· Research
What the Latest Research Tells Us
We have better accessibility testing tools than ever before.
We have mature international accessibility standards. Browsers and assistive technologies continue to improve. Developers have access to accessible component libraries, automated testing, browser extensions, linting tools, artificial intelligence, and decades of documentation explaining how to make the web more inclusive.
Governments are also putting considerably more weight behind digital accessibility requirements.
So, is the web becoming more accessible?
By some commonly cited measurements, particularly large-scale automated testing, we're moving backward.
In February 2026, WebAIM tested the homepages of one million of the world's most popular websites. Its automated testing detected more than 56 million accessibility errors, averaging 56.1 errors per homepage.
More strikingly, 95.9% of those homepages contained automatically detectable WCAG failures.
A year earlier, that number was 94.8%.
After several years of gradual improvement, the 2026 results reversed that trend.
That statistic deserves attention, but it doesn't tell the entire story.
Research published over the past year has examined accessibility across government, higher education, libraries, e-commerce, organizational policy, artificial intelligence, and the way researchers themselves evaluate accessibility for people with disabilities.
Taken together, the research paints a more complicated—and more useful—picture of web accessibility in 2026.
The problem isn't simply that developers occasionally forget alternative text.
We're still making many of the same basic accessibility mistakes we've been making for years. At the same time, websites are becoming dramatically more complex. Organizations continue to struggle to make accessibility part of their normal processes. Automated testing can identify enormous numbers of problems while still being unable to tell us whether a website is genuinely usable by a disabled person.
And now AI-generated and AI-assisted code is entering that already complicated environment.
For this review, we examined web and digital accessibility research published primarily between September 2025 and September 2026, supplemented with earlier research where necessary to establish longer-term trends.
Here's what the research tells us.
1. Most of the Web Still Has Detectable Accessibility Problems
The scale of the problem is difficult to appreciate until you look at a sufficiently large sample.
WebAIM's annual Million project provides one of the largest recurring measurements available. In 2026, researchers again evaluated the homepages of one million popular websites using the WAVE accessibility engine.
The results were substantial:
- 56,114,377 detected accessibility errors
- 56.1 errors per homepage
- 95.9% of homepages with detected WCAG failures
- A 10.1% increase in detected errors compared with 2025
Those numbers require an important qualification.
Automated testing cannot identify every accessibility problem. WebAIM itself explicitly warns that the absence of detected errors does not mean that a page is accessible or WCAG conformant.
The Million project therefore isn't a million complete accessibility audits.
It measures problems that software can identify with a high degree of confidence.
In some ways, that makes the results more concerning.
We are still failing the basics
Six categories accounted for 96% of all errors detected in the 2026 analysis:
| Accessibility issue | Homepages affected |
|---|---|
| Low-contrast text | 83.9% |
| Missing alternative text | 53.1% |
| Missing form labels | 51.0% |
| Empty links | 46.3% |
| Empty buttons | 30.6% |
| Missing document language | 13.5% |
Even more remarkably, these have remained the six most common categories of detected errors for seven consecutive years.
These aren't obscure WCAG edge cases.
Contrast. Alternative text. Labels. Links. Buttons. Page language.
We know how to fix these things.
That leads to a more interesting question than whether inaccessible websites still exist:
Why do we keep building them?
2. Progress Has Been Painfully Slow—and 2026 Reversed Some of It
Looking only at one year can make accessibility statistics seem static. Looking backward tells a more interesting story.
WebAIM reported detectable WCAG failures on:
- 97.4% of tested homepages in 2021
- 96.3% in 2023
- 94.8% in 2025
- 95.9% in 2026
There had been progress.
It just wasn't particularly fast.
Then 2026 reversed some of that improvement.
Several individual measures moved in different directions.
The percentage of pages with missing alternative text decreased from 55.5% to 53.1%, and pages missing a document language decreased from 15.8% to 13.5%.
