Keep in mind these things don’t really know anything. They’re good at saying things that seem to fit the situation because that’s how they’re trained. They are like that person you may know that thinks he knows everything and will just say stuff that sounds right to them. The only difference is the ai is a lot more practiced than the human. Google’s llm may have some filtering done on the output to at least make sure that all of the books it recommends are real though it wouldn’t surprise me if there’s a fake one in the list somewhere. These things are prone to “hallucinations” which some lawyers found out the hard way.
The only things JSON has over xml is that it’s easier to write a parser for it and the format is less verbose and less complicated. There are extensions to JSON that can add features that xml has and the JSON spec doesn’t have. Overall the xml spec is bigger and has more features but that also makes it overkill for many of the cases that it would be used in.