I feel like the things you list had more human input, as built, and more scope to take human feedback into account to amend any issues.
The latter could of course be similarly used to refine AI versions, but as cost-cutting is a major attraction, this seems unlikely to happen unless the AI is do poor that the errors cost too much otherwise.
As things stand, we now must guess what a customer service bot or search engine might understand, framing our terms to fit our beliefs about how massed groups might make the same enquiry. Relatively simple tech questions are met, not with the conversations initiated by people with similar queries, but with reams of links to material offering solutions to almost the opposite problem. I.e.: “how do I disable x on y?” must be asked “disable x y?”, but you’ll struggle to find any link which isn’t “y is great, learn about the cool new x feature on y”, x is great, enable x", “what to do if you cannot enable x”. Maybe some bots telling you why wanting to disabling x is futile & wouldn’t you like to learn how great x is? “Your life will suffer if you disable x. You won’t be able to do things you never ever & would never want to do if you disable x” Which, for some things, there’s some validity to some parts of that, but for disabling a bloatware messaging app? Not so much, and potentially just indicative of terrible architecture.
Perhaps I should be optimistic that AI will rapidly patch this type of issue, being able to return responses that have a good analysis of the query, but feel it is more likely that flawed AI will hijack the whole.
It has been a good run, but swathes of us have lost what were standard means of seeking information.