Published: 2 June 2026
AI Is Changing Accessibility — But Not in the Way People Think
AI is reshaping digital product development at remarkable speed, and accessibility is no exception. We now have tools that can generate alt text, detect WCAG failures, summarise research sessions, and even propose design patterns. But there’s a growing misconception in the industry — the idea that AI can “do accessibility for you.” It can’t. Not even close.
AI accelerates accessibility work, but it doesn’t replace human judgment. Teams that understand the boundary between automation and expertise get the best results — and avoid the biggest risks.
My Experience With AI Tools So Far
Note these are some of my increasing familiarity with AI tools — not an exhaustive list. There are many I haven’t described here and no doubt many others that I haven’t tried yet.
ChatGPT
Like many people, ChatGPT was my first real exploration into AI. I used it mostly for content suggestions and personal tasks — even sourcing replacement equipment after leaving ABC. Helpful, yes. But not central to my accessibility work, so I’ll leave it there.
Microsoft Copilot
Copilot has been far more integrated into my professional workflow.
- Content shaping: For a recent GAAD speaking engagement, I drafted my talk and used Copilot to refine it for delivery. It didn’t replace my thinking — but it dramatically reduced prep time.
- Design ideation: Our AccessUX logo concept? AI suggested. It provided a solid creative direction that our designer could then execute properly.
- Accuracy is not guaranteed: When I needed to make structural changes to our website, Copilot gave me confident, detailed instructions. They looked right. They sounded right. They were not right.
After several rounds of “fix this based on the error I just got,” I found myself in a rabbit hole of broken code and false confidence.
The solution was simple: ask a human expert.
Cursor
From a technical perspective, Cursor has been the most promising tool I’ve used. It can generate fully coded systems from well considered prompts, and has been a huge time saver for me in prototyping and testing accessibility patterns. Yet even Cursor makes authoritative sounding assumptions about accessible markup such as which elements can have aria-label, how to structure an accordion, and so on. And those assumptions aren’t always correct. But it’s a powerful accelerator when used with the right oversight.
It has also been invaluable as I prototype a tool to streamline end-to-end product accessibility workflows - watch this space!
What AI Does Well Today
AI is genuinely useful — especially for speed. With the right oversight, it can meaningfully accelerate accessibility workflows.
Rewrite and refine content
If you draft the core message, AI can shape it for different audiences, formats, and delivery styles.
Generate first pass alt text
AI can describe what’s visually present. But it still struggles with meaning. Treat it as a starting point, not a final output.
Detect common WCAG failures
Automated tools reliably catch many recurring issues:
- Missing labels
- Low contrast
- Empty links
- Incorrect heading order
- Missing form associations
This removes a huge amount of manual checking.
Analyse user testing data
AI can summarise patterns, cluster behaviours, and highlight recurring issues. It’s a powerful accelerator for research teams.
Support code linting and component governance
AI can detect anti-patterns in code and suggest improvements — especially useful in large codebases.
What AI Cannot Reliably Do
1. Judge meaningfulness of alt text
AI can describe what is in an image, but not why it matters. That nuance is often the whole point.
2. Evaluate cognitive load or UX clarity
Accessibility isn’t just technical — it’s experiential. AI can’t feel friction.
3. Validate keyboard interactions
AI cannot reliably detect:
- Keyboard traps
- Incorrect focus order
- Broken modals
- Dynamic state changes
4. Interpret ARIA patterns
ARIA is complex. AI often misapplies roles or attributes.
The Risks of Over-Reliance on AI
Teams that treat AI as a replacement for accessibility expertise face real consequences.
1. False confidence
AI may say a page is “accessible” when it’s not.
2. Legal exposure
Actual usage breakdown and the implications of incorrect or misleading content on people with disabilities aside, they can present real legal risk for even well-intentioned organisations. An example is incorrect Accessibility Conformance Reports (ACR) with misleading accessibility claims that result in procurement — especially for ICT buyers in government and enterprise.
3. Missed high-impact failures
Automation can only catch around 30% of WCAG issues. The rest require human testing.
A Better Model: AI + Human Expertise
The winning pattern is simple: AI for speed. Humans for accuracy.
1. AI handles the first pass
- Automated scans
- Draft alt text
- Code linting
- Pattern detection
2. Humans verify and refine
- Assistive tech testing
- UX evaluation
- Semantic review
3. Integrate AI into your pipeline
- Pre-commit hooks
- CI/CD gates
- Design system governance
- Research analysis
This is where AI shines — as an accelerator, not an authority.
Where AI Is Heading
AI is a powerful accelerator — not a substitute. Teams that use AI wisely deliver faster, reduce rework, and improve quality. However teams that rely on AI alone create risk.
The future of accessibility is human-led and AI-supported — and the organisations that embrace that balance will build better, more inclusive products.
Services
Digital Product: Learn how AccessUX helps product teams build accessibility into their design and development process, from early design to delivery.