Web Accessibility Statistics 2026: The Numbers That Hold Up

The web got less accessible in 2026. WebAIM's annual scan of the top million home pages found 95.9% had detected WCAG 2 failures, up from 94.8% a year earlier and the first increase since 2020, after five straight years of small improvements (WebAIM Million, 2026).

Web accessibility statistics are unusually full of rubbish. The same handful of numbers get copied between blogs for years after their source went offline, lost its context, or turned out to be a vendor's marketing estimate. So this post has one rule: every figure below is traced to a primary source, with its methodology and the date it was measured. Where a famous number doesn't survive that test, we say so and explain where it came from. There's a whole section on the ones you should stop quoting.

TL;DR: 95.9% of the top million home pages have detected WCAG failures, up from 94.8% and the first rise since 2020 (WebAIM, 2026). Low-contrast text is the largest single mover, spreading from 79.1% to 83.9% of pages. Errors per page also rose, to 56.1, but pages grew 22.5% bigger, so error density actually fell. In the US, cognitive disability is the largest category at 13.9% of adults against 5.5% for vision (CDC BRFSS, 2022). Most accessibility tooling serves the smaller group. Pages using ARIA average 40.7% more errors than pages without it.

How Many Websites Fail Accessibility Testing?

Almost all of them. In 2026, 95.9% of the top one million home pages had detected WCAG 2 failures, with an average of 56.1 errors per page, up about 10% from 51.0 in 2025 (WebAIM Million, 2026). WebAIM runs the WAVE engine over the Tranco top million, evaluating the rendered DOM after scripts and styles have applied.

Be careful how you read that spike. 2026 is not the worst year on record. At 95.9% it's identical to 2024, and 2020 was worse at 98.1% (WebAIM Million, 2021 and 2024 reports). The 1.1-point rise precisely cancels the 1.1-point gain made in 2025, so the web is back where it was two years ago. One year of progress erased, not an all-time low. Anyone telling you the web has never been less accessible is overselling it.

Citation capsule: WebAIM's 2026 scan of the top one million home pages found 95.9% had detected WCAG 2 failures, up from 94.8% in 2025, the first increase since 2020 after five straight years of improvement (WebAIM Million, 2026). The 2026 figure returns the web to its 2024 level rather than setting a record; the worst year in the series was 2020, at 98.1%.

Did the Web Get Worse, or Just Bigger?

Both, and they're different numbers measuring different things. This is where nearly every write-up of the WebAIM data goes wrong, so it's worth slowing down. The 95.9% and the 56.1 are not the same story.

95.9% is a rate: the share of home pages with at least one detected failure. A page needs only a single error to count. The largest single mover behind it was low-contrast text, which spread from 79.1% to 83.9% of pages (WebAIM Million, 2026), a 4.8-point jump that dwarfs every other change.

56.1 is a count: the average number of errors per page. That rose about 10% from 51.0. But the average home page also grew 22.5% bigger in a single year, to 1,437 elements (WebAIM Million, 2026).

Do the division and something surprising falls out. Error density was 3.9% of page elements in 2026, against roughly 4.3% in 2025. Per element, markup actually got slightly better. Developers didn't get worse at writing accessible components; they shipped a lot more components, and even a slightly improved failure rate applied to a much bigger page yields more bugs. Home pages gained 264 elements in twelve months.

Be careful about how far you push that, though, because the two numbers are decided by different populations. The count is an average across every page. The rate is settled entirely at the margin, by the small share of pages that previously had zero detected errors, and a bigger page gives a previously-clean page more chances to trip its first error. So page growth plausibly nudges the rate too. WebAIM doesn't publish a decomposition, so neither we nor anyone else can tell you exactly how the 1.1 points split between contrast and sheer page size.

What we can say is which lever to pull. The rising error count is a component-sprawl problem. The rising failure rate has contrast as its biggest single component. Treat them as one number and you'll fix the wrong thing.

What Actually Fails?

Six things, overwhelmingly. Six failure types account for 96% of all detected errors, and they've been the same six for seven consecutive years (WebAIM Million, 2026).

Look at which way each one moved, because there's a pattern in it. Only two improved: missing alt text (55.5% to 53.1%) and missing document language, which has fallen steadily to 13.5%. Four got worse, led by low-contrast text.

Here's the pattern, and it isn't the obvious one. It's tempting to say "what gets linted gets fixed", but that can't be right: form labels, empty links and empty buttons all got worse, and all three are caught by axe and by eslint-plugin-jsx-a11y at author time. Plenty of linting, no improvement.

