De AI Polish

Quick Start

Detect AI-generated expressions in articles:

bash

# Full-text scan and detection
/de-ai-polish detect @article.md

# Direct modification
/de-ai-polish fix @article.md

Core Detection Rules

I. Absolute Prohibitions (Must Remove)

The following patterns should be removed in all cases, as they are the most obvious AI traces:

1. Contrast Structures

Patterns to remove:

```
不是.*而是
```
(Not...but...)
```
并非.*而是
```
(Not...but...)
```
不在于.*在于
```
(It's not about...but about...)
```
不再是.*而是
```
(No longer...but...)
```
与.*是两回事
```
(X is not the same as Y)
```
与.*不等同于
```
(X is not equivalent to Y)
```
看似.*实则
```
(Seems like...but actually... e.g., Seems like a metaphor, but actually talks about production methods)
```
看起来.*其实
```
(Looks like...but actually... e.g., Looks like X, but it's actually Y)

Problem Analysis:

Uses contrast structures to create strong反差, revealing AI traces
Forces concepts into unnatural comparisons
Should be changed to direct statements or progressive relationships

Rewriting Methods:

Split into two independent declarative sentences
Or change to natural relationships like progression or parallelism
Avoid forced contrasts for emphasis

Synonym Repetition (New):

Detection Pattern: Using multiple synonyms to refer to the same thing within a short distance

Examples:

```
主人公…主要角色…中心人物…英雄
```
(Protagonist...main character...central figure...hero - referring to the same person in the same paragraph)
```
系统…平台…工具…解决方案
```
(System...platform...tool...solution - frequent word changes referring to the same thing)

Problem Analysis: AI has repetition penalty code, leading to overuse of synonym replacement, which seems deliberate instead of natural.

Rewriting Method: Use the most accurate term consistently, and reuse the same term when context is clear.

2. Stylized Conjunctions

Already prohibited in writing rules:

```
首先
```
,
```
其次
```
,
```
此外
```
,
```
但是
```
,
```
因此
```
,
```
然而
```
(First, second, in addition, but, therefore, however)
```
一方面
```
,
```
另一方面
```
(On one hand, on the other hand)
```
综上所述
```
,
```
总而言之
```
(In summary, to conclude)

AI Vocabulary Bank (New):

```
深入探讨
```
,
```
深入分析
```
(In-depth discussion, in-depth analysis - overused)
```
强调
```
,
```
凸显
```
,
```
彰显
```
(Emphasize, highlight, demonstrate - over-repeated)
```
持久的
```
,
```
持久的证明
```
(Durable, durable proof)
```
增强
```
,
```
增强其
```
(Enhance, enhance its)
```
培养
```
,
```
培养/促进
```
,
```
促进
```
(Cultivate, cultivate/promote, promote)
```
获得
```
,
```
获得.*的认可
```
(Gain, gain recognition for)
```
相互作用
```
,
```
复杂的相互作用
```
(Interaction, complex interaction)
```
复杂/复杂性
```
,
```
错综复杂的
```
(Complex/complexity, intricate)
```
格局
```
,
```
整体格局
```
,
```
格局的演变
```
(Pattern, overall pattern, evolution of pattern - overuse of abstract nouns)
```
织锦
```
,
```
社会织锦
```
(Tapestry, social tapestry - overuse of abstract nouns)
```
宝贵的
```
,
```
宝贵的经验
```
(Valuable, valuable experience)
```
充满活力的
```
,
```
充满活力的社区
```
(Vibrant, vibrant community)

Superficial Analysis Phrases (New):

```
突出/强调/彰显…，确保…
```
(Highlight/emphasize/demonstrate..., ensuring...)
```
反映/象征…，为…做出贡献
```
(Reflect/symbolize..., contributing to...)
```
培养/促进…，涵盖…
```
(Cultivate/promote..., covering...)
```
展示…，体现…
```
(Show..., embody...)

Problem Analysis: AI adds these superficial analysis phrases at the end of sentences to create false depth, but they have no actual concrete content.

