We don't have any experience with accessibility, so Claude will know more than us.
Quick aside. Claude is AI, in case you just woke up from a deep three-year sleep.
Anyways, I've heard someone say this the other day. And you know what? They weren't wrong!
Claude probably has more stored accessibility information than all the people (myself included) in most rooms I've been in. It can explain ARIA, suggest descriptive text, point out missing labels and recite or summarise WCAG.
But that’s not the same as experience.
Experience is knowing when ARIA makes things worse. Experience is noticing that a keyboard flow technically works but feels just awful. It's understanding that "passes automated checks" and "is usable" are two different things.
When you already have some accessibility knowledge, AI can be useful. It can speed up research, remind you of things you forgot and give you a first draft, a checklist or a plain-English explanation of a confusing guideline.
But when nobody in that room knows about accessibility, AI's answers can feel like certainty.
And certainty is dangerous when you have no way of testing it.
If nobody on the team knows accessibility, AI doesn't become the expert. It becomes the most confident voice in that room.
I doubt the problem is that it gets accessibility wrong all, most or some of the time. I think the problem is that you may not know when it does.