What Automation Can and Can't Fix in Accessibility

As AI shows up in more of the tools we use every day, it’s natural to wonder how well LLMs can write and test accessible code. Automated tools that find and fix issues play an essential role. It’s the only practical way to find and fix issues across thousands of pages as a site changes, and it resolves common issues faster and more consistently than any manual process could. 

But it can’t do everything on its own. 

To get the most from automation, it helps to understand where it excels and where human judgment still matters. The question worth asking isn’t “does AI fix accessibility?” It’s “which parts of accessibility can be reduced to a rule a machine can check?”

How Accurate is Automated Accessibility Testing?

Automation covers a defined and measurable share of WCAG, not all of it. AudioEye’s automation, for example, catches 37 of the 55 WCAG 2.2 Level AA success criteria, which leaves 18 criteria that no automated tool can fully evaluate on its own. That remainder is not a gap in the technology. It’s the set of criteria that requires a person to make a judgment.

Accuracy also varies widely from tool to tool. On rule-based checks, like color contrast, flagged issues are almost real. The bigger difference is how much each tool finds. Two scanners testing the same site, for example, can return very different results because some detect more issues than others. AudioEye, for instance, detects 2.5x more issues than other leading tools. A scan is only as accurate as the tool running it, so a lower issue count doesn’t always mean a more accessible site.

So when a vendor claims its automation covers most accessibility issues, it is worth asking what that figure is measured against. Most WCAG criteria, or just the ones the tool already knows how to check?

The more serious failure mode is a site that returns a clean report yet remains unusable for screen readers, because nothing in that result signals that further review is needed. A clean scan is worth having, but read it narrowly: it tells you the rule-based issues are handled. It does not tell you the site meets WCAG 2.2 Level AA.

Where the Human Expert Layer Takes Over

The human expert layer takes over on assistive technology testing, custom component validation, and judgment calls, which are the WCAG criteria that automation may flag, but can’t resolve.

Concretely, that work includes:

• Assistive technology testing: Walking through key user flows with a screen reader and checking that it announces elements on the page correctly. A page that has valid markup doesn’t guarantee a usable screen reader experience.

• Custom component validation: Using interactive components with keyboard commands or assistive technology, checking for focus management, state announcement, and escape behavior.

• Editorial judgment: Reviewing alt text for purpose, not just presence, assessing reading order against content hierarchy, and evaluating whether error messages and link text are actually useful.

• Fixing what automation flags but can’t repair: Automation detects more than it can actually repair. Those issues are real and still have to be fixed, which means someone has to complete them.

• Conformance documentation: Producing defensible evidence of what was tested, how, and against which success criteria.

This isn't a one-time pass. Sites change, components get rebuilt, and content publishes continuously, so the expert layer has to run alongside the automated one.

Closing the Gap Automation Leaves

Any vendor can tell you what automation catches. The more useful question is what happens to everything it misses. That’s the work that either gets done for you, gets passed to your team, or, worse, is ignored entirely. 

AudioEye's platform shows what was fixed automatically and what was routed for human review separately, rather than folding both into a single pass/fail score. Flagged issues route to Expert Audits, where AudioEye's specialists and assistive technology users identify the high-impact issues automation can't, and to Custom Fixes, where experts write and deploy the code so those issues get closed instead of added to your backlog.

That distinction has real stakes. Pairing automation with Expert Audits and Custom Fixes gets you to roughly 97% issue coverage and delivers 400% more protection from valid legal claims than automation-only tools. AudioEye Assurance backs that coverage: if a claim comes in tied to a WCAG violation inside your covered scope, AudioEye's team supports your response directly.

Automation has a real and bounded limit. AudioEye builds its coverage and its legal protection on top of that limit rather than pretending it isn’t there. 

Ready to see what automation can fix on your site, and what it can’t? Talk to an expert now.