3. Concluding Gesture Sentences

Expressions to detect:

```
方向已经明确
```
(The direction is clear)
```
未来可期
```
(The future is promising)
```
拭目以待
```
(Wait and see)
```
这只是开始
```
(This is just the beginning)
```
让我们期待
```
(Let us look forward to)

Formulaic Outlook (New):

Detection Keywords:

```
尽管其…面临若干挑战…
```
(Despite facing several challenges...)
```
尽管存在这些挑战
```
,
```
尽管面临挑战
```
(Despite these challenges, despite facing challenges)
```
挑战与遗产
```
,
```
机遇与挑战并存
```
(Challenges and legacies, opportunities and challenges coexist)
```
未来展望
```
,
```
下一步计划
```
(Future outlook, next steps)
```
我们将继续关注
```
,
```
值得关注
```
(We will continue to pay attention, worthy of attention)

Problem Analysis: Many AI-generated articles include formulaic "challenges" and "future outlook" sections, which look identical and lack originality.

Rewriting Method: Remove empty outlooks, end with specific facts or plans, or conclude directly.

4. Colloquial and Casual Expressions

Vocabulary to detect:

```
稳
```
,
```
挺稳
```
,
```
很稳
```
,
```
靠谱
```
(Steady, very steady, reliable)
```
飘
```
,
```
摇摆
```
,
```
不稳
```
(Unsteady, wavering, unstable)
```
搞定
```
,
```
没问题
```
(Get it done, no problem)
```
迟早得
```
,
```
肯定
```
,
```
必须得
```
(Sooner or later, definitely, must)
Overuse of absolute terms:
```
肯定
```
,
```
绝对
```
,
```
一定
```
(Definitely, absolutely, must)
```
跑偏
```
,
```
很顺
```
,
```
挺顺
```
(Deviate, very smooth, quite smooth - overly colloquial state descriptions)

Casual Expression Patterns:

```
写长了
```
,
```
做多了
```
,
```
搞定了
```
(Wrote too long, did too much, got it done - overly casual phrases)
```
迟早得拆
```
,
```
肯定会
```
,
```
必须得
```
(Sooner or later it needs to be split, definitely will, must - colloquial expressions)
```
接下来要盯的东西
```
,
```
接下来要关注的
```
(Things to focus on next, what to pay attention to next - colloquial expressions)
Using short colloquial phrases instead of professional explanations

Rewriting Directions:

Use more formal and specific expressions
"迟早得拆" → "Need to consider splitting" or "Should be split"
"肯定行" → "Usually feasible" or "Generally effective"
"跑偏" → "Deviate from expectations", "Have deviations"
"很顺" → "Smooth", "Fluid"
"接下来要盯的东西" → "What we need to focus on next is"
Maintain professionalism, avoid overly casual language

Copula Avoidance (New):

Detection Keywords:

```
作为/代表/标志着/充当 [一个]
```
(Act as/represent/signify/serve as [a])
```
拥有/设有/提供 [一个]
```
(Have/equipped with/provide [a])
```
位于.*的中心
```
,
```
坐落于
```
(Located at the center of, situated in)

Problem Analysis: AI uses complex structures to replace simple "is/have", making the language stiff and unnatural.

Rewriting Method:

"作为X的Y" → "X is Y" or "This Y is X"
"设有Z个W" → "Has Z Ws"
"坐落于" → "Located in" or "In"

Ingratiating Tone (New):

Detection Patterns:

```
好问题！
```
,
```
您说得完全正确！
```
(Great question! You are absolutely right!)

这就是一个很好的观点

您提到的…是相关的

(That's a great point, what you mentioned is relevant)

Overly positive, ingratiating opening remarks

Problem Analysis: The ingratiating/obsequious tone of chatbots is pasted into formal text.

Rewriting Method: State the core content directly, remove ingratiating expressions.

5. Absolute, Philosophical and Dramatic Expressions

Expressions to detect:

```
并不是一个绝对命题
```
(It is not an absolute proposition)
```
本质上
```
,
```
从根本上说
```
,
```
从本质上讲
```
(Essentially, fundamentally, in essence)
```
必然
```
,
```
无疑
```
,
```
绝对
```
,
```
必须
```
(Inevitably, undoubtedly, absolutely, must)
```
真正的
```
,
```
核心的
```
,
```
关键的
```
(True, core, key - overused)
```
推到底
```
,
```
刺痛感
```
(Push to the extreme, sense of sting)
```
这词听着玄
```
,
```
道理其实很朴素
```
(This term sounds mysterious, but the logic is actually simple)
```
天然
```
,
```
天生
```
,
```
自然而然
```
(Naturally, inherently, spontaneously - absolute natural expressions)
```
立刻
```
,
```
马上
```
,
```
立即
```
(Immediately, right away, instantly - overly absolute time expressions)
```
其实
```
(Actually - overused, seems artificial)
```
其实已经能看清
```
,
```
其实很简单
```
(Actually can already see clearly, actually very simple - expressions with "actually")
```
很现实
```
(Very realistic - AI-style absolute judgment)