Meanwhile, low-contrast text increased from 79.1% to 83.9%. Missing labels increased from 48.2% to 51%. Empty links and empty buttons also became more prevalent.
This doesn't mean every aspect of web accessibility suddenly deteriorated, nor does automated homepage testing represent the accessibility of the entire web.
But it does undermine a comforting assumption:
Better technology and greater awareness do not automatically produce better accessibility.
Another finding may help explain why.
3. The Web Is Getting Much More Complicated
The average homepage tested by WebAIM in 2026 contained 1,437 elements.
That's a 22.5% increase in a single year.
Since WebAIM began this research in 2019, the number of homepage elements has nearly doubled.
Think about what that means in practice.
- More components
- More interactions
- More third-party scripts
- More generated markup
- More widgets
- More dependencies
- More design-system components
- More opportunities for something to go wrong
WebAIM found a detected accessibility error on approximately one out of every 26 homepage elements in 2026.
ARIA provides another interesting example.
ARIA exists to communicate semantics and behavior to assistive technologies when native HTML alone isn't sufficient.
In 2026, WebAIM detected more than 133 million ARIA attributes—more than 133 per homepage on average.
ARIA usage increased 27% in one year and is now more than six times higher than it was in 2019.
Yet pages containing ARIA also had more detected accessibility errors on average than pages without it.
That does not mean ARIA causes inaccessible websites.
WebAIM explicitly cautions against that conclusion. More complex websites are both more likely to use ARIA and more likely to contain accessibility problems.
But the relationship illustrates something important:
Using accessibility technology is not the same thing as creating an accessible experience.
A sophisticated component containing numerous ARIA attributes can still be less accessible than a correctly implemented native HTML control.
Technical sophistication doesn't guarantee accessibility sophistication.
4. Higher Education Provides a Useful Warning
Several studies published in 2026 examined university websites, and the results reinforce the larger pattern.
A March 2026 study evaluated 993 pages from 250 Chinese universities using three automated accessibility tools against WCAG 2.2 Level A and AA.
The axe evaluation alone identified 36,283 issues, averaging 36.54 per page. Nearly 70% of identified issues related to the WCAG Perceivable principle, and most involved foundational Level A criteria.
Research in the United States found similar problems on a smaller scale.
A 2026 exploratory audit examined homepages and Disability Services pages at 20 U.S. universities. Across 160 page-and-testing-module scans, researchers identified 1,626 accessibility alerts.
Color contrast accounted for more than half.
Perhaps most strikingly, accessibility barriers were found even on Disability Services pages—the pages specifically intended to help disabled students access institutional resources.
The problem becomes even clearer when researchers look beyond homepages.
A September 2026 study evaluated 200 library pages from 100 U.S. public universities, comparing library homepages with the search-result interfaces where students actually conduct research.
Search-result pages contained approximately 2.5 times as many accessibility errors as the homepages.
University prestige didn't predict accessibility performance.
That's an important finding.
An accessible front door isn't enough if the room behind it is inaccessible.
5. Accessibility Is an Organizational Problem
For years, discussions about web accessibility have disproportionately focused on developers.
Use semantic HTML. Add alternative text. Label your inputs. Check color contrast. Make everything keyboard accessible.
Those practices remain essential.
But recent research increasingly examines what happens around the developer.
- Who owns accessibility?
- Who receives training?
- Who checks procurement?
- Who establishes organizational standards?
- Who tests releases?
- Who monitors existing content?
- Who has the authority to stop an inaccessible product from shipping?
A 2026 study of local-government online platforms specifically examined organizational factors affecting accessibility. The researchers found relationships involving website age, oversight, government structure, and organizational learning.
Another study published in August 2026 examined digital-accessibility readiness among 20 leading U.S. public universities.
All 20 publicly referenced WCAG 2.1 AA, while three committed to WCAG 2.2 AA. Researchers found substantial activity involving training, accessibility offices, governance structures, evaluation tools, remediation, monitoring, and institutional policy.
