If you want an LLM to help with prose without sanding off your voice, I like this boundary: let it flag passive voice, filler words, and paragraphs that might move, but don’t take its suggested wording or encouragement. The warning that praise can reinforce weak first-draft choices is sharp; I’m less convinced by the absolute ban on using any suggested word, which the piece argues for through experience rather than evidence.

Most LLM writing advice starts with generation; this prompt has the model interview the author instead, one question at a time, with up to two rounds of follow-up. I like the guardrails that check new answers against earlier ones and make the final outline trace back to what the author said; it’s a thoughtful way to use the friction of planning, though the piece offers a workflow rather than evidence that it improves finished posts.