Making Course Materials Accessible with AI

Accessibility work includes more than adding machine-generated alt text. A useful text alternative depends on the image’s purpose and context: a decorative image may need no descriptive text, while a chart may require its main trend and values. Captions should convey spoken dialogue and relevant non-speech audio, remain synchronized, and identify speakers when needed. Scanned documents may need OCR, headings, reading order, table structure, and meaningful links. AI can draft these elements or flag issues, but may hallucinate visual details, omit important data, misidentify speakers, or produce captions with timing errors. Reviewers should compare descriptions with the source, test documents with assistive technologies, and include users with disabilities in evaluation when possible. W3C’s WCAG defines criteria such as text alternatives for non-text content and captions for prerecorded media; legal requirements and applicable standards depend on context and jurisdiction. A generative tool’s declaration of “accessible” is not a conformance assessment. Institutions should maintain original files, check automated changes, and ensure students can request accommodation through established channels. Privacy matters when recordings or student materials are sent to third-party services. Design for the specific learning purpose: a slide chart description should communicate the finding a learner needs, not merely enumerate visual features. AI can reduce repetitive work, but quality depends on human review and testing with actual course workflows. Testing should reflect actual student tasks.