Using ML to determine the order of matches is absolutely fine, but to "digest" the internet and cook up an answer to what I'm looking for without proper sourcing? I do not want that. I don't want to try and guess what biases the language model might have. It's way easier for me to gauge the bias of another human, and for that, I need to be sent to a page a human has written.
(Of course, I realize that the "blog written by a bot" genre of writing is also becoming more convincing, making this whole thing harder...)