AI 喜欢通过”打磨”把所有东西变成平均值。好的打磨 Prompt 要针对具体弱点(被动、AI 口头禅、行话、死副词),同时不抹平作者声音。
适合哪些场景
把 AI 当编辑的独立写手、跑最后一遍的内容团队、发布前压稿的市场、自动化机械活的文字编辑。
什么时候不建议这样写 Prompt
初稿别用——太早。法律 / 医疗也别用——需专家。
Prompt 结构公式
每个打磨 Prompt 都要带这六个要素:
- 读者:一个具体的人。
- 目标:一个动作——读完 / 点击 / 同意 / 分享。
- 语气:2-3 个锚定形容词。
- 限制:字数、禁用短语、必含事实。
- 格式:段落、列表、小标题、表格。
- 示例:1-2 条语气示例——匹配 voice 的最强杠杆。
这套 Prompt 适合用在哪
- 压缩
- 去行话
- 匹配语气
- 段落流畅度
- AI 口头禅清理
12 个可直接复制的 Prompt 模板
1. 压缩 20%
Tighten this draft by 20% without losing meaning. Cut: dead adverbs, hedges (probably, perhaps, somewhat), filler phrases ("at the end of the day"), redundant pairs ("each and every"). Keep voice. Output side-by-side: cut count + final version.
2. 去行话
Audience: `{nonExpertAudience}`. Find jargon in this draft. For each: replace with plain English OR explain in a parenthetical the first time. Don't kill specificity — keep terms the audience uses.
可替换变量: nonExpertAudience
3. AI 口头禅清理
Find AI clichés in this draft: "In today's fast-paced…", "delve into", "unlock the power of", "navigate the complexities", "harness", "robust". Replace each with the specific thing the cliché was avoiding.
4. 匹配品牌语气
Brand voice anchors: `{anchorAdjectives}`. Sample voice line: `{voiceSample}`. Re-polish this draft to match. Output: changed lines only as diff.
可替换变量: anchorAdjectives 锚定形容词, voiceSample 语气示例
5. 主动语态
Find passive-voice constructions in this draft. For each: replace with active voice unless passive is correct (unknown actor, action emphasised). Don't do every passive — judge case by case.
6. 段落流畅度
Audit paragraph flow: (1) Paragraphs that abandon their topic mid-way, (2) Paragraph breaks in wrong places, (3) Transitions missing where the topic actually shifts. Output: where to split / merge / bridge.
7. 标题 + 首句强化
Rewrite the headline + first 2 sentences. Headline: 8-12 words. First sentence: hooks. Second sentence: pays off the hook. Don't start with "In this article…" or rhetorical questions.
8. 念出来测试
Read this draft aloud. Flag every sentence that: (a) Trips on a comma, (b) Has > 25 words, (c) Has consecutive long sentences (rhythm issue), (d) Forces re-reading. Suggest splits / cuts.
9. 句长多样性
Audit sentence-length variety. Flag stretches of 4+ same-length sentences. Vary by alternating short and long. Don't make everything short — that becomes choppy.
10. 套话替换
For each cliché, generate 3 specific replacements. Use real nouns / numbers. Example: "moving the needle" → "doubling weekly trial sign-ups". Skip clichés that genuinely have no replacement.
11. “每段砍一句”
Apply the discipline: from every paragraph, cut at least one sentence. Choose the weakest. Output: the cut line per paragraph + the leaner version.
12. 语气校准
I think this draft is `{currentTone}` but I want `{targetTone}`. Show me 5 specific paragraphs where the tone is off-target and rewrite each to land on target. Don't rewrite the whole piece.
可替换变量: currentTone 当前语气, targetTone 目标语气
容易踩的坑
- 受众模糊——输出泛泛。
- 没语气锚——所有版本一个味。
- 没限制——字数、禁用、长度。
- 不给示例——语气最强信号。
- 相信初稿——AI 落安全中间值。
- AI 口头禅不删。
- 事实不二审——AI 偶尔自信地错。
优化技巧
- 一定给 1-2 条语气示例。
- 狠下限制。
- 发布前念出来。
- 砍不承重的副词形容词。
- AI 出前两稿,人改第三稿。
- 锚定真实读者中的一个人。
- 标题单看能否传递信息。
实操加深
使用这些 prompt 时,不要只替换一个主题词就直接交付。围绕「校对与语气打磨 Prompt:12 个发稿前最后一关模板」先补齐受众、渠道、长度、语气、参考样例、禁止样式和成功标准,再让模型输出 2 个不同版本做横向比较。好的结果应该能被另一个人直接复用,而不是只有顺滑但空泛的表达。
如果输出看起来像通用模板,下一轮要增加一个真实场景、一个反例和一个可检查指标,例如点击率、转化动作、字数、平台限制或品牌禁区。这样改出来的内容才更像可用资产,而不是一次性的灵感草稿。
FAQ
- 文章多长?: 按渠道——注意力短的地方就短。
- AI 写全稿?: 前两稿 AI,第三稿人。
- 多久刷新?: 受众或事实变了,或季度重看常青内容。
- 不审就发?: 别——AI 自信但不一定对。
- 单一 voice 还是分渠道?: 品牌 voice 一个,渠道在轴内漂移。
- 这套 Prompt 能复用吗?: 能——换受众 / 目标 / voice 即可。