avoid-ai-writing · 一页纸速查表
文本版 · 供搜索与朗读
avoid-ai-writing 一页纸速查表 · 53 类 AI-isms + 45 type + 中文 ZH 规则
avoid-ai-writing · 一页纸速查表
v3.16.0 · 53 类 prose pattern · 45 detector type · 49 issue weight · 中文 ZH 规则 + 工程实践
仓库conorbronsdon/avoid-ai-writing
版本v3.16.0 (2026-07)
规模2448 ⭐ / 226 🍴
LicenseMIT
依赖零
平台10+ Agent
图例
53 类 prose pattern
49 个 issue weight
评分公式
强 corroborator
中文 ZH 规则
CI workflow
发布前 P0 清单
两个 skill 的 diff
6 条反哺行动
参考文献
图例与三组划分
P0 发布事故
P1 结构性 AI 痕迹
P2 风格判断题
Tier 1 词汇(见词即报)
Tier 2 词汇(同段 2+ 才报)
Tier 3 词汇(密度阈值)
三组故意不强制对齐:仓库 README 公布的 53 个 prose pattern(面向人)、detector 暴露的 45 个 issue type(执行体)、SKILL.md 的 ### 章节(包含 writer-side 测试)。差异在 detector/CATEGORIES.md 中维护映射,CI 强制反漂移。
53 类 prose pattern(按 8 组)
1. Content Patterns(7)· 严重度 P0/P1
P0 1. Significance inflation
普通事实外挂"历史意义"尾巴。标志着转折点 / 具有里程碑意义 → 删后事实不受损即气球
P1 2. Notability name-dropping
借名号替代证据。被 NYT/YC/红杉同时看好 → 一个有上下文的来源胜过四个名号
P1 3. Superficial -ing analyses
尾链 -ing 装深刻。从而彰显 / 体现着 / 折射出 → 用具体事实代替
P1 4. Promotional language
推销性语言。nestled within / vibrant / thriving / breathtaking → 说事实
P0 5. Vague attributions
无来源归因。专家认为 / 研究显示 / 业内普遍认为 → 补作者/报告/日期
P1 6. Formulaic challenges
套路化挑战。Despite challenges... 持续繁荣 → 点名挑战与应对
P1 7. Novelty inflation
新奇性注水。没人谈 / 首次提出 / 大家都忽略 → 用现象级证据
2. Language Patterns(7)· 严重度 P1/P2
P1 8. Word replacements (3 tiers)
Tier 1 见词即报(leverage→use、commence→start、robust→strong、seamless→smooth、delve→explore...)
Tier 2 同段 2+(foster、navigate、empower、ecosystem...)
Tier 3 密度阈值(innovative、scalable、significant...)
P2 9. Copula avoidance
系动词回避。serves as / features / boasts → is / has
P1 10. Synonym cycling
同义词轮换。开发者/工程师/实践者/建设者 → 重复最准确的那个
P1 11. Template phrases
模板短语。a [adj] step towards [adj] infrastructure → 说具体结果
P1 12. Filler phrases
填充短语。In order to / Due to the fact that → to / because
P1 13. False ranges
虚假范围。从 Big Bang 到暗物质 → 列实际主题
P1 14. Parenthetical hedging
括号式对冲。工具(像 X 和 Y) → 直接列出或删
3. Structure Patterns(9)· 严重度 P1/P2
P1 15. Formatting
破折号 / 粗体过用 / emoji 标题 / bullet 重。em-dash ≥ 1/1000 words 报
P1 16. Sentence structure
It's not X — it's Y + 三段式 + 多重否定 → 直接肯定句
P2 17. Structural issues
段落均匀 / 句长均一 / 完美语法 → 变长 + 碎句 + 疑问
P2 18. Transition phrases
Moreover / Furthermore / In today's [X] → and / also 或重组
P1 19. Inline-header lists
**速度:** 速度提升了… → 直接写观点
P1 20. Title case headings
中文不适用(英文 Title Case)。
P1 21. Numbered list inflation
7 个理由 / 5 个建议 → 砍到 2-3 个真正成立的
P1 22. False concession
假装权衡。虽然 X 有局限，但仍然卓越 → 真权衡
P1 23. Rhetorical question openers
修辞性问句开场。如果有一种更好的方法呢? → 直接抛主张
4. Communication Patterns(9)· 严重度 P0/P1
P0 24. Chatbot artifacts
Certainly! / Hope this helps! / Let me know if… → 整段删除
P1 25. "Let's" constructions
Let's explore / Let's break this down → 直接开始
P0 26. Cutoff disclaimers
截至我的知识更新 / 基于有限资料 / 无法访问实时数据 → 查证后重写
P2 27. Generic conclusions
未来可期 / Only time will tell / 拭目以待 → 具体收尾或删
P2 28. Emotional flatline
What surprised me most / I was fascinated to discover → 写出触发情绪的事实
P1 29. Reasoning chain artifacts
Let me think step by step / Breaking this down → 先结论,后证据
P1 30. Sycophantic tone
Great question! / You're absolutely right! → 删
P1 31. Acknowledgment loops
You're asking about… / To answer your question… → 直接答
P2 32. Confidence calibration
It's worth noting / Interestingly / Surprisingly → 让事实说话
5. Meta Patterns(4)· 严重度 P2
P2 33. Excessive structure
200 字 5 个 header → 合并,改用具体标题
P2 34. Rhythm and uniformity
句长 15-25 全一致 → 短长交错 + 碎句 + 问句
P2 35. Over-polishing
每处不规则都被砂平 → 保留自然不流利
P1 36. Rewrite-vs-patch threshold
5+ 词汇 + 3+ 模式 + uniform rhythm 三项同时命中 → 停止 patch,整篇重写
6. Structural Detection v3.4(6)· 严重度 P1
P1 37. Tier 3 phrases (multi-word boilerplate)
同 phrase ≥2 → tier3-phrase;≥3 distinct → tier3-phrase-cluster(权重 12)。the integration of / decentralized compute / community-driven / long-term sustainability
P1 38. Future-narrative closers
may become one of the most important narratives of the next market cycle → 改成可证伪预测
P1 39. Hedge-stacked predictions
could potentially create / may eventually unlock → 留一个 hedge,删其余
P1 40. "Real/actual" adjective inflation
real on-chain tokenomics / actual reward sustainability → 删空心加强词,加具体主张(carve-out: 显式对比时合法)
P0 41. Hashtag stuffing
6+ 个 hashtag 几乎必为 LLM 社交输出 → 限 2-3 个 specific tag 或删
P1 42. Bare noun-phrase bullets
5+ 短 NP 列表无动词无数值 → 改写为完整 claim(carve-out: 真实列表内容合法)
7. AI-tool fingerprints v3.5-3.8(9)· 严重度 P0/P1
P0 43. Unfilled placeholders
[Your Name] / [INSERT SOURCE] / 2025-XX-XX → 发布事故,必删
P0 44. Chatbot citation markup
citeturn0search0 / oai_citation / contentReference[oaicite:0] → smoking gun,必删
P0 45. AI-tool URL parameters
utm_source=chatgpt.com / utm_source=copilot.com → 删跟踪参数
P0 46. Speculative gap-filling
maintains a low profile / likely began his career → 删猜测或补来源
P1 47. Hyphenated-pair overuse
high-quality, well-architected, future-proof solution → 砍到真正起作用的修饰
P1 48. Infomercial engagement hooks
The catch? / The kicker? / 你猜怎么着 → 删钩子,直说事实
P1 49. Low vocabulary diversity (TTR)
≥200 tokens 且 TTR