Yep, I enabled it. The rest of the UI is fine as it seems to use regular HTML so it will show up in the accessibility tree. The problem is the text editing area itself which uses canvas.
I think this is actually a great example on the potential pit-falls of going with this approach. This isn't really accessible and shows how even a large company with a lot of resources can miss the mark.
For example, if I try to have VoiceOver (control + option + w) read the word I am currently on in Google Docs, it doesn't work, Using VoiceOver's search (control + option + f in Safari) doesn't work, basically nothing actually seems to work from what I can tell. A visual user can easily scan a document and find a heading, where a VoiceOver user might search to jump to it instead.
While they do provide some keyboard shortcuts to help navigate a document easier, this is not a valid substitute for supporting the screenreader the way users know how to use / are proficient with.
While a user of a screenreader could technically use it, it isn't usable (or accessible) in any real sense of the word. If you were to translate this to a visual user experience, imagine you could only see one letter at a time and can go forwards/backwards one character.
It seems like they are using an `aria-live` region to speak where you are rather than having a separate DOM representation of the content and syncing the state (which can be an engineering effort most companies / libraries cannot afford invest enough to get completely right).
Disclaimer: I am not a primarily a screenreader user, though I do have some experience using them.
Disclaimer: I used to work at Google, though not on Docs.