AI and Accessibility: The Promise, the Pitfalls, and Why Overlays Aren't the Answer
AI is reshaping digital accessibility — but automated overlays don't make sites compliant. An honest look at what AI can and can't do for accessibility.
Artificial intelligence is reshaping digital accessibility — in some ways for the better, in others creating new problems we're still learning to navigate. Automatic captions, AI-generated alt text, real-time sign language translation, and voice interfaces have given more people access to digital content than ever before. But the same AI wave has also produced a class of products that promise to 'fix' accessibility automatically — and consistently fail to deliver.
Where AI Is Genuinely Helping
Caption generation has been the clearest AI accessibility win. Services like Google's Live Caption, YouTube's automatic subtitles, and Microsoft Azure AI Captioning have made video content accessible to deaf and hard-of-hearing users at scale — something prohibitively expensive through human transcription alone.
AI-powered alt text generation — used by tools like Microsoft Seeing AI and Google Lookout — helps content teams at scale. For images that would otherwise be uploaded without descriptions, AI alt text is meaningfully better than nothing. It's a floor, not a ceiling: human review remains essential for complex, contextual, or brand-specific images.
The Overlay Problem
Accessibility overlays — JavaScript widgets that claim to make any website WCAG-compliant at the click of a button — have been one of the most aggressively marketed technology categories of the last five years. The accessibility community's verdict is unambiguous: overlays do not make websites accessible.
Screen reader users — the very users overlays claim to help — have documented that overlays frequently conflict with their assistive technology, making sites harder to use, not easier. Several overlay vendors have faced lawsuits from plaintiffs who argued the overlay made the site less accessible than before.
AI can augment accessibility. It cannot automate it. The difference matters enormously.
Evaluating AI Accessibility Tools Honestly
Automated testing tools — axe, WAVE, Lighthouse — can catch roughly 30–40% of WCAG failures. AI-powered tools are improving, but the gap is significant. For organisations buying AI accessibility tools, ask: What percentage of WCAG success criteria does this tool actually test? Can it provide a WCAG-referenced audit report? Has it been tested with real assistive technology users? What happens when I remove the tool — is the underlying site accessible?
The Responsible Path Forward
Use AI for what it does well: scale, speed, and rough-cut detection. Use human expertise for what requires judgment: remediation validation, user testing with assistive technology, and producing legally defensible documentation. The combination is powerful. Either alone is insufficient.
Back to all posts