Accessibility is more than finding violations

Accessibility has traditionally involved a combination of automated tools, manual testing, expert reviews, and documentation. As AI becomes increasingly capable, it can take on some of the repetitive parts of this process—helping designers and accessibility specialists work faster.
 

But there is an important distinction:

Finding an accessibility issue is not the same as understanding its impact.
 

A tool may identify a missing label, insufficient color contrast, or a possible ARIA issue. But understanding whether an interaction is actually usable requires context.
 

• How does the experience behave with a keyboard?

• Does the focus order make sense?

• Will a screen reader user understand what is happening?

• Does the solution work within the user's actual workflow?

• Could fixing one accessibility issue introduce another usability problem?
   

These are areas where human judgment remains essential.
 

Where AI can help

I see AI as an accelerator for accessibility work, particularly when dealing with repetitive analysis and documentation.
 

Analyze

AI can help identify potential accessibility issues across large amounts of UI content, code, or design output.

​

Suggest

It can suggest possible solutions—for example, alternative color combinations, semantic structures, or ARIA approaches.

 

Document

AI can help turn findings into structured accessibility reports, annotations, remediation guidance, and developer-ready documentation.

 

Organize

It can help summarize findings, group similar issues, and highlight patterns that might otherwise take significant manual effort to identify.

This doesn't eliminate the accessibility review.

 

It creates more time for the parts that require expertise.

​

But accessibility needs context

Consider a simple example.

 

An AI system identifies that an interactive element doesn't have an accessible name.

It might suggest: Add an aria-label.

Technically, that could address the immediate issue.

 

But an accessibility specialist needs to ask:

• Is the element actually interactive?

• Does it need to be exposed to assistive technology?

• Is there already visible text that should provide the accessible name?

• Is the element semantically correct?

• Will the interaction make sense when navigated with a keyboard?

• How will it be announced by different screen readers?

• Is the interaction itself understandable?

 

The correct solution isn't always “add ARIA.”

Sometimes the better solution is to change the component, markup, interaction, or information architecture.

​

That's where accessibility becomes UX.

 

The 30/70 model

I think about AI-assisted accessibility through a simple working model:

​

30% - AI assistance

Analyze · Suggest · Organize · Document

AI can take care of repetitive, time-consuming activities and help surface potential issues faster.

 

70% - Human expertise

Interpret · Validate · Decide · Advocate

Accessibility professionals bring the context needed to determine whether a solution actually works for people.

 

AI helps find and organize the work. Humans decide what good accessibility looks like.

 

The 30/70 split isn't a universal measurement. It's a way of thinking about where AI should support the accessibility process and where human judgment should remain in control.

 

What still needs human testing?

Some accessibility behaviors are difficult to evaluate through automation alone.

 

Keyboard interaction

• Does the focus order follow a logical sequence?

• Can users reach every interactive element?

• Does focus remain visible and meaningful?

Screen reader experience

• Does the interface communicate the right information?

• Are dynamic updates announced appropriately?

• Does the user understand what changed and what action is expected?

Focus management

• What happens when a modal opens?

• Where does focus go?

• Where does it return when the modal closes?

Zoom and reflow

• Does the interface remain usable when content is magnified?

• Does information become hidden, clipped, or difficult to navigate?

Real-world workflows

Even if individual components pass accessibility checks, can someone successfully complete the entire task?

 

This is why accessibility cannot be reduced to automated scanning or a list of WCAG criteria.

 

Accessibility is not a checklist

WCAG provides an important foundation. Automated tools provide valuable support. AI can make the process faster. But compliance alone doesn't guarantee a good experience.

 

A product can technically meet a requirement and still leave users confused.

It can have correct ARIA attributes and still have a poor interaction model.

It can pass an automated scan and still be difficult to navigate with a screen reader.

 

The real question is not simply:

“Does this pass?”

 

It is:

“Can people actually use this?”

That shift - from compliance to usability - is where accessibility becomes part of UX rather than a final quality check.

AI should reduce repetitive work - not replace judgment

The most valuable opportunity I see isn't replacing accessibility specialists with AI.

It's giving them more time to do the work only humans can do well.

Instead of spending hours manually organizing findings or documenting repetitive issues, an accessibility specialist could spend more time:

• Evaluating complex interactions

• Testing real workflows

• Working with designers and developers

• Understanding assistive technology behavior

• Advocating for inclusive product decisions

• Thinking about the experience from the user's perspective

In other words:

Let AI handle more of the repetition so humans can focus more on the reasoning.

The future of accessibility is collaborative

I don't see AI and accessibility expertise as competing forces.

I see them working together.
 

AI can be the accelerator.
Accessibility expertise remains the navigator.
The user remains the reason.
 

The best accessibility workflow will combine automated analysis, AI-assisted exploration, expert validation, and real-world testing.
 

Because ultimately, accessibility isn't about making a product pass a checklist.

It's about making sure people can use it.
 

A final thought

As designers, we have an opportunity to use AI thoughtfully—not just to create things faster, but to create better experiences for more people.
 

The question shouldn't be:

“Can AI do accessibility?”
 

It should be:

“How can we use AI to make accessibility better?”
 

That's where I believe the real opportunity lies.