When properly and thoughtfully implemented, generative AI can serve as an accessible mode of interaction. A key facet of accessibility is providing several ways for a user to perceive or gather information to accommodate for varying styles of behavior, so AI can accommodate those who prefer to enter or select from prompts. Here are some guidelines that will help ensure that generative AI can be used as a helpful mode of discovery, designed for all potential users:

• Ensure that AI-generated imagery has relevant and descriptive alt text
• Ensure that the AI interface is keyboard navigable
  – Can a user who only uses a keyboard successfully submit or select a prompt/prompt suggestion, choose from generated options, etc.?
• Ensure that a screen reader provides clear guidance to a user
  – Will a screen reader user know when generation is in progress? Will a screen reader user be aware of the generated options being presented, and how to select one if applicable?
• Ensure proper labels on AI prompt inputs
  – A visual label clearly associated with an input (directly above, directly preceding, or within) is preferred. Otherwise, use aria-label or aria-labelledby when associated with a matching id.
• Use clear language and provide headings to break up long chunks of text
• Allow the user to edit prompt submission or “go back” and try again
  – Tolerance for error is a crucial part of accessibility. A user should not be locked in to a submission or selection
• While not a requirement, providing concise and relevant prompt options for a user can help those who might not already have a prompt in mind
• Even better: allow for speech input, and allow the user to edit what they’ve dictated

As the world of generative AI continues to expand, so will accessibility considerations. This post will be updated with inevitable new findings.

Further reading:

https://hbr.org/2023/08/designing-generative-ai-to-work-for-people-with-disabilities

https://teaching.cornell.edu/generative-artificial-intelligence/ai-accessibility