Using AI for Accessibility Compliance
A critical deadline is approaching for higher education institutions in the USA. There’s an April 24th, 2026 deadline for public colleges and universities to meet the Web Content Accessibility Guidelines. This means that all academic course content and institutional website materials must meet these guidelines, which is a shift from the prior rule that emphasized accommodation on request standards rather than institution-wide compliance. While the task may seem daunting, artificial intelligence tools offer promising solutions to help institutions meet these requirements more efficiently.
Meeting the ADA Challenge: AI Solutions for Equations
During a recent workshop, an attendee, who taught statistics, asked if AI could help generate text that could allow screen readers to read handwritten equations. An attendee, who taught statistics, asked if AI could help generate text that could allow screen readers to read handwritten equations. For disciplines such as mathematics, statistics, and chemistry that rely heavily on handwritten equations and complex mathematical notation, meeting ADA requirements can pose significant challenges, particularly when converting these visual elements into screen reader-friendly formats.
I suggested that generative AI could help with this. In theory, you should be able to paste a file or screenshot of an equation into generative AI, then ask it to convert it into text. (As I understand it, there are three main approaches: alt-text, LaTeX, and MathML, all of which should be readable by screen reader software. There are advantages and disadvantages to each approach, which are beyond my knowledge, so I won’t get into those here.) A chemistry professor who was attending my workshop gave it a try and said, even though the image file that she uploaded was not the clearest, AI did a pretty good job of converting the equation into text.
Here’s an example of how this might work. By the way, I’m using Gemini here because I think it does the best job of reading images, but you could experiment with other AI tools. The first step is to paste in a screenshot of an equation. I could also have uploaded an image file. Next, I simply ask Gemini to convert the image into alt-text, LaTeX, and MathML. Of course, if you only needed one of these, you could simplify the prompt a little bit.
In just a few seconds, Gemini came back with the following.
This was a pretty simple test, but it shows what’s possible. The Alt-Text and LaTeX seem spot on. I don’t know MathML, but I’m assuming it’s correct although I really don’t know. I tried a second test with a little more complicated formula and got equally solid results.
If you have dozens or even hundreds of such handwritten equations to convert, you could use generative AI to help you write a Python script that would do this in a more automated way. This could get a bit complicated, and you would probably need an API key to an AI service. But if the job is big enough, the investment may be well worth the time. There are some tricks that can keep the cost down. If you’re interested in learning more about this, just add a comment requesting more information and I’ll be happy to share what I’ve discovered.
AI for Alt-Text Generation
Many of us don’t have equations to deal with, thankfully. But we may have large batches of PowerPoint files that are heavy with images. In order to be compliant, those images will need to have alt-text, which is a text description of the image. The slides for my undergraduate class are not particularly image-heavy, but it’s not unusual for me to have 20 to 25 images in a slide deck. Multiply that by a dozen or so slide decks, and it’s a pretty tedious job to go through and create the alt-text for each one of those images.
It turns out that Gemini is pretty good at creating the alt-text from a slide deck. The process is quite simple. You can either upload a PowerPoint file, or if you happen to use Google Slides, you can simply link to a Google Slide deck from within Gemini, then use a pretty straightforward prompt. There’s an example below.
I like to keep prompts simple when I can. So I thought I would try a simple prompt first. Good news! It worked pretty well. Here’s part of Gemini’s response.
A couple of these are pretty interesting. Gemini nailed all three descriptions, even getting the tricolor Rough Collie right. (Yes, I was having fun with image generators when I made that slide.)
Slide 16 has a little cartoon border collie image I use to point out particularly important points in the slides. Gemini’s description is spot on.
It literally took less than two minutes to do all of this. Now, I still need to go in and add the alt-text, which would be a matter of copying and pasting into PowerPoint. I’ve tried a couple of different ways to automate that using Copilot, but it didn’t work. It is possible to write a Python script that would add these alt-text images, but I have not tested that yet. For now, it’s good enough for me if I can just substantially cut the time down, even if I can’t get 100% of the way there.
It’s worth pointing out that there’s an awful lot going on here. First, Gemini has to use some sort of image recognition to realize what the images are, even to realize that they are images. Then, it has to create reasonable text to describe the image. Finally, it has to tie that description to individual slides - that’s a lot for AI to get right. In my tests, there are always some small problems, but they’re small enough to not deter me from using this approach.
Another important point here is that none of this is particularly complicated. You saw how simple the prompt was, and it worked really well. Now, if you wanted to automate this completely, it would take a little bit of time to figure out the coding to get it right, but that may be well worth the investment, especially if you’re going to be doing this for an entire department or for a large set of classes.
Remember that the accuracy of the alt-text is on you. So, be sure to double-check, especially for important images such as explanatory diagrams. (It would also be a tragedy if a Rough Collie was misinterpreted as a Sheltie. Both breeds would be insulted.)
My final point is that I really didn’t know if this was going to work. I thought it might. It made sense that AI could handle this, but I didn’t know. I just tried it. If I tried it and it didn’t work, there wouldn’t have been a bunch of time lost. In this case, it did work pretty well. So don’t be afraid to try things with AI. Some things will work, some things won’t. But as the saying goes, you’ll never know until you try.
Want to continue this conversation? I’d love to hear your thoughts on how you’re using AI. Drop me a line at Craig@AIGoesToCollege.com. Be sure to check out the AI Goes to College podcast, which I co-host with Dr. Robert E. Crossler. It’s available at https://www.aigoestocollege.com/follow.
Looking for practical guidance on AI in higher education? I offer engaging workshops and talks—both remotely and in person—on using AI to enhance learning while preserving academic integrity. Email me to discuss bringing these insights to your institution, or feel free to share my contact information with your professional development team.