The 30% Problem: What the USWDS AI Takeover Teaches Us About Automated Accessibility Testing

When the team that now runs the U.S. Web Design System (USWDS) announced it would “automate as much accessibility testing as possible” with AI code-review tools, accessibility professionals pushed back almost immediately. The reason is a number every WordPress site owner should know: automated tools can reliably catch only about 30% of accessibility problems, according to former USWDS engineering lead Matt Henry. The rest needs human judgment.

What Happened to USWDS

USWDS is the open-source design system that many U.S. federal websites use to build accessible pages. According to FedScoop’s reporting, in September 2026 Treasury CIO Sam Corcos and Treasury AI engineer Ryan Parker took over leadership of the project. They said they plan to automate accessibility testing using AI code-review tools such as OpenAI’s Codex and CodeRabbit.

Reports also describe contributors being removed from the repository and comments being restricted. A coalition of developers and former government officials has told GSA that websites are “in jeopardy”. Nextgov reports that technologists are creating their own copy of the system. The governance dispute is still developing, so we’ll focus on the lesson that applies to everyone: what automated testing can and can’t do.

Why Automation Only Goes So Far

Automated scanners are very good at checking things that have a clear, machine-readable answer. They struggle with anything that depends on meaning or context. The “roughly 30%” figure is an expert estimate, but it matches what most accessibility practitioners see in practice.

What automation does well

• Missing alt attributes on images
• Form inputs with no associated label
• Insufficient color contrast on solid backgrounds
• Missing page language or document title
• Invalid or duplicate ARIA attributes and IDs
• Empty links and buttons

What still needs a human

• Whether alt text actually describes the image’s purpose
• Whether the keyboard focus order makes logical sense
• Whether error messages help someone recover from a mistake
• Whether a custom widget is usable with a screen reader
• Whether captions and transcripts are accurate
• Whether instructions rely on color, shape, or position alone

A tool can confirm that an image has alt text. It cannot reliably confirm that “image123” or “photo of person” is the right description for that image. AI code review can improve on rule-based checks, but it can also be confidently wrong, and it is not a substitute for testing with real assistive technology.

What This Means for Your WordPress Site

Few WordPress site owners are building a federal design system, but the same temptation applies to you: run one scan, see a green score, and consider the job done. Lawsuit data we’ve covered before shows that having a tool installed doesn’t protect a site that still has barriers. Plaintiffs and their experts test with screen readers and keyboards, not just scanners.

The practical takeaway is to treat automation as a fast, repeatable first layer and to plan for the rest.

A Realistic Testing Workflow

1. Scan regularly. Run an automated scan after every theme, plugin, or content change to catch the machine-detectable issues early.
2. Fix the code, not just the symptoms. Correct the underlying markup rather than layering something on top that hides the problem.
3. Tab through your key pages. Navigation, forms, checkout, and contact pages should work with a keyboard alone, with a visible focus indicator.
4. Try a screen reader. Even 15 minutes with NVDA, VoiceOver, or TalkBack on your most important flow will surface problems no scanner flags.
5. Review content by hand. Check alt text, link text, headings, and captions for meaning, not just presence.
6. Listen to users. Publish an accessibility statement with a contact method and respond to feedback.

Be Skeptical of “Fully Automated” Claims

Whether a claim comes from a government agency, an AI vendor, or an overlay company, be wary of anyone promising complete accessibility without human involvement. No tool, including ours, can guarantee WCAG conformance on its own. Good tools are honest about that boundary: they find what can be found, help you fix it, and tell you what still needs manual review.

Take Action

LEWCA’s WCAG scanner finds real code-level issues in your WordPress site, and its toolbar gives visitors helpful display and reading controls. LEWCA Pro adds AI-powered code fixes, scheduled scanning, and compliance reports, so the automatable layer stays covered while you focus on the manual testing only people can do. See the pricing page to compare plans, or download LEWCA and run your first scan today.