AI and Healthcare Accessibility Audits: Why Human Expertise Still Matters

Accessibility is a critical consideration for health plans. It affects whether members can independently access care, understand their benefits, and navigate enrollment—and it directly affects CMS compliance, Star Ratings, and CAHPS performance. When digital experiences fall short, the people who bear the costs are older adults, people with disabilities, and those with limited English proficiency. 

The business impact is significant. Studies show that 71% of users with disabilities will leave a website that is not accessible, which can affect enrollment, member satisfaction, digital engagement, and long-term retention. Accessible digital experiences also help members compare plans, access provider information, complete enrollment, and use self-service tools more effectively.

As organizations look for ways to improve accessibility, many are turning to AI-powered tools for accessibility audits, consultation, and remediation. While AI can be a valuable assistant, it’s not the shortcut it may appear to be. Accessibility remains a human-centered discipline, and AI cannot reliably own accessibility work or create accessible experiences on its own.  

Why AI Falls Short in Accessibility Audits 

AI can be a useful tool for accessibility work, but it has limitations that make it difficult to rely on without human oversight. 

The training data problem: AI models have been trained on vast amounts of internet content, much of which is not accessible. The recent 2026 WebAIM Million survey, which evaluates the accessibility of the homepages of the top million websites, found only marginal gains. The overwhelming majority of the internet remains inaccessible, meaning many large language models (LLMs) have been trained on content that does not meet accessibility standards.

The instruction problem: Good AI output requires expertise in the subject to guide it where you want. If you’re turning to AI because you lack accessibility knowledge, you’re not equipped to give it the direction it needs and won’t know when you’ve arrived. Simply prompting AI to “build a great product” and shipping the result is how businesses fail fast. We shouldn’t make the same mistake in accessibility.

The "looks good enough" problem: Flawed output often appears fine to an untrained eye, but not to an industry expert or someone who actually relies on assistive technology. We tend to trust AI most where we’re least equipped to catch its mistakes.

The stakes problem: Digital accessibility compliance is frequently a legal requirement, and even the best-in-class pre-AI tools cap out at about 50% issue detection. AI hasn't meaningfully moved that needle, and experimental tools plus legal compliance don't mix. Effective accessibility testing still requires expert review, assistive technology validation, and human judgment.

Can AI Ever Fully Replace Human Accessibility Expertise?

AI will continue to improve, but many accessibility challenges are not simply technical problems to solve. Accessibility is fundamentally about user experience, which requires understanding how real people interact with digital content. Questions about usability, clarity, and navigation often depend on context and judgment rather than a single correct answer. 

Accessibility testing also involves real-world interactions across different browsers, operating systems, and assistive technologies, each with its own behaviors and limitations. While AI can analyze portions of a digital experience, it can’t fully replicate the experience of a person using a screen reader, keyboard controls, or other tools. 

As AI capabilities advance, it will likely become a more effective assistant for accessibility professionals. However, greater capability does not guarantee reliability. For organizations responsible for compliance and user access, human expertise will remain an essential part of the process.

Where AI Adds Value

One of the most significant advances has been in automated captioning. With high-quality audio, AI-generated captions have improved dramatically over the past decade and can now achieve near-human levels of accuracy in many situations. While human review may still be required for legal or compliance purposes, AI can substantially reduce the time and effort involved in creating captions. 

AI is also improving the tools available to designers and developers. Platforms such as Stark integrate directly with design environments like Figma, providing features such as color contrast analysis, vision simulators, and accessibility annotations that help teams identify potential issues earlier in the development process. 

Perhaps most importantly, AI is expanding what's possible for people with disabilities themselves. Tools such as Be My Eyes use AI-powered vision capabilities to help blind and low-vision users interpret and navigate the world around them, offering greater independence in situations where a user may not want to rely on a stranger, or where the mental load of that dependence simply isn’t practical.

The most successful applications of AI tend to support accessibility efforts rather than replace them. AI can help people work more efficiently, identify potential issues, and increase access to information. What it cannot reliably do is replace the expertise, judgment, and real-world testing required for comprehensive accessibility remediation and evaluation.

Keep AI in the Passenger Seat

AI is a capable assistant, but accessibility requires an expert practitioner. Human expertise, real assistive technology testing, and genuine user understanding simply can’t be automated away. This is especially true when legal compliance and member access are in play.

For Medicare Advantage plans, the cost of inaccessible experiences shows up in CAHPS scores, retention, and enrollment outcomes. TransPerfect helps health plans close the gap—efficiently, compliantly, and with the human oversight that this work demands. 

Ensure your health plan is accessible for all members. Connect with our accessibility experts  to evaluate, remediate, and move forward with confidence.