This is, I think, a really important topic. As a society (I mean all humans here) we don't yet have sufficient muscle memory for asking questions of the flawed oracle that is LLMs.
We have deep cultural memory -- about a quarter century -- of asking Google for things and, for most of that time, getting back pointers to sources, and those sources being either reliable (e.g. respected newspaper, government website), or detectably suspect (some random blog you've never heard of, known propaganda site, The Onion...).
Google's insane escapade of substituting the responses of an incredibly weak LLM model for the job we've been relying on it for for 25 years is especially unfortunate, because "Google says..." was, while imperfect, a reasonable approximation for a quick reality check in 2010. Today people say "Google says..." followed by whatever the stupid AI Overview model has output.
I get that Google thinks they're saving a ton of money by not using anything close to a frontier model since they run this on billions of searches a day. The risk to Google is that people start to catch on that asking even free-tier ChatGPT is at least twice as likely to give back a correct answer, notice that Google barely provides webpage results other than AI slop anyway, and just stop using Google.com entirely.
Anyway bringing it back to the quote, Google's used to having no responsibility for anything, 'we're just a search engine showing you pointers to other people's stuff.' But I actually hope that, due to liability problems, that habit will be beaten out of them and they'll make a shift to, especially outside the context of an actual chatbot, reduce reliance on their own AI output to 'answer questions with search results.' It's too risky to just show people, who are used to getting back mostly facts, to replace that with mostly BS coming from that same endpoint. Even with the fine print.