A better split is per-document rules versus per-element rules. <html lang="en"> is a single attribute on a single tag, once per page. Fix it and it stays fixed, however big the page grows, which is why it's the only failure with a clean, unbroken improvement record. The interactive per-element rules behave the opposite way: labels, links and buttons scale with element count, so a page that gained 264 elements gets 264 new chances to ship an unlabelled one. All three lost ground in 2026. And colour contrast isn't really a code problem at all: it's a design decision made in Figma months before anyone runs a scanner, and no linter can veto a brand palette.

Missing alt text is the honest exception, and it's the most interesting number on the chart. It's a per-element rule, so the framework above says it should be getting worse with page growth. It improved anyway, 55.5% to 53.1%. The plausible reason is that alt text is the one per-element rule with a decade of CI checks, linter rules, and CMS fields that nag you at upload time behind it. Enforcement at the point of authoring beat page growth. Nothing else among the per-element rules has anything like the same tooling, and nothing else among them improved.

If you want the fixes, our guide to fixing common WCAG violations walks through all six with code, and the GitHub Actions guide covers gating them in CI.

The ARIA Paradox: More ARIA, More Errors

Pages that use ARIA have 40.7% more detected errors than pages that don't: 59.1 against 42.0 (WebAIM Million, 2026). ARIA adoption is now at 82.7% of home pages, up from 74.6% in 2024, and the error gap has widened over the same period, from 34.2% to 40.7% (WebAIM Million, 2024 and 2026 reports).

The honest caveat, which WebAIM makes itself: this does not prove ARIA causes the errors. Pages that use a lot of ARIA tend to be more complex, and complexity brings its own bugs.

But there's direct evidence of misuse sitting alongside the correlation. 5.7% of home pages use role="menu", and 22% of those menus introduced accessibility barriers through incomplete markup and missing keyboard interactions (WebAIM Million, 2026). Meanwhile home pages carry an average of 23.3 aria-hidden="true" instances each (WebAIM Million, 2026). A lot of the page is being hidden from assistive technology, and not always on purpose.

The underlying reason is that ARIA describes behaviour without implementing any of it, so a role is only ever a claim you still have to honour in code. We break down the specific patterns, and the fixes, in ARIA mistakes developers make.

Citation capsule: Home pages using ARIA averaged 59.1 detected errors in 2026 against 42.0 for pages without it, a 40.7% penalty, widening from a 34.2% gap in 2024 even as ARIA adoption rose from 74.6% to 82.7% of pages (WebAIM Million, 2026). WebAIM cautions this is correlation rather than proven causation, but 22% of the pages using role="menu" introduced barriers through incomplete menu markup.

How Many People Have a Disability?

This is where most statistics posts quietly mislead you. There is no single answer, and the four figures below do not disagree with each other. They measure four different things.

Pick whichever number flatters your argument and you can claim disability affects anywhere from a sixth to well over a quarter of people. The honest version: the differences are almost entirely definitional, not real. A "significant disability" (WHO) is a much higher bar than "serious difficulty concentrating" (CDC), and the population bases differ too, since the CDC figure covers adults only.

So quote whichever one matches your market, but say which definition you're using. That single sentence is what separates a credible statistic from a decorative one.

Which Disabilities Are Most Common?

Not the ones you'd guess from how the industry talks. In the US, cognitive disability is the largest single category at 13.9% of adults, well ahead of mobility (12.2%) and more than double vision (5.5%) (CDC BRFSS, 2022, age-adjusted prevalence).

Now compare that to where the industry points its effort. Screen readers, ARIA, alt text, and semantic markup dominate the tooling, the conference talks, and the WCAG success criteria, and they mostly serve blind and low-vision users: the 5.5%. The 13.9% with cognitive disabilities get plain language, consistent navigation, generous timeouts, and error recovery, which are the criteria that are hardest to test automatically and easiest to skip.

The UK data points the same way. Among working-age disabled adults, the most common impairment types are mental health (46%) and mobility (40%) (DWP Family Resources Survey, 2024-25).

Two honest caveats, since this post is meant to hold itself to its own standard. The CDC's cognitive domain is broad and self-reported ("serious difficulty concentrating, remembering or making decisions"), so it is not a like-for-like swap with blindness, and population size is not the same as severity of web barrier. And mobility, at 12.2%, is well served by the keyboard-operability criteria that often get filed under "screen reader work" anyway.

None of this argues for doing less screen-reader work. It argues that "accessibility" has quietly become a synonym for "screen reader support" in most engineering orgs, while the largest group of affected users sits outside that synonym entirely.

