The way I see it is, the usefulness of straight LLM generated text is indirectly proportional to the importance of the work. If someone is asking for text for the sake of text and can’t be convinced otherwise, give 'em slop.

But I also feel that properly trained & prompted LLM generated text is a force multiplier when combined with revision and fact checking, also varying indirectly proportional with experience and familiarity with the topic.