Write the draft yourself, let an LLM find the flaws, then fix every one of them in your own words
The most useful thing a language model can do for your writing is refuse to write any of it.
The method runs in three steps and keeps the authorship where it belongs: you write the first draft, you hand it to a model only so it can flag what is weak, and then you rewrite those passages yourself before asking a fresh session which version actually reads better. Two rules hold the whole thing together, that you may not use a single word an LLM suggests to you, and that you ignore its encouragement, because a model will praise a draft that still needs rethinking. The post is candid about where the help is real, in the mechanical catches: overused passive voice, repeated phrases, filler words, and a paragraph sitting three paragraphs from where it belongs. No special software is required, and while the author built a small custom tool for it, the piece says an ordinary chat window runs the same loop.