I think that math and programming may be the odd exceptions, as they employ their own language-like constructs. There might be a misconception in the name of LLMs - we say that they’re language models, but really they’re token models, some of which may be human language words, while others may represent other things.
As for the discussions, I agree that I don’t have a distinct narrative in my mind during one, but I also noticed that I don’t really know what exactly I’m going to say when I start a response. So it also feels like the act of responding is actually heavily involved in creating the response, rather than just putting it into words.
BTW, I’ve always wondered if people really think differently, or we just describe it in different ways. I guess we’ll never really know.