A step-by-step workflow for using AI on real writing tasks: brief first, draft second, human edit last — with the checkpoints that keep the result sounding like you.
2026-09-16 · 约 4 分钟读完
作者：林知微（CPCX 撰稿人 · 写作与职场）
I had AI rewrite one paragraph of my own draft “to sound better” — it stripped out every specific example and the paragraph said nothing. That failed rewrite is the reason the edit checklist in this article exists.
Handing a writing task to an AI with no brief is like handing one to an intern on their first morning with no onboarding: whatever comes back will be fluent, generic, and about eighty percent wrong in ways that matter. The usual reaction — “AI writing sounds fake” — describes the missing process more than the tool.
The workflow below takes a real task from blank page to publishable text in three stages. Its one rule: the AI drafts, but you own the decisions — audience, structure, and what stays in the final version.
If you cannot write the brief, you are not ready to write the draft — with or without AI. The brief is the actual thinking; the rest is typing.
Expect to use a fraction of the first draft, not all of it. What AI drafting is genuinely good at: getting past the blank page, offering a structure to react to, producing variants of a sentence you are stuck on. What it is not good at: your specifics. It does not know what your team actually did last week or what your customer actually said, so it will write around the gaps — smoothly, and sometimes inventively.
Read the draft once, mark what is usable, then rewrite the marked parts with real details. Concretely: keep its skeleton, replace its abstractions with your facts, and delete anything that sounds right but says nothing.
Editing with an AI in the loop has no natural endpoint — it will always find something to change, and a fifth pass can easily undo what the third one fixed. Decide in advance what “done” means for the piece: the brief is satisfied, every paragraph passes the specifics test, and one full read-through out loud produces no stumbles. Then stop.
A closing habit worth keeping: after your final edit, run one last check against the brief alone, not the draft. Does the finished text actually do what you set out in stage one? Drafts drift, and the brief is the only anchor that does not.
The strongest role for AI is after you have a version you believe in: ask it to critique the draft against the brief, point out where it loses the reader, or flag paragraphs that promise something the piece never delivers. Critique is a task where outside perspective genuinely helps — and the model has no ego invested in the text.
Used this way, the tool stays in the right place: fast drafts and honest feedback on your structure, while judgment, facts, and voice remain yours. That is also the difference between writing that reads like everyone’s and writing that reads like yours.