John B: How Can Programmers Use AI to Evaluate Accessibility?

Vibe Check – John B Asked: How Can Programmers Use AI to Evaluate Accessibility?

John B sent in this excellent question:

“What are techniques that us programmers can use with AI to help us evaluate our code for accessibility?”

Love this one—because yes, AI can help. But also? It has limits.
AI can flag things. It can guide you. But it cannot replace what really matters: human judgment, manual testing, and intentional design thinking.

Let’s break it down.

A personal story – the gold-standard fail

I once reviewed a government web app that looked great on paper. It passed WCAG 2.1 AA. It was fully Section 508 compliant. Every image had alt text. Every form field had a label. The tab order followed a logical path. ARIA roles were applied correctly. If you ran it through any accessibility checker, it came out squeaky clean.

But when I tested it with assistive tech?

The keyboard navigation was linear—but the experience didn’t follow the visual layout, which made it confusing.

Modals didn’t trap focus properly even though aria-modal="true" was used.

Screen readers missed important success messages entirely.

All links were labeled—but used vague text like “Click here” or “Learn more.”

Technically? Accessible.
Real-world experience? Frustrating, disjointed, and emotionally taxing.

That’s when it hit me: compliance is not the same as clarity. You can check every box and still build something unusable.

Elevate the vibe – AI can support, but it can’t solve

Here’s the truth: AI is great at finding some problems. But it doesn’t understand context, emotion, flow, or people. It won’t catch if your content overwhelms someone with anxiety, or if your tooltip can’t be triggered on a touchscreen. It won’t know if your interaction makes sense from a blind person’s perspective.

So when using AI, the goal isn’t to hand off accessibility—it’s to enhance your own process. Think of it as backup—not a replacement.

Vibe boost – ways programmers can use AI to support accessibility

1. Run AI-powered code reviews

Tools like GitHub Copilot or CodeWhisperer can suggest accessibility attributes like aria-label, alt, or semantic HTML.

Use it to:

Generate cleaner HTML structure

Recommend keyboard-accessible components

Flag potential a11y issues early in dev

But always:
Review suggestions with a critical eye—AI often guesses and may apply ARIA where it’s not needed.

2. Pair AI with accessibility linters and tools like axe

While not AI, tools like axe DevTools are essential. They scan your site for real accessibility issues—missing labels, bad contrast, broken landmarks, etc.

Use axe to:

Scan pages inside your browser

Integrate checks into your CI pipeline

Catch low-hanging WCAG violations automatically

And remember:
Axe doesn't cover everything. It won't catch poor UX, emotional overload, or misleading labels. You still need human review.

3. Use AI as a personal explainer

When you’re stuck on what an attribute or pattern means, AI can break it down in plain language.

Ask AI:

“What does aria-live do and when should I use it?”

“What’s the difference between aria-labelledby and aria-describedby?”

“How can I make this button accessible to screen reader users?”

This is a great way to learn instead of copy/pasting blindly.

4. Ask AI from an accessibility lens

AI isn’t just for code—it’s also great for exploring perspectives. Try asking:

“How would someone who is blind interact with this form?”
“What challenges might someone with ADHD face using this page?”
“Does this component work with just a keyboard or voice control?”

These questions push AI to consider human-centered scenarios—not just syntax.

5. Use AI to refactor for semantic HTML

You can give AI a messy block of divs and ask it to refactor using semantic tags (<nav>, <main>, <form>, etc.).

Just remember: AI doesn’t understand meaning. You still need to test and verify that it behaves correctly.

Self-reflection – check your dev + AI accessibility strategy

Ask yourself:

Am I using AI to speed up, or to replace my thinking?

Have I validated AI-generated code manually or with real assistive tech?

Have I tested the full experience, not just the technical structure?

Did I ask how different kinds of people would actually use this?

If you're only relying on AI and automated tools, you're likely missing the bigger picture—and possibly introducing more harm than help.

Vibe in action – what actually makes the difference

Step 1: Use AI to support—but not substitute—your process

It’s a helper, not a decision-maker.

Step 2: Always follow with manual testing

Use screen readers, keyboard navigation, and magnification tools. Test across devices and contexts.

Step 3: Prioritize real code reviews

AI can’t catch nuance like a seasoned dev or accessibility specialist can. Human reviews catch what automation won’t.

Step 4: Don’t just check boxes—test the experience

What does the page feel like? What happens when something goes wrong? Can people recover gracefully?

Vibing out

John B, thank you for this smart and important question. Yes—AI can absolutely help us catch accessibility issues faster. But we can’t let it replace the most critical part of the process: the human touch.

The best accessibility work comes from thinking like a person, not just checking like a machine. That’s why nothing replaces manual testing, code reviews, and asking questions like:
“How would someone who is blind—or anxious, or distracted, or overwhelmed—experience this?”

If you’re reading this and have an accessibility question of your own, head over to the A&Q question board and send it in. I’ll break it down with you, no matter how technical or theoretical it is.

We don’t build for checklists.
We build for people.
That’s the vibe.