That's encouraging.
It also reveals how much infrastructure sustainable accessibility requires.
An organization can employ developers who understand WCAG and still repeatedly produce inaccessible products if accessibility is absent from design requirements, procurement, content workflows, quality assurance, management, vendor agreements, and governance.
That changes the way we should think about remediation.
If an organization fixes hundreds of accessibility defects every year but continues creating hundreds more, the problem isn't its ability to repair websites.
The problem is the system producing them.
6. Accessibility Laws Are Getting Stronger
2025 and 2026 have also been significant years for accessibility regulation.
In the European Union, the European Accessibility Act began applying on June 28, 2025, bringing common accessibility requirements to covered products and services including e-commerce, banking services, electronic communications, computers, smartphones, e-books, and other technologies.
In the United States, the Department of Justice's updated ADA Title II rule established WCAG 2.1 Level AA as the technical standard for covered state and local government web content and mobile applications.
The implementation timeline has changed since the rule was originally issued.
In April 2026, the Department of Justice published an Interim Final Rule extending the compliance dates. Larger state and local government entities now have until April 26, 2027, while smaller entities and special district governments generally have until April 26, 2028.
The standard didn't disappear.
The implementation window got longer.
That gives organizations additional time—but also removes one excuse for waiting until the last minute.
Does legislation actually produce accessible websites?
Recent research has started examining that question directly.
A 2026 study compared accessibility and legal frameworks across e-commerce websites in the United States, United Kingdom, Canada, Australia, and South Africa.
The researchers found substantial implementation gaps and differences in regulatory clarity and enforcement.
But the relationship isn't simple.
The technical accessibility measures used in the study did not show statistically significant differences between countries across several tested categories.
That nuance matters.
Stronger laws may influence organizational behavior, accountability, investment, and policy without immediately producing perfectly accessible websites.
A regulation can establish a requirement.
It cannot write useful alternative text.
It cannot manually test a keyboard interaction.
It cannot make an organization train its employees effectively.
It cannot automatically make accessibility part of development culture.
Law can change incentives.
Organizations still have to change systems.
7. Automated Accessibility Testing Is Essential—and Incomplete
There's an important caveat running through nearly every large accessibility study:
Computers cannot fully test accessibility.
Automated tools are excellent at identifying certain classes of problems.
- They can detect an image without an
altattribute. - They can calculate many contrast failures.
- They can identify unlabeled form controls.
- They can find empty buttons.
- They can flag many forms of invalid ARIA.
What they cannot reliably determine is whether the experience actually makes sense.
An image can have alternative text and still have terrible alternative text.
A heading structure can technically exist while communicating an incoherent hierarchy.
A button can have an accessible name that poorly describes its purpose.
A keyboard user might technically be able to reach every control while navigating through an exhausting interface.
A screen reader can announce every component while the overall workflow remains confusing.
This distinction matters when interpreting accessibility research.
When WebAIM reports that 95.9% of homepages contain detectable WCAG failures, it does not mean that the remaining 4.1% are accessible.
It means that automated testing did not detect the WCAG failures it is capable of identifying.
Accessibility conformance and accessible user experience overlap.
They aren't identical.
8. We Also Need to Ask Who Accessibility Research Represents
One of the more interesting findings from recent literature isn't about websites at all.
It's about accessibility research itself.
A 2026 review examined 18 peer-reviewed studies involving website and mobile accessibility for people with visual impairments.
The researchers found that the literature did not evenly cover WCAG requirements or research methodologies.
None of the analyzed studies addressed Level AAA criteria.
Only seven evaluated assistive-technology usage among people with visual impairments, and only three of the 18 studies used mixed methodologies.
That exposes an important limitation.
"People with disabilities" isn't one user group.
A blind screen-reader user and a person with low vision can encounter different barriers.
