By Renea Davis-Leathers, Communications, Technology, and Training Specialist, Systemwide Procurement
Accessibility remediation is usually slow, manual work. A collaboration between the systemwide AI Community of Practice, UCOP Systemwide Procurement, and UC Irvine found a faster path: configure an AI tool to review content against WCAG 2.1 accessibility standards and UC brand guidelines, then let it handle the tedious first pass.
At UCOP, that took the shape of a custom GPT, the UC Brand & WCAG 2.1 Accessibility Reviewer. Give it a PDF or a document from the Microsoft Office Suite and it will provide issues and suggested changes. Point it at HTML content and it rebuilds the output to better meet WCAG 2.1 standards while staying on UC brand. The tool is useful, but it isn’t perfect. And the approach is not tied to one product. Any AI platform that accepts custom instructions and knowledge files can be configured the same way.
The way it got built is the better story, because no one person built it.
The Problem: Strong Content, Inaccessible Delivery
This journey began with an automated IT hardware supply chain forecast from UC Irvine OIT Procurement. The analysis was solid, but the report’s structure made it difficult to consume and unsuitable for publication. When evaluated with WAVE, WebAIM’s free accessibility checker, it earned a score of 5.8 out of 10—a reminder that even great content can fall short when accessibility is overlooked.
The Turning Point: Borrow a Solution and Iterate
Instead of starting from scratch, we assembled insights from multiple contributors and iteratively refined the solution:
- Andrea King, Systemwide Associate Director, Talent Development, shared an initial GPT build through the AI Community of Practice group.
- Brian McNeilly, Web Accessibility Specialist, consulted on the WCAG 2.1 requirements.
- I (Renea Davis-Leathers, Communications, Technology, and Training Specialist) found a GitHub repo with the full WCAG 2.1 standard as JSON and added it to the GPT as a machine-readable knowledge source.
- Justin Sabo, Product Manager, Automation and AI, and I rewrote the GPT instructions to define standard, predictable outputs.
The result: the same report, now corrected, scored 10 out of 10 on WAVE. The win did not come from one person being clever; it came from a collaboration loop.
Two Lessons Worth Stealing
- Stop Thinking in PDFs. Treat Standards Like Reusable Data.
Our first attempt scraped the WCAG site into large PDFs, which maxed out the GPT’s knowledge sources and left us with a fragile setup. One structured JSON (JavaScript Object Notation) file—a structured, machine-readable format—did the job better.
- AI Agents Need Constraints, Not Just References.
The UC brand guidelines were designed for people, not machines. While the full color palette offers flexibility for human designers, it can introduce unnecessary ambiguity for an AI. Instead, we used the UC Brand Accessible Color Guide as a focused reference, helping the tool make choices that were both on brand and accessible.
Try It, and Get Out of Your Silo
The recipe is simple: a machine-readable standard, a constrained set of brand rules, and instructions that define the output. It works on any AI platform your location supports.
- If you are UCOP staff, use the UC Brand & WCAG 2.1 Accessibility Reviewer in the UCOP ChatGPT workspace.
- If you prefer a different AI, take a look at how we structured the tool and rebuild it for your own team: View the GPT structure.
- Check your own pages with WAVE: WAVE Accessibility Evaluation Tool.
- Check out all the accessibility tools available to you on the UCOP site: UCOP Accessibility Tools and Testing Resources.
As a final note, if you are experimenting with AI on your own, find a buddy. Swap prompts, peer review each other’s builds, or block 30 minutes for pair prompting. Solo experiments are critical, but shared learning is how we scale.
Contact
Renea Davis-Leathers
Communications, Technology, and Training Specialist
Systemwide Procurement | UC Office of the President
UC Tech Project Series
Tell us about how you and your team is using AI or about a tech project in general (it does not have to involve AI) your team has completed, is working on, or is preparing to work on. Your story, if accepted for publication, will be shared out on the UC Tech News site, in the UC Tech News newsletter (room/timing permitting), and on the UC Tech News LinkedIn page.
Please use the UC Tech News blog intake form to submit your UC Tech Project story: https://app.smartsheet.com/b/form/be650a2cf625471684bd94d18122d17b
Questions? Email us at UCTechNews@ucop.edu.
