[eyelidlessness]

I don’t understand, the screen reader is doing the correct thing. People tweet math, code, all sorts of things that aren’t prose formatted with hacky unintended use of unicode. Maybe some form of AI could detect that, but I wouldn’t bet accessibility on it.

The real issue is that there’s a general use case for basic formatting of real prose, and perhaps unicode should accommodate that. But expecting screen readers to understand these kinds of workarounds is unrealistic, and expecting it despite knowing it isn’t supported is user hostile.