AI can generate a helpful first draft of alt text for image descriptions; however, it may overlook important context, misidentify people or objects, or produce unsafe or inaccurate descriptions, which can negatively impact content accessibility.

Recommendation: If you wish to use AI to generate alt text, consider it as an assistive drafting tool, not the final authority. AI generated content is NOT a replacement for human judgement.

• Please review your responsibilities, policies, etc. and do not use university data in any non-university provided system. Review the University's stance on AI. 
• For best practices for image description and alt text, please refer to the "How to Implement" section of "Image Descriptions and Alt text" Fundamentals Guide.

Limitations of AI-Generated Alt Text:

• May miss context and intent - AI usually sees only the pixel content. It can’t reliably infer the author’s intent (why the image is included) or the image’s role in surrounding content, which are essential for meaningful alt text.
• Provide incorrect or misleading descriptions - AI models sometimes mislabel objects, actions, locations, or people. An incorrect alt attribute can be worse than none, because it conveys false information to assistive technology users.
• Overly verbose or under-specific output - Some AI alt text is too long for quick screen-reader consumption; other times it’s too vague (“a man standing”) and lacks the details needed for understanding.
• Bias and stereotyped language - AI models can reflect societal biases or use insensitive phrasing (e.g., describing appearance unnecessarily).
• Legal/compliance uncertainty - AI output alone does not guarantee accessibility compliance.