AI Is Everywhere in Digital Accessibility, That Doesn’t Mean It’s Working

AI has quickly found its way into digital accessibility. Eighty percent of professionals now use it for accessibility work, according to new research from Level Access.

What’s less settled is what they should actually be using it for.

Across the software development life cycle, AI adoption is remarkably scattered. Thirty-two percent use it to generate accessibility compliance documentation, 29% to check designs for accessibility issues, 29% to fix those issues, 28% to test code and 28% to identify accessibility requirements during planning. No individual use case has emerged as the clear standard.

That experimentation is happening as AI reshapes the speed and sequence of software development itself.

Among respondents using AI, 99% said it has accelerated at least one stage of the development cycle. Sixty-one percent reported faster design, 58% faster planning and 51% faster development. Just 30% said the same for quality assurance and testing. AI has also shifted design earlier in the process for 64% of respondents, compared with just 23% who said QA and testing moved earlier.

The result is an increasingly compressed development cycle in which organizations can produce digital experiences faster than they can necessarily validate them.

And despite widespread adoption, the research suggests simply adding more AI isn’t what separates the strongest accessibility programs from everyone else.

AI Adoption Alone Isn’t Moving the Needle

Level Access surveyed 2,530 professionals involved in creating or maintaining digital experiences for its eighth annual State of Digital Accessibility Report, conducted by independent research firm Sapio Research.

The researchers compared organizations that reported improvement across every accessibility metric they formally track with the broader survey population.

AI adoption was higher among those organizations, but only modestly. Ninety percent of the organizations improving across all of their tracked metrics use AI for accessibility, compared with 80% overall.

Other differences were far larger.

Among those improving across the board, 88% described their accessibility training as highly effective, compared with 49% overall. Eighty-one percent said accessibility was embedded in their organizational culture, versus 43% overall. Eighty-one percent reported clear and frequent communication about accessibility between teams, compared with 40% across the full sample.

They were also nearly twice as likely to start addressing accessibility during planning, 67% versus 35%.

In practice, AI appears to work differently depending on the infrastructure around it.

Among organizations with all seven markers Level Access uses to assess program readiness, 29% said AI helped them prioritize accessibility issues more effectively. That fell to 5% among organizations with none of those markers. Twenty-nine percent of the most mature organizations said AI helped them address accessibility more proactively during development, compared with 10% among organizations without the underlying practices in place.

Those readiness markers include governance, communication, training, integration into development workflows, executive support, procurement requirements and a culture of shared responsibility for accessibility.

“AI doesn’t create a strong accessibility program any more than buying a new Mac makes you a professional photographer,” said Corbb O’Connor, director of thought leadership at Level Access. “The organizations seeing the biggest gains aren’t replacing fundamentals with AI. They’re using AI to make good practices scale farther and move faster.”

Automation Isn’t Eliminating the Human Work

The rise of AI also hasn’t translated into the kind of cost reduction that has accompanied some other enterprise AI deployments.

Only 6% of respondents said they spend less on accessibility as a result of AI adoption. Twenty-two percent said they rely less on manual evaluation, suggesting that some automated work is being replaced without eliminating the need for accessibility investment overall.

Human oversight remains particularly important to the organizations with the most developed accessibility programs.

Eighty-nine percent of professionals overall said it is critical for people to remain closely involved in accessibility processes as AI advances. Among respondents at organizations with all seven program-readiness markers, that figure climbs to 97%.

“The true measure of accessibility is the experience of people with disabilities, not the output of a tool,” said Jon Avila, chief accessibility officer at Level Access. “Organizations that succeed will use AI to scale their efforts while ensuring that human expertise and the perspectives of people with disabilities continue to guide decisions, validate outcomes, and shape inclusive experiences.”

That becomes particularly important when AI is generating code and content from an internet where inaccessible experiences remain widespread. WebAIM’s 2026 analysis of the top one million home pages found detectable WCAG failures on 95.9% of them, up from 94.8% the previous year.

AI can accelerate creation, but it does not inherently make what it creates accessible.

Earlier Doesn’t Necessarily Mean Better

There is some evidence that accessibility is entering the development process sooner.

Thirty-five percent of respondents said their organizations begin addressing accessibility during planning, up from 28% in 2025. Fewer organizations now wait until QA, after launch or later stages to begin the work.

But the report cautions against reading that shift entirely as evidence of more proactive accessibility programs.

AI is compressing the early stages of development and changing when design, writing and development happen. Accessibility may therefore be moving earlier because the development process itself is moving earlier.

That distinction becomes important when QA isn’t keeping pace.

If accessibility remains primarily a testing exercise at the end of an increasingly fast development cycle, organizations have less time to identify and remediate problems before release. The alternative is embedding accessibility requirements into the same AI-enabled workflows now speeding up planning, design and development.

Accessibility Is Also Becoming a Buying Requirement

The consequences are increasingly commercial, not just technical.

Eighty-three percent of respondents said proof of digital accessibility is required most or all of the time when purchasing digital products, up from 75% in 2025. Forty-two percent said it is always required, compared with 31% last year.

One respondent described asking AI vendors to demonstrate how their output performs against WCAG 2.2 criteria and whether their systems generate properly tagged and structured content by default.

That raises the stakes for enterprise technology companies selling AI-enabled products. Accessibility isn’t only something their customers may evaluate after implementation. It is increasingly becoming part of whether a product gets purchased at all.

Still, many organizations have already put the formal pieces of an accessibility program in place. Ninety-one percent have an accessibility policy, 97% have someone accountable for accessibility and 84% have a dedicated budget.

The larger problem is translating that infrastructure into everyday practice. Only 60% say their accessibility policy has a major operational impact. And while 89% provide accessibility training, fewer than half describe that training as highly effective.

AI is putting those programs under a new kind of pressure. It gives organizations the ability to create more digital experiences, more quickly, without automatically improving the systems responsible for making those experiences accessible.

For companies already doing the underlying work, AI can extend it across more of the development process. For everyone else, faster development may simply mean accessibility problems can scale faster too.

The State of Digital Accessibility Report was based on a June and July 2026 survey of 2,530 professionals. Fifty-two percent of respondents were based in the U.S., 43% in the U.K. and European Union, and 97% worked for organizations with at least 1,000 employees.