AI and Accessibility in the Digital Experience Life Cycle

I’ve previously argued that there simply aren’t enough people in the world to make the entire internet accessible without automation. I stand by this point. Designers, developers, and content authors cannot practically manage all aspects of accessibility, given the scope of digital content creation today. This legacy approach continues to lead to burnout and hinder broader access for people with disabilities.

Artificial intelligence (AI) is reducing reliance on authors to manually ensure accessibility, enabling teams to work more efficiently by automating aspects of accessible experience creation and remediation. In the second installment of my three-part series on AI and accessibility, I’ll explore the opportunities that AI presents for streamlining and enhancing accessibility at key stages of the digital experience life cycle. I’ll also discuss the long-term potential implications of these process changes for digital accessibility standards.

(If you missed part one of this series, on the impact of AI on people with disabilities, you can find it here.)

AI in accessible design, development, and content authoring

AI tools are already making it easier for teams to create new, accessible experiences. Many of these tools are still nascent, so designers, developers, and content authors must be mindful of their limitations when incorporating AI into their workflows. However, as AI technologies become more sophisticated, they're poised to have a profound impact on the accessibility of the digital world.

AI’s emerging role in user interface (UI) personalization has been met with enthusiasm among some influential voices in user experience (UX) design, including Jakob Nielsen, who covered the topic in his article, “Accessibility Has Failed: Try Generative UI.” Generative UI and UI personalization tools may enhance accessibility by allowing users to adjust their preferences, including collapsing distracting content, simplifying the interface, and adjusting fonts and colors for easier reading. However, generative UI and UI personalization tools alone cannot ensure accessibility—and these technologies must be considered an enhancement to, not a replacement for, an interface that conforms with a baseline standard such as the Web Content Accessibility Guidelines (WCAG).

I firmly believe that, to serve users equitably, designers must offer a standard level of accessibility to everyone, rather than expecting people with disabilities to opt into a "separate but equal” custom interface for basic access. Additionally, ensuring conformance with accessibility standards is essential for providing users with a consistent experience across websites, and a predictable experience when they visit the same site, so they don’t have to learn a new layout each time they return. 

Finally, website owners should not make assumptions about individual users’ needs based solely on their disability. For example, just because a person is blind doesn’t mean an audio interface is optimal for them. They, for example, may rely on braille and need to input data through tactile means (e.g., a keyboard interface) rather than by voice. By building experiences that are usable for as many people as possible, and then allowing users to personalize these experiences, site owners can ensure their efforts to enhance accessibility through personalization don’t unintentionally create barriers.

That said, generative UI and UI personalization tools hold promise in solving for the fact that accessibility standards aren't “one size fits all.” Users may have conflicting needs and preferences, and considerations that improve access for some users, but create barriers for others, are deliberately excluded from WCAG. By offering users tailored experiences, generative UI and UI personalization tools can help website owners meet individuals' unique access needs beyond what is achievable through conformance with accessibility standards.

When it comes to development, tools like GitHub's Copilot are already assisting developers in writing more accessible code. Because these tools are limited by the data they have been trained on, however, they often require guidance to produce accessible outcomes. As AI models are increasingly trained on accessibility-focused data, their output improves. It's a misconception that AI can currently generate fully accessible code autonomously; ongoing input and corrections from developers are necessary. But with thoughtful prompt engineering, developers can achieve quality results.

For content authors, one of the most promising applications of AI to support accessibility is generating alternative (alt) text. While critics have pointed to errors made by early AI systems as proof that human writers provide higher-quality descriptions, it's worth considering how frequently human-authored alt text is either poorly written or completely missing. In many of these instances, AI could potentially match or even exceed the quality of descriptions written by humans, even accounting for inaccuracies. AI can also be a useful tool in crafting a first, more descriptive version of alt text that a human author can then edit. Moreover, AI has the capability to consider the surrounding content of a page when generating alt text, a practice that has been improving over the years.

AI in issue remediation

Endpoint automation, also referred to as automated remediation, corrects accessibility issues after code is developed. This technology—which may include AI—is sometimes met with criticism, as advocates fear it might absolve developers from their responsibility to address accessibility at the code level.

But the practical reality remains that without automated solutions, many existing services would remain inaccessible to those who need them most. It’s also not a wholly new concept: over 15 years ago, I developed scripts for the JAWS screen reader to make specific systems usable for blind employees by modifying the Document Object Model of Internet Explorer to incorporate HTML ARIA attributes to correct accessibility issues. The progress we’ve made since this early form of automated remediation underscores the potential of future AI innovation to further accessibility.

While automation should not be seen as a panacea, it is critical in contexts where manual revision is unfeasible because of an ever-increasing backlog of issues or on a site where content remains static. Technology today can safely incorporate many accessibility fixes without breaking accessibility. A few examples include:

• Automatically adding 'skip links' to allow users to bypass blocks of repeated content

• Automatically detecting and setting the language of the page to assist users of text-to-speech and screen readers

• Adjusting the meta viewport to allow for browser zoom to 200%

• Preventing auto-playing audio and video content, which may distract users

These automated fixes can significantly enhance usability for individuals, including those who rely on assistive technologies.

AI content creation and web accessibility standards

As we move into the future, digital accessibility standards will need to evolve to accommodate the increasing capabilities of AI. These future standards will have to focus on the outcomes—what needs to be achieved—rather than prescribing specific methods. For instance, instead of mandating that all images must have manually provided alternative text, future standards could require that the purpose of an image must be discernible and conveyable in various accessible formats—text, audio, or braille—by commonly available technology. This approach would ensure that the intent and information conveyed by an image are accessible to users who cannot visually perceive it, regardless of whether the alternative text is explicitly provided by the content creator or determined by a machine.

Of course, many AI technologies for accessibility are not yet universally available or consistent. Until these technologies become uniformly reliable and widely accessible, standards are unlikely to change in this way.

The future of AI in accessible experience creation

In the near term, I believe that the primary use of AI will not be fully autonomous. Rather, AI will act in a supportive role, providing suggestions to be acted upon by a user, such as an author. Furthermore, there is an emerging trend toward ensuring that end-users are aware of AI involvement in the content they interact with. This trend toward greater transparency and the careful curation of AI-generated content is likely to intensify, especially as legal challenges arise from instances where unmonitored generative AI has delivered inaccurate information.

In summary, while AI tools and automation are transforming accessibility, a balanced approach that includes both proactive and reactive measures is essential. Designers, developers, and content authors must continue to strive for accessibility from the outset, complemented by AI-driven aids and endpoint solutions that improve efficiency to ensure that all content is universally accessible.