Everett Zufelt
VP, Strategic Partnerships & Emerging Technology
Orium
Ajinkya Kunjir
Research Lead
Title TBD
Orium
AI-assisted coding tools—such as GitHub Copilot, Cursor, and Claude Code—significantly accelerate interface development through natural-language prompts. However, these systems are trained on large corpora that include years of inaccessible patterns. As a result, they frequently produce visually polished yet inaccessible interfaces.
This session presents a practical method for "accessibility prompt engineering": techniques that guide AI tools toward output aligned with WCAG, ARIA best practices, and inclusive interaction patterns. Through demonstrations and real-world examples, attendees will learn how to shape prompts that yield semantic HTML, meaningful names/roles/states, robust keyboard interaction, and verifiable accessibility. The session also covers how to validate AI-generated code and how to operationalize a reusable prompt library so accessibility becomes a default outcome rather than a post-hoc fix.
By the end of this session, participants will be able to:
-
Diagnose common accessibility failures produced by AI-generated UI code.
-
Write effective prompts that enforce semantic structure, keyboard operability, and sufficient color contrast.
-
Apply ARIA judiciously (and minimally) to complement, not replace, native semantics.
-
Validate outputs using automated checks (axe, Lighthouse, pa11y) and targeted manual testing.
-
Establish a team prompt library and review workflow to maintain accessibility quality at scale.
Orium • A11YTO Conf 2025