This really hasn't matched my experience, and maybe I'm just blind to it. I have absolutely not found any LLM on the market able to produce text or organize my ideas well, and I spend all days cajoling LLMs into producing text for a living.
I've found it very useful for proof-reading, and calling me out on blind-spots. I'll tell ChatGPT, Claude, and Anthropic that my drafts were written by a contractor and I need a rating out of ten to figure out how much I should pay him. They come back with ideas. They often wildly disagree. I will absolutely ask it to redraft stuff to give me inspiration, or to take a stab at a paragraph to unblock me, but the work produced is almost always dreck that needs heavy rework to get to a place I'm happy with. I will ask its opinion on if an analogy I have created works, but I've found if I ask it for analogies by itself it rarely comes up with anything useful.
I've found it immensely useful for educating myself. For me, learning needs to be interactive to stick. I learn something by asking many clarifying questions about it: "does that imply that" and "well isn't that the same as" and "why like this instead of like that" until I really get it, and the models are beautiful for that. It doesn't -- to me -- feel like I'm atrophying my thinking skills when I do this, it feels like I am managing to implant new and useful concepts because I can really sink my teeth into them.
In short, I think it's improved my writing by challenging me, and I think it's helped me understand complex topics much more efficiently than I would have done by banging my head against a text book. My thinking skills feel sharper, not weaker, from the exercise.