, and markup you cannot edit is markup you cannot break. If you are never going to change anything structural, that limitation is working in your favour.

The framework breakdown runs the other way. By JavaScript framework, Next.js home pages (23,863 sampled) averaged 40.9 detected errors, 27.1% below the sample average, with Astro at 9.0 and React at 43.5. And in commerce the gap is wider still. Among ecommerce platforms, Shopify home pages (42,516 sampled) averaged 75.1 detected errors, 33.9% above the sample average, with Magento at 75.8.

Read all of that carefully. Association, not causation. CMS choice correlates with site age, budget and builder skill, none of which were controlled for. A newer framework attracts newer sites built by better-resourced teams, and that alone would produce the same pattern. These figures describe populations of sites. They are not evidence that any particular agency, including this one, builds accessible websites.

The legal picture, stated carefully

Federal district courts recorded 9,718 civil filings under the nature-of-suit code ADA-Other in the twelve months ending 30 September 2025, up from 9,030.

Filings, not judgments, and the code covers more than website accessibility. Report it as filing volume and never as legal advice or as a prediction that any given business will be sued. We are engineers, not lawyers. What we can tell you is what the court statistics count and what they do not, and ADA-Other is a broad code. It is not a count of web accessibility lawsuits specifically.If your exposure is a real question for your business, that is a conversation for a lawyer in your jurisdiction, not for a web studio's blog.

What we will say is that the vendor version of this statistic is usually presented without either caveat. A figure in the high thousands, attached to the word lawsuits, attached to a product page, is a sales instrument rather than a measurement. The number is real. What it counts is narrower than the sentence around it usually implies, and you are entitled to know which is which before anyone quotes you for a fix.

What to do with this, in order

The same order we use for performance work, for the same reason: measure the site you actually have before deciding what to change about it.

• ✓
  Run an automated check on your own most important pages, not just the home page. Home pages are the least representative page on most sites, and every figure in this article is home-page data.
• ✓
  Fix contrast, alternative text and form labels first. They are the most common failures in the data and the cheapest to change, whatever platform you are on.
• ✓
  Test with a keyboard. Tab through a form and a navigation menu without touching the mouse. This takes two minutes and catches things no automated tool reports.
• ✓
  Only then consider whether anything structural needs to change. Most sites in this data do not have an architecture problem, they have an unchecked-pages problem.

One thing worth naming, because it comes up in every quote: an overlay widget that promises instant compliance is not the same as fixing the page. It sits on top of markup it did not author and tries to correct it at runtime. Whatever you think of the category, an overlay does not change the contrast ratio your designer chose or write the alternative text nobody wrote. Those remain your decisions, on your pages.

What this data cannot tell you

Every figure above comes from automated testing, and Automated testing detects a subset of WCAG failures. A page with no detected errors is not necessarily conformant. A page can pass every automated check and still be unusable with a screen reader, because the things that matter most, such as whether alternative text is accurate or whether a reading order makes sense, cannot be judged by a machine.

It also cannot tell you about your site. This is a million home pages, sampled once, in February 2026. It establishes what is normal across the web. It establishes nothing at all about the pages your customers actually use.

The home-page-only sampling matters more than it sounds. A home page is usually the most designed and least functional page on a site. The pages where accessibility failures actually cost someone something are checkouts, booking forms, account settings and search results, and none of those are in this data. The reasonable inference is that the figures here are a floor rather than a ceiling, but that is an inference, and we are labelling it as one.

And it cannot tell you about people. Every number on this page counts machine-detectable defects in markup. Not one of them counts a person who gave up on a form. That measurement exists, but it is made by watching people use your site, and no million-page scan is a substitute for it.

Sources

Every figure in this article, with its method and the date we verified it:

• The WebAIM Million 2026 , WebAIM, Institute for Disability Research, Policy and Practice, Utah State University. Published 2026. Verified 2026-08-13. Automated evaluation with the WAVE stand-alone API against the rendered DOM after scripting and styles were applied. Sample: 1,000,000 home pages, tested February 2026.
• Federal Judicial Caseload Statistics , Administrative Office of the United States Courts. Published 2025. Verified 2026-08-13. Case counts by nature-of-suit code, twelve months ending 30 September 2025.

Frequently asked questions

Related reading: WordPress vs Next.js covers the same CMS and framework data in the context of a platform decision, and Core Web Vitals explained applies the same measure-before-you-change discipline to performance.