Is it more likely to be clear and reliable if it is AI-written, or are features associated (both directly and by correlation) with clear writing increasingly misperceived as “AI tells” because they are also favored in LLM training?

I don't find the LLM written stuff very readable because after one too many "real"s or "The X Dilemma" my brain shuts off. It's not even voluntary, it just does that on its own.