Monday, September 29, 2025, 4:00 PM - 4:45 PM EDT
We all know what a challenge it can be to keep a large site up to the newest accessibility standards. At Santa Clara University, we have over 250 non-expert contributors using our CMS. We also had 37,000 images without captions and an eight-year-old backlog of html issues. Training content editors helped, but it wasn't enough, so we turned to AI.
In this session, we'll describe how we leveraged AI to generate accurate, meaningful image captions at scale, while trying to avoid embarrassing (and sometimes problematic) missteps AI is known for. From describing a headshot of one of our Board of Regents as "portrait of an elderly man" to an AI overly eager to classify race, we had to continually refine our prompts to get appropriate, inclusive alt text. We'll share what worked, what didn't and how we fine-tuned the system to improve results.
But images were only part of the challenge. We also tackled accessibility issues in rich text HTML blocks — cleaning up WCAG violations that content editors unknowingly introduced. We fed the content blocks into the AI to identify errors and fix them in a way that aligns with best practices while keeping content intact.
This talk will share real-world insights, technical takeaways and maybe even a few laughs from our AI's more "creative" attempts.