What Do Screen Reader Users Actually Use?

JAWS and NVDA, mostly, and they'd rather you fixed your website than improved their software. WebAIM's most recent Screen Reader User Survey (n=1,539) puts primary desktop screen reader share at JAWS 40.5%, NVDA 37.7%, VoiceOver 9.7%, with Narrator at 0.7% (WebAIM Screen Reader Survey #10, 2024).

Three findings worth building around:

• 91.3% use a screen reader on a mobile device, where VoiceOver leads at 70.6% and TalkBack at 34.7%. Mobile is not a secondary surface for these users.
• 71.6% navigate a long page by headings first. Your heading hierarchy is the primary navigation for most screen reader users, which is a good reason to treat it as an interface rather than as styling.
• 85.9% say more accessible websites would have a bigger impact than better assistive technology. The users themselves are telling you the fix belongs in your code, not in their software. That's the most direct available rebuttal to accessibility overlays and to the "AI will patch it at runtime" pitch.

One caveat you won't find in other statistics posts, because it requires actually checking: there is no Survey #11. As of July 2026, Survey #10 (fieldwork December 2023 to January 2024) is still the latest (WebAIM, 2026). The industry's most-cited assistive technology dataset is two and a half years old. If you see "2026 screen reader statistics" anywhere, they are either recycled from Survey #10 or invented. WebAIM also notes the sample was uncontrolled and may not represent all screen reader users.

Which Sectors Are Worst?

Retail and sport; and the regulated sectors are the best. Government home pages average 42.4 errors, 24.4% below the overall mean, with non-profit (43.0) and education (48.9) close behind (WebAIM Million, 2026).

The sectors under the heaviest legal obligation are the most accessible, and the ones selling you trainers are the worst. It's tempting to read that as proof that regulation works, and it might be, ahead of the ADA Title II deadline and a year into European Accessibility Act enforcement.

But hold that lightly, because there's an obvious confound and it's the one this post already raised. Government and education sites are also far less bloated than retail sites: fewer third-party widgets, less ad tech, less marketing JavaScript. Given that errors scale with element count, the sector ranking is roughly what page complexity alone would predict, with no reference to the law at all. Two explanations, same data, and this dataset cannot separate them.

It does at least sit consistently with where the litigation lands. Shopping sites are among the most broken in this dataset, and website accessibility filings rose 27% in 2025 to 3,117 in US federal courts (Seyfarth Shaw, 2026), overwhelmingly against retail defendants.

What Does the Legal Data Show?

Filings are climbing. Plaintiffs brought 3,117 federal website accessibility lawsuits in the US in 2025, up 27% from 2,452 in 2024, and website cases were 36% of all federal ADA Title III filings (Seyfarth Shaw, 2026). We break that down in ADA website lawsuits.

Europe produced the year's most significant ruling, and it did not come from a regulator. On 4 June 2026, the Tribunal judiciaire de Caen ordered Carrefour to make its website and mobile app fully accessible within six months, with a daily penalty for delay (Droit Pluriel, 2026). The case was brought by the disability organisations apiDV and Droit Pluriel, not by a state enforcement body.

The detail that should worry anyone tracking a compliance percentage in a dashboard: Carrefour argued it met 71% of RGAA criteria and that this was sufficient. The court rejected that outright, holding that accessibility is an obligation of result. A site cannot be somewhat accessible.

One thing we could not find, despite looking: any publicly confirmed fine issued by an EU member-state regulator under national European Accessibility Act law. Regulators in the Netherlands, Sweden and Ireland are running audits and processing complaints, but the first real teeth came from private litigation in France. Several vendor blogs describe the Carrefour case as regulator enforcement. It wasn't. See our EAA developer guide for what the law actually requires.

What About Mobile Apps?

Honestly? Nobody knows, and that's the finding. There is no WebAIM Million for mobile apps. No large-scale, repeatable, methodologically transparent scan of app accessibility exists, which is remarkable given that 91.3% of surveyed screen reader users also use one on a mobile device.

The best available study is small. ArcTouch expert-tested 50 apps (25 iOS, 25 Android) across five industries against four assistive technologies, scoring each 0-100 (ArcTouch, 2025). Only 2 of the 50 scored above 85; 9 failed outright. Counterintuitively, Android averaged 56 and iOS 48, so the platform with the better accessibility reputation did worse in this sample. Shopping apps were the worst category at 41.

Treat that as a signal, not a census. It's an agency's own study with n=50, which is more transparent than most vendor research but nowhere near a WebAIM-scale dataset. If you want a genuinely useful research project in this space, that's the gap.

