"The process is built on a fundamental belief: that the maker of the work knows something about it that no one else can know."
— Liz Lerman《Critical Response Process》(2003)
"Can you review this doc?""This is a rough draft — I only need to know if the argument holds. Ignore wording and typos for now."明确「这一轮要哪一层」，是把读者的时间用在刀刃上，也避免在将被删掉的句子上较真。
"What do you think? Is it good?""Where did your attention drop? Which sentence did you re-read? What's the one line you remember?"把「好不好」这种主观打分，换成「在哪走神、在哪卡壳」这种可回答的诊断——读者的身体反应不会撒谎。
AI 是随叫随到、永不疲倦的读者，能立刻指出哪里不清楚、哪里逻辑断裂——但它的品味是平均值。用它做诊断，别让它代笔；一旦让它「帮你改」，你的声音就被磨成 AI 腔。
"Rewriting is the essence of writing well: it's where the game is won or lost."
— William Zinsser《On Writing Well》(1976)
原理解读
Zinsser 说写作的胜负在修订——而修订需要反馈循环。AI 把这个循环的成本压到几乎为零：凌晨三点、改到第七稿，它都在。它最强的用法是当读者，不是当作者：让它指出哪句话会让人困惑、哪个论点缺支撑、哪里你以为说清了其实没有。关键在指令——要它诊断而非改写。一旦说「帮我改改」，它会顺手磨平你的棱角，换上四平八稳的 AI 腔。还要记住它的盲点：它倾向附和，不会真反对你的核心观点，也没有你的语境和品味。用它扫盲点，把判断权留给自己。
修改示范
（对 AI 说）「帮我润色这篇文章，让它更好。」「不要改写。以一个挑剔读者的身份，指出三处最可能让人困惑或失去兴趣的地方，每处说明为什么，但保留我的用词和语气。」「润色」= 交出方向盘，换来平庸的 AI 腔；「指出三处困惑」= 它当读者、你当作者，反馈精准且你仍掌舵。
"Make this email sound more professional.""Don't rewrite. Tell me where a busy reader might misread my intent, and which sentence buries the ask."前者会把邮件改成千篇一律的商务腔；后者让 AI 当「忙碌读者」替你压力测试，你据此自己改。
"When people tell you something's wrong or doesn't work for them, they are almost always right. When they tell you exactly what they think is wrong and how to fix it, they are almost always wrong."
— Neil Gaiman (2012, 致敬 Stephen King 的反馈原则)
"You misunderstood — that part was intentional.""Good to know you stumbled there. I'll figure out the fix — thanks for flagging it."「你误解了」会让对方闭嘴，下次不再给你真话；承认症状、自己留着开药，才能持续拿到诚实反馈。
这是真实的风险。AI 反馈廉价、即时、不会让你尴尬，确实诱人。但它有两个真人不可替代之处：一是真实利害——同事、用户读了真会做决定，AI 不会；二是意外——真人会带来你和 AI 都想不到的视角，而 AI 倾向给「平均答案」。健康的做法是分工:用 AI 做高频的第一轮自查，把真人的精力省给高价值的关键稿。把 AI 当陪练，把真人留给正赛。