Overemphasis on Significance (New):

```
作为.*的体现/证明/提醒
```
,
```
充当.*的证明
```
(As an embodiment/proof/reminder of, serve as proof of)
```
标志着.*关键时刻
```
,
```
见证了.*历史
```
(Marks a critical moment, witnesses history)
```
是.*不可或缺的组成部分
```
,
```
是.*的重要体现
```
(Is an indispensable part of, is an important embodiment of)
```
凸显/强调/彰显了其重要性/意义
```
(Highlights/emphasizes/demonstrates its importance/significance)
```
反映了更广泛的
```
,
```
象征着其持续的/永恒的/持久的
```
(Reflects the broader, symbolizes its continuous/eternal/durable)
```
为.*做出贡献
```
,
```
为.*奠定基础
```
(Contributes to, lays the foundation for)
```
关键转折点
```
,
```
不断演变的格局
```
(Critical turning point, evolving pattern)

False Scope (New):

```
从.*到.*
```
(From...to... - check if it's on a meaningful scale)
Example: "从大爆炸到暗物质" (From the Big Bang to dark matter) - meaningless wide scope
Problem: Forces "from X to Y" structure where X and Y are not on the same scale

Problem Analysis:

Overuse of absolute terms makes language stiff
Negative absolute expressions like "并不是..." (It is not...)
Piling up philosophical big words makes content empty
Dramatic expressions are too emotional and lack restraint

Rewriting Method:

Replace absolute expressions with flexible ones
"并不是一个绝对命题" → "There is no simple yes/no answer to this question"
"本质上" → "Usually", "To some extent"
"必然" → "Tend to", "Mostly", "May"
Replace dramatic terms with milder expressions
Keep room for interpretation, don't over-commit to conclusions

6. Rigid Parallel and Symmetric Structures

Structures to detect:

Three or more parallel structures like "既要...又要...还要..." (Not only...but also...and...)
Forced parallel sentence structures
Rigid symmetric expression structures

Specific Examples:

"既要说明'我是谁'，又要说明'我在做什么'，还要说明'我怎么做'" (Not only explain 'who I am', but also 'what I do', and 'how I do it')
"讲清楚口味、讲清楚底线、讲清楚习惯" (Clarify taste, clarify bottom line, clarify habits - three "clarify" parallel phrases)
Overly rigid "一方面...另一方面..." (On one hand...on the other hand...)

Problem Analysis:

Forces parallelism for formal neatness
Seems deliberate and mechanical
Lacks the natural variation of human language

Rewriting Method:

Break rigid structures, use more natural expressions
"既要...又要...还要..." → Change to declarative sentences or mixed structures
Avoid three or more identical parallel structures
Let language have natural ups and downs

7. AI Transition Phrases

Transition phrases to detect:

```
先把.*摆出来
```
,
```
摆上台面
```
(First lay out..., put on the table)
```
不妨把.*拆成
```
,
```
不妨
```
(Might as well split..., might as well)
```
这两个问题一问
```
,
```
答案就清晰了
```
(Ask these two questions, and the answer becomes clear)
```
这话对了一半
```
,
```
这话没错
```
(This is half right, this is correct)
```
我后来发现
```
,
```
我也想过
```
(I later found out, I also thought about it)

一个直接的原因

一个现实的原因

一个根本的原因

(A direct reason, a practical reason, a fundamental reason)

一个很现实的原因是

一个很直接的原因是

(A very practical reason is, a very direct reason is)

```
原因很简单
```
,
```
原因很直接
```
(The reason is simple, the reason is direct)

Collaboration Communication Traces (New):

Detection Keywords:

```
希望这对您有帮助
```
,
```
希望这能帮到您
```
(Hope this helps you, hope this can assist