Built Without Us: Why People with Disabilities Are Still Left Out of AI Tool Design
AI-powered hearing aids, exoskeletons, and health monitors are shipping faster than ever. The people they’re built for are still too often absent from the room where they’re designed.
AI is showing up across nearly every category of assistive and adaptive technology this year: hearing aids that process sound through dedicated AI chips, exoskeletons with adaptive gait-training algorithms, health wearables that track vitals in real time for people managing chronic conditions. A recent industry roundup aimed at people with disabilities, caregivers, and families catalogued a genuinely wide range of active products — Phonak’s Sphere Infinio platform, Oticon’s Intent hearing aids built on a deep neural network trained on millions of sound samples, Ekso Bionics’ EksoNR rehabilitation exoskeleton, ReWalk’s FDA-cleared powered exoskeleton for spinal cord injuries, and Cyberdyne’s HAL system used in rehabilitation settings abroad.
That same roundup includes a caution worth sitting with: what serves one person may not serve another, and people with disabilities have often been excluded from the design and development of the very technologies built in their name. It’s a familiar pattern to anyone who has spent time in disability advocacy circles — and it’s the same principle behind IAAP’s CPACC credential, which grounds accessibility professionals in the lived-experience foundations of the field before it gets to technical checklists at all.
A Pattern Older Than AI
This isn’t a new problem AI introduced — it’s an old problem AI is now scaling. Assistive technology has a long history of being engineered by clinicians and technologists working on behalf of people with disabilities rather than alongside them, producing devices that solve the problem an outside expert imagined rather than the one the actual user experiences day to day. The disability rights movement’s long-standing principle, “nothing about us without us,” exists precisely because that gap has real consequences: devices that are technically impressive and practically unused, or that solve a clinical metric while ignoring the priorities — comfort, independence, dignity, cost — that matter most to the person wearing or using them.
Why AI Raises the Stakes
When a hearing aid’s sound processing or an exoskeleton’s gait algorithm is trained on data, the question of whose experience shaped that training data becomes just as important as whose experience shaped the industrial design. A model trained primarily on one population’s movement patterns, speech patterns, or usage habits will perform best for people who resemble that training population — and quietly worse for everyone else, in ways that are much harder to spot than an uncomfortable strap or a confusing button layout. That’s a much more diffuse, and much less visible, version of the same exclusion.
How Accessible Web Helps: This is exactly why we built our Assistive Technology Testing service around real people, not just automated checks. Our team of individuals with disabilities tests your website or web app with the actual tools they use every day — so you find out how your product performs for real users before they do, not after.
What This Means Beyond Hardware
The same lesson applies directly to organizations building AI-powered features into websites and digital products — AI writing assistants, automated image descriptions, chatbots, summarization tools. It’s tempting to treat accessibility testing as a QA step that happens after an AI feature ships. The more durable approach is to involve users with disabilities in testing and feedback while the feature is still being shaped, the same way user research works for any other product decision — not as a compliance checkbox, but because the people who will actually rely on the tool are the ones most qualified to say whether it works.
The Takeaway
The pace of AI-powered assistive innovation is genuinely good news. But “built for people who have a disability” and “built with people who have a disability” are different claims, and only one of them reliably produces tools people actually want to use. Any organization evaluating or building AI features — hardware or web-based — should ask early who was in the room, not just what the product can technically do.
Ready to Make Accessibility Happen?
Stop guessing how real users experience your site. Get real feedback, then a clear path to fix what you find.
Start your free 14-day trial of Accessible Web RAMP — no credit card required. Try RAMP free for 14 days →
Sources
AI and Disability in 2026: A Comprehensive Guide, AmeriDisability — https://www.ameridisability.com/ai-and-disability-in-2026-a-comprehensive-guide-for-people-with-disabilities-caregivers-seniors-and-families