So can someone with limited dexterity. Someone who cannot use a mouse. Someone with dyslexia. Someone with a cognitive disability. Someone who is Deaf or hard of hearing. Someone using speech input. Someone with multiple disabilities.
A website can perform well against the accessibility problems researchers most commonly measure while still creating substantial barriers for people whose needs receive less attention.
That's another reason automated WCAG statistics should be treated as one piece of accessibility evidence rather than a complete measurement of accessibility.
9. And Then AI Arrived
Artificial intelligence creates one of the more interesting contradictions in current accessibility research.
AI has significant potential to improve accessibility.
- It can assist with alternative-text generation.
- It can help identify accessibility problems in code.
- It can transform and simplify content.
- It can assist captioning and transcription.
- It can help developers understand WCAG failures.
- It can suggest repairs.
A systematic literature review published in 2025 began with 277 studies and identified 31 key works examining AI methods related to web accessibility. The researchers found promise in accessible-content generation and particular coding applications while also identifying significant gaps.
A separate systematic mapping study published in 2026 analyzed 53 studies involving AI and web accessibility. Among the areas represented were alternative-text generation, accessible HTML generation, accessibility testing, automatic correction, conversational navigation, and accessibility-barrier remediation.
So AI can help.
But AI also dramatically increases our ability to generate digital content and software.
A developer who once manually produced several hundred lines of interface code can now generate considerably more in minutes.
If the system generating that code doesn't reliably account for accessibility—or nobody tests the result—we haven't eliminated the accessibility problem.
AI may also accelerate the production of inaccessible content when accessibility safeguards are absent.
That creates an important question for the next several years:
Can we make accessible output the default rather than something users have to explicitly request?
That's a much higher bar than demonstrating that an AI system can produce accessible code.
10. What Does the Research Actually Tell Us to Do?
The studies differ in methodology, geography, sector, disability populations, and measurement techniques.
But several themes repeatedly emerge.
Fix basic accessibility before chasing sophisticated solutions
If low contrast, missing alternative text, unlabeled forms, empty links, and empty buttons still dominate large-scale accessibility testing, organizations don't need to wait for an AI accessibility revolution to make meaningful progress.
Start with good HTML.
Reduce unnecessary complexity
Every additional component, dependency, interaction, and abstraction introduces another opportunity for an accessibility failure.
Complexity isn't inherently inaccessible.
But complexity should have to justify itself.
Test automatically—but never only automatically
Automated accessibility testing belongs in development pipelines.
So do manual keyboard testing, screen-reader testing, zoom and reflow testing, and testing involving disabled users.
Test the task, not just the homepage
The university-library research provides an excellent example.
A homepage can perform relatively well while the interface where the user actually completes their task performs much worse.
Test registration. Test checkout. Test search. Test account creation. Test forms.
Test the workflows people came to the website to use.
Make accessibility organizational
Accessibility can't belong exclusively to the developer who happens to know WCAG.
Designers, developers, content creators, procurement teams, managers, QA teams, vendors, and leadership all influence accessibility outcomes.
Build accessibility in instead of repairing it later
Accessible components, design systems, content standards, procurement requirements, automated tests, and release criteria can prevent entire categories of recurring problems.
Measure user experience, not just conformance
Passing an automated test doesn't prove that somebody can successfully use the product.
Eventually, somebody has to use it.
Final Takeaway: We Know More Than Enough to Do Better
Perhaps the most frustrating conclusion from the latest accessibility research is that we're not waiting for some undiscovered technical breakthrough.
We already know how to make a button accessible.
We know how to label a form.
We know how to provide alternative text.
We know how to create sufficient contrast.
We know how to structure a document.
We know how to support keyboard navigation.
We know how to test with assistive technology.
The difficult problems haven't disappeared. Accessibility will always involve situations requiring judgment, research, testing, and—critically—the expertise and experiences of disabled people.
But millions of the barriers researchers continue finding aren't difficult problems.
