Accessibility

The Rule That Governs Everything Here

AI can be a useful assistant in accessibility work. It can help identify potential barriers, suggest improvements, and support tasks that might otherwise require significant time and effort.

However, AI should support—not replace—human judgment. Accessibility depends on context, purpose, and how people actually experience content. An AI tool may identify a potential issue or suggest a solution, but its output should not be treated as confirmation that content is accessible.

Use AI to help you identify, review, and remediate potential accessibility barriers, then evaluate the result using appropriate accessibility practices and tools.

For broader guidance on creating accessible digital content, accessibility standards, and remediation practices, visit the Maricopa Digital Accessibility site.

The examples below highlight areas where AI can be helpful, along with limitations and considerations to keep in mind.

Where AI Helps, and What to Check

What AI does well: describes what is visibly in an image, quickly, at scale. For a backlog of undescribed images this is a real improvement over nothing.

Where it fails: alt text is not a description of an image. It is a description of what the image is doing on that page. AI does not know that. The same photograph needs entirely different alt text in a history course and a photography course.

What to check every time:

  • Does it convey the point the image is making, or just list what is pictured?
  • Is it the right length? AI tends toward over-description; if the image is decorative, the correct alt text is empty.
  • Does it include text that appears in the image? AI often misses or misreads embedded text.
  • For charts and diagrams: is the data conveyed, not just “a bar chart showing results”? A complex graphic usually needs a text description nearby, not only alt text.
  • Are people described appropriately, without guessing at race, gender, age, or disability?

What AI does well: automatic speech recognition has improved substantially and handles clear, standard-accent speech in good audio well.

Where it fails, reliably:

  • Names. Student names, faculty names, place names. Almost always wrong.
  • Technical and discipline-specific vocabulary — precisely the words that carry the meaning in a lecture.
  • Accented speech and non-native English, which is an equity problem: the students most likely to depend on captions are frequently the ones whose instructor’s speech the system handles worst.
  • Speaker identification in discussions.
  • Punctuation, which changes meaning more than people expect.

What to check: every proper noun, every technical term, and anything where a mis-transcription would change meaning. Auto-captions without review do not meet accessibility requirements — they are a starting point.

What AI does well: rewriting dense material more simply, producing summaries, and generating multiple versions at different levels.

Where it fails: simplification loses conditions and exceptions. A policy rewritten in plain language can quietly drop the “unless,” the deadline, or the qualifying circumstance — and the result is clear, readable, and wrong.

What to check: compare against the original specifically for anything conditional. If the material has legal or policy consequences, someone who knows the policy reviews the simplified version.

Worth noting: plain language benefits everyone, not only readers with disabilities. This is among the highest-value uses of AI in this section.

What AI does well: converting content between formats, producing audio versions, restructuring material for different presentation, drafting summaries alongside full text.

Where it fails: structure. A document that looks organized may have no real heading structure underneath, which is what a screen reader actually uses. AI conversions frequently produce visually-formatted output with no semantic structure at all.

What to check: real headings rather than large bold text, tables with proper headers, lists marked as lists, and reading order that makes sense when the visual layout is removed.

  • Generated images need alt text like any other image, and the person who generated them is the one who knows what they are for.
  • AI-generated documents often have no heading structure — the output looks organized without being structured. Slide decks created in Gemini Notebook are just images, not actual slides with text on them making them impossible to view with a screen reader without the necessary alt text on each slide.
  • Interactive AI features added to a course or service may not be keyboard-navigable or screen-reader compatible. A chatbot dropped into a page can make that page less accessible than it was.
  • Requiring an AI tool assumes every student can access it — reliable internet, a compatible device, an account, and a tool that works with their assistive technology. Any required tool needs an alternative path.
  • AI detection tools used on student writing raise a documented equity problem: higher false-positive rates for non-native English speakers, and concerns about writing produced with assistive technology.

A Short Checklist

Before AI-assisted material reaches students:

  1. Alt text — reviewed for context, not just accuracy. Decorative images marked as decorative.
  2. Captions — names and technical terms corrected.
  3. Headings — real heading structure, properly nested, no skipped levels.
  4. Links — descriptive text, never “click here” or a bare URL.
  5. Color — never the only way information is conveyed.
  6. Simplified text — checked against the original for lost conditions.
  7. Any required tool — confirmed to work with assistive technology, with an alternative available.

Getting help

MCCCD Disability Resources can advise on specific accessibility questions and review material you are unsure about. If you are building something students will be required to use, involve them early — a conversation at the design stage costs far less than a rebuild.

For AI tools specifically, accessibility is one of the review gates covered in Choosing the Right Tool. Questions can also go to the ARC at ai-resource-center@domail.maricopa.edu.