Replying to [comment:6 tj.vantoll]:
Replying to [comment:5 joern.zaefferer]: > Have you tried the key combo in JAWS that tells it to read descriptions? As far as I remember, JAWS tends to be less verbose on tooltips, and you have to tell it to actually read tooltips. Might be a good idea to compare with regular non-async tooltips to be sure. I had an example where synchronous tooltips were read by JAWS and asynchronous ones were not. That being said I am no JAWS expert and I did not experiment with its key combos to read the descriptions. I also did not try NVDA so there's definitely some more exploration that could be done here. Perhaps @dylanb could give us some more specific information about this.
VO (iOS), NVDA and JAWS all read titles and aria-describedby automatically. VO on Mac OS X is a little different in that the default settings cause a delay and a bunch of junk information to be read before the "hint" is read. IMHO, most users will turn off that junk announcement and also reduce the delay - especially if they have learned using NVDA or JAWS where this is the default.
From a user perspective, how would a blind user know that there is some "hidden tooltip" that they can have read out by pressing some key? if you would insert the tooltip into the document immediately after the element itself, then the user could discover it by doing DOM navigation - however this would make the tooltip not work naturally inside forms (when applied to form elements) because the natural navigation mode for most users there is forms mode, not DOM nav.