Five Accessibility Statistics You Should Stop Quoting

Here's the section that should make this post worth linking to. Every number below is widely repeated, and every one falls apart when you chase it to its source.

1. "$6.9 billion is lost annually to inaccessible e-commerce." This comes from a Nucleus Research study commissioned by Deque and announced in 2019. Per Deque's own write-up, it interviewed 73 blind or severely vision-impaired US adults, and the $6.9bn referred to potential North American revenue at the top ten retailers, with the calculation never shown (Deque, 2019). Deque's page now states the report is no longer available, so the primary source is offline entirely. A seven-year-old, vendor-commissioned, n=73 study that nobody can now read is not a statistic. Stop using it.

2. "71% of users with disabilities leave a website they find difficult." Real study, routinely mangled. It's the UK Click-Away Pound survey of 2016, and it found that 71% of UK disabled internet users with access needs click away. Not 71% of disabled people, and not global. The 2019 follow-up survey, a fresh sample rather than a correction, found 69% (Click-Away Pound 2019). So the number is a decade old, and the more recent wave of the same research says something different.

3. "The click-away pound is worth £17.1 billion." That figure is from the 2019 Click-Away Pound report, not the 2016 one that gets cited alongside it (2016 estimated £11.75bn). It's an extrapolation, users multiplied by average online spend, and the authors say so plainly, cautioning that the figures "are extrapolated from the Survey's findings so care must be taken when considering them." It's a modelled market-size estimate rather than measured lost revenue, and the data behind it is from 2019.

4. "Only 3% of the web is accessible." The number was roughly right for its vintage; the word "accessible" is the lie. It traces to WebAIM's share of pages with no detected errors, which sat around 2-3% in the 2021 and 2022 reports and is 4.1% today. But WebAIM explicitly warns that "absence of detected errors does not indicate that a page is accessible", and that true WCAG 2 A/AA conformance is "certainly lower than 4.1%" (WebAIM Million, 2026). So the stat quietly asserts the one thing its own source denies: that an automated scan can certify a page as accessible. Cite the no-detected-errors figure by its real name, or don't cite it.

5. "1 billion people, 15% of the world, have a disability." That's the 2011 WHO/World Bank figure. WHO's current fact sheet (March 2023) says 1.3 billion, 16%, one in six (WHO, 2023). If a post still quotes 1 billion in 2026, it hasn't checked a source in fifteen years.

And one we're turning on a source we'd have liked to use. The widely-cited shift-left economics (a defect costing "$350 less" to fix in development than in QA, and production fixes running to "$800 or more") comes from a Deque blog post with no stated methodology, no sample size, and no cited study, attributed to experience with "hundreds of content production teams". The direction is almost certainly right. The precision is not earned. A post arguing that every number needs a source can't then quote an unsourced one approvingly, even when it agrees with it.

Frequently Asked Questions

What percentage of websites are not accessible?

95.9% of the top one million home pages had detected WCAG 2 failures in 2026, up from 94.8% in 2025 (WebAIM Million, 2026). Only 4.1% had no detected errors, and WebAIM cautions that true WCAG 2 A/AA conformance is "certainly lower than 4.1%", since automated tools catch only a subset of failures.

How many people have a disability?

It depends entirely on the definition. WHO says 16% globally experience significant disability; Eurostat says 23.9% of EU adults report an activity limitation; the UK says 25% under the Equality Act; the CDC says 28.7% of US adults report a functional difficulty. These measure four different things and are not comparable to each other.

What are the most common accessibility errors?

Six types cause 96% of all detected errors: low-contrast text (83.9% of home pages), missing alt text (53.1%), missing form labels (51.0%), empty links (46.3%), empty buttons (30.6%) and missing document language (13.5%) (WebAIM Million, 2026). All six are automatically detectable. Five are quick code fixes; contrast usually means changing a design token, so it needs a designer in the room.

Does using ARIA make a site more accessible?

Not automatically. Home pages using ARIA averaged 59.1 detected errors in 2026 against 42.0 for pages without it, a 40.7% penalty (WebAIM Million, 2026). WebAIM notes this is correlation, not proven causation, but 22% of pages using role="menu" introduced barriers through incomplete markup. Bad ARIA is worse than no ARIA.

Which screen readers do people use?

JAWS leads at 40.5% of primary desktop use, then NVDA at 37.7% and VoiceOver at 9.7% (WebAIM Screen Reader Survey #10, 2024). 91.3% also use a screen reader on mobile, where VoiceOver leads. Note this data is from December 2023 fieldwork; no newer WebAIM survey exists as of July 2026.