They're old problems.
The technology surrounding the web has transformed enormously.
Our frameworks have changed. Our devices have changed. Our browsers have changed. Our design practices have changed. Our development tools have changed.
Artificial intelligence can now generate an entire interface from a sentence.
Yet we're still forgetting to label the button.
The 2026 research doesn't tell us that accessibility has failed.
It tells us something more useful:
Accessibility doesn't improve merely because our technology improves. It improves when we deliberately make it part of how that technology is designed, built, purchased, tested, maintained, and measured.
Maybe the next major improvement in web accessibility won't come from another tool.
Maybe it will come when accessibility stops being something we fix and becomes part of what we mean when we say something is finished.
References and Research Reviewed
- WebAIM. The WebAIM Million: The 2026 Report on the Accessibility of the Top 1,000,000 Home Pages. 2026. Annual automated WAVE analysis of one million popular homepages. View the 2026 WebAIM Million report .
- WebAIM. The WebAIM Million: 2025. 2025. Used for year-over-year and longitudinal comparisons. View the 2025 WebAIM Million report .
- Liu, H., Sun, J., He, Z., Huang, H., Liu, X., & Nie, L. Web accessibility of university websites in China: An empirical study. Journal of King Saud University Computer and Information Sciences, 2026. Evaluation of 993 pages from 250 universities using three automated accessibility tools. https://doi.org/10.1007/s44443-026-00544-6
- Adedoja, T. Exploratory automated audit of visual and structural accessibility gaps in US university homepages and Disability Services pages with the Accessible Name and Description Inspector. Discover Education, 2026.
- Badiuzzaman, M., Jiang, Z., Lee, J-S., Cumming, T. M., et al. Accessible for all? A longitudinal evaluation of the digital accessibility of all Australian university websites. Humanities and Social Sciences Communications, 2026.
- Evaluating the accessibility of U.S. public university library webpages: A comparative analysis of homepages and search results pages. The Journal of Academic Librarianship, 2026. https://doi.org/10.1016/j.acalib.2026.103293
- Frandell, A. & Allgood, M. Breaking barriers: Evaluating accessibility in local government online platforms for individuals with disabilities. Government Information Quarterly, 43(1), 2026. https://doi.org/10.1016/j.giq.2026.102111
- Cudjoe, E. K. & Asilevi, P. A. Toward universal access to information: an analysis of digital accessibility policy readiness in U.S. public universities. Universal Access in the Information Society, 2026. https://doi.org/10.1007/s10209-026-01379-8
- The status quo of 'WCAG' reporting in research on technology use by people with visual impairments: key gaps and underexplored areas of study. 2026. Systematic review of 18 peer-reviewed studies. https://doi.org/10.1080/12460125.2026.2688526
- Examining the Relationship Between Legislation and Web Accessibility Compliance: From Theory to Practice. International Journal of Human–Computer Interaction, 2026. Mixed-method comparison involving 56 e-commerce websites across five countries. https://doi.org/10.1080/10447318.2026.2623225
- Vera-Amaro, G. & Rojano-Cáceres, J. R. Towards accessible website design through artificial intelligence: A systematic literature review. Information and Software Technology, 186, 107821, 2025. https://doi.org/10.1016/j.infsof.2025.107821
- Campoverde-Molina, M. & Luján-Mora, S. Artificial intelligence in web accessibility: A systematic mapping study. Computer Standards & Interfaces, 96, 104055, 2026. https://doi.org/10.1016/j.csi.2025.104055
- U.S. Department of Justice, Civil Rights Division. Accessibility of Web Content and Mobile Apps Provided by State and Local Government Entities. ADA Title II rule and 2026 Interim Final Rule extending applicable compliance dates. View the DOJ guidance .
- European Commission. European Accessibility Act. Directive accessibility requirements applicable to covered products and services beginning June 28, 2025. View the European Commission overview .