Passing the Scan, Failing the User: AI and Accessibility - Learning Trends & Innovations SIG (Virtual)
AI can generate content, format a slide, and confidently tell you a document passes WCAG 2.2 AA… right before it gets the accessibility wrong. This session was inspired by eight hours of vibe coding a single accessible one-pager: what AI got right, where it quietly broke reading order and heading structure, and why a perfect automated accessibility score isn't the same as what works for real people.
In this session, we will discuss how to bring intentional, human-centered accessibility practices into AI-assisted workflows, using AI to support learning and development without letting it decide what "accessible enough" means. You'll leave with a practical way to evaluate AI-generated content, know when to trust automated checkers (and when not to), and keep people at the center of your design decisions.
Learning Objectives / Takeaways
1. Identify risks of using AI and vibe coding to support accessibility and inclusion, including bias, error, and false confidence in "passing" results.
2. Evaluate and correct AI-generated content for reading order, heading structure, and semantic accuracy.
3. Generate and verify alt text and image descriptions using popular AI-supported tools and models.
4. Discover how AI-powered tools can help you create closed captions and transcripts.
5. Build a workflow that combines automated accessibility checkers with real assistive technology testing.
Sarah Mercier, MBA, CPACC, is CEO of Build Capable and a learning technology strategist. She translates complex technical concepts and research into real-world practice grounded in human-centered design. Sarah is the editor of Design for All Learners and an international speaker on accessibility, the ethics of AI, data strategy, and learning ecosystems.
ATD Capability Model:
Personal: Compliance & Ethical Behavior, Lifelong Learning
Professional: Instructional Design, Technology Application
Organizational: Future Readiness
