As I sometimes do when I'm researching for a blog post, if a comment or an idea piques my interest, I immediately jump down a rabbit hole. The DigiCol A11Y Summit was no exception.
While watching the presentation titled The future of document accessibility starts here, given by Mac Clemmens of CivicPlus and Shawn Jordison, The Accessibility Guy, I heard Mac Clemmens say, You can get PAC2024, you can get PAC2026. And off I went down the rabbit hole. I wanted to know more about PAC2026.
One of the first resources in the search results was a video from, you guessed it, The Accessibility Guy titled PAC 2026 | AI Is Here | The Accessibility Guy News. He explained the new AI tab that has been added to PAC2026 and said:
My understanding of the tool is that it's going to help us identify whether or not something should be marked as a paragraph versus being marked as a heading. And I think this could be pretty valuable for users that are getting into accessibility for the first time.
That last sentence got my attention.
I think this could be pretty valuable for users that are getting into accessibility for the first time.
My immediate reaction was, Wow. That's an interesting perspective on AI and accessibility that I haven't really seen discussed before. Most of the conversation around AI in accessibility focuses on what it gets wrong, what it automates, or whether it's trustworthy enough to rely on. But what about its potential as a learning tool for people who are just getting started?
The Case for AI as An Apprenticeship Tool
Accessibility has always had a steep learning curve with WCAG success criteria, semantic HTML, keyboard behavior, screen reader quirks, etc. Traditionally, you couldn’t skip learning those accessibility elements before conducting a meaningful, more accurate accessibility assessment.
But with AI tools that can identify when content should be structured as a heading rather than a paragraph or explain why a color contrast ratio fails, newcomers have a lower barrier to entry.
It has been argued that machine learning can meaningfully reduce the manual effort accessibility testing requires, making it more approachable for people without deep expertise, while cautioning that AI can't replace human assessment.
This is an interesting viewpoint. AI as an apprenticeship tool. One that explains an issue, ties it to the user experience, and points to the relevant guideline, turning a novice into a practitioner faster than before.
I’ll admit, I would have welcomed AI as a learning tool when I first started out in accessibility. 😮💨
Where AI Reaches Its Limits
The counterargument is just as strong.
A 2026 empirical study of LLM-based web accessibility repair found that while more than 99.7% of AI-generated fixes were syntactically valid and 80.2% reduced accessibility violations, fewer than 26% completely resolved the underlying accessibility issue. The authors concluded that LLMs are effective for partial remediation but are not yet reliable enough to replace human expertise or rule-based validation.
The W3C's own guidance is clear on this issue. Reliable WCAG evaluation requires both automated testing and human evaluation, performed by people who understand how users with different disabilities actually navigate the web. That knowledge isn't something a tool can hand over. It’s acquired over time through experience, deep understanding, training, and consistent practice.
Knowledge Transfer vs. Responsibility Transfer
The real question for any AI accessibility tool is whether it's teaching or just delegating. Is it informing the user that Here's the problem and why it matters, or is it simply stating Here's the problem and the code to paste?
For someone new to accessibility, that’s a huge difference.
The first fosters a practitioner; the other creates someone who is extremely good at clearing the remediation queue without necessarily understanding why.
AI can certainly help lighten the learning curve for accessibility, but it can’t develop sound accessibility judgment. That requires human experience, critical thinking, and the ability to evaluate accessibility in context.
Resources
- Understanding Conformance
- Enhancing Accessibility with AI and ML
- Best AI Accessibility Tools: What Actually Works
- LLM Based Web Accessibility Repair: An Empirical Study of Detection, Remediation, and Cost
A human author creates the DubBlog posts. The AI tools Gemini and ChatGPT are sometimes used to brainstorm subject ideas, generate blog post outlines, and rephrase certain portions of the content. Our marketing team carefully reviews all final drafts for accuracy and authenticity. The opinions and perspectives expressed remain the sole responsibility of the human author.
