I'm blind - ai can help with some stuff, not with other. my favorite is the lack of labels on buttons or text that is supposed to be a table but is not marked as such. the ai should be able to get inside the page code and rewrite it into something usable which is let's say a heavy ask for most ai-s nowadays. the thing must get an enormous context of react js and apply surgical changes on the regular. maybe possible, but heavy, slow, and expensive, and every blind person will have to do it separately.

The better option is for the ai to review the code during development and suggest suitable accessibility measures compared to the spec and common guidelines.

I wouldn’t even use AI to generate image descriptions (because it doesn’t have enough context to convey the meaning, aside from being wrong many times). Designers and developers should see it as part of their trade to build inclusive products, not try to patch it up as an afterthought with Accessibility overlays or AI-based “fixes”.

You can certainly use AI to tick boxes.

Need a caption on an image or video? AI will generte it for you! But the quality is a different question. Is the caption correct? Is the caption usefull? In my test with images it was roughly 60/40 on the correctness.

On the other hand. If you force a company to provide those captions. They might use the cheapest provider. If they do a better job is anyones guess.

First thought that comes to mind, the AI will modify the content, do what the user sees is not the same as what others see. So how can you garantee that the AI does not mess up on some part of content ?

It can definetly help, like with describing images that have no description. But I doubt it can deliver good and consistent results to make whole websites accessible where developer have paid no attention to this topic at all.

From an efficiency standpoint it would be way better when devs run this once and not every visitor.

Whether AI would be a good solution or not, it then pushes the cost to the consumer which would be unfair and create disparities.

Automated caption for videos is probably something that .. somewhat works.