The Bottom Line

Read the 2026 web accessibility statistics carefully and they say two separate things, not one. The failure rate rose to 95.9%, back to its 2024 level, and low-contrast text is the largest single mover behind it. The error count rose to 56.1 per page, but pages grew 22.5% bigger, so per element the markup actually improved slightly (WebAIM Million, 2026). Conflate those two and you'll fix the wrong thing.

The most useful findings here aren't the headline. The failures that improved are the ones with enforcement at the point of authoring, a document-level rule for page language and a decade of tooling for alt text, while the interactive rules lost ground to page growth. The largest group of disabled users is cognitive, not visual. And screen reader users themselves say, by 85.9%, that fixing websites would help them more than improving their software. Start by scanning your own site, fix the six failures behind 96% of errors with our common WCAG violations guide, and gate them in CI using the GitHub Actions setup.

And if you quote a statistic from this page, quote its source too. That's the whole point of it.

Sources

• WebAIM, The WebAIM Million: The 2026 Report on the Accessibility of the Top 1,000,000 Home Pages, retrieved 2026-07-13, https://webaim.org/projects/million/ (source for the 2025 and 2026 data points in the trend chart)
• WebAIM, The WebAIM Million 2024 Report, retrieved 2026-07-13, https://webaim.org/projects/million/2024 (source for the 2023 and 2024 data points)
• WebAIM, The WebAIM Million 2022 Report, retrieved 2026-07-13, https://webaim.org/projects/million/2022 (source for the 2022 data point)
• WebAIM, The WebAIM Million 2021 Report, retrieved 2026-07-13, https://webaim.org/projects/million/2021 (source for the 2019, 2020 and 2021 data points)
• WebAIM, Screen Reader User Survey #10 Results, retrieved 2026-07-13, https://webaim.org/projects/screenreadersurvey10/
• WebAIM, Screen Reader User Survey index (confirming no Survey #11 as of July 2026), retrieved 2026-07-13, https://webaim.org/projects/screenreadersurvey/
• World Health Organization, Disability fact sheet, 7 March 2023, retrieved 2026-07-13, https://www.who.int/news-room/fact-sheets/detail/disability-and-health
• Eurostat, Population with disability (EU-SILC, 2024 reference year), retrieved 2026-07-13, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Population_with_disability
• UK Department for Work and Pensions, Family Resources Survey financial year 2024 to 2025, published 26 March 2026, retrieved 2026-07-13, https://www.gov.uk/government/statistics/family-resources-survey-financial-year-2024-to-2025/family-resources-survey-financial-year-2024-to-2025
• US Centers for Disease Control and Prevention, Disability and Health Data System (DHDS), Prevalence of Disability Status and Types, BRFSS 2022 age-adjusted prevalence, retrieved 2026-07-13, https://data.cdc.gov/resource/s2qv-b27b.json
• Seyfarth Shaw LLP, Federal Court Website Accessibility Lawsuit Filings Bounce Back in 2025, retrieved 2026-07-13, https://www.adatitleiii.com/2026/03/federal-court-website-accessibility-lawsuit-filings-bounce-back-in-2025/
• Droit Pluriel, Décision du Tribunal judiciaire de Caen, apiDV et Droit Pluriel c. Carrefour, retrieved 2026-07-13, https://droitpluriel.fr/decision-du-tribunal-judiciaire-de-caen-apidv-et-droit-pluriel-c-carrefour-une-decision-de-justice-majeure-pour-laccessibilite-numerique/
• ArcTouch, The State of Mobile App Accessibility, retrieved 2026-07-13, https://arctouch.com/state-of-mobile-app-accessibility
• Deque Systems, Research Shows the Internet is Unavailable to Blind Users (origin of the $6.9bn figure), retrieved 2026-07-13, https://www.deque.com/blog/research-shows-internet-is-unavailable-to-blind-users/
• Deque Systems, Doing the Numbers: Digital Accessibility and Shifting Left, retrieved 2026-07-13, https://www.deque.com/blog/doing-the-numbers-digital-accessibility-and-shifting-left/
• Click-Away Pound Survey 2016 Final Report (origin of the "71%" figure), retrieved 2026-07-13, https://www.clickawaypound.com/cap16finalreport.html
• Click-Away Pound Survey 2019 Final Report (origin of the 69% figure and the £17.1bn extrapolation), retrieved 2026-07-13, https://www.clickawaypound.com/downloads/cap19final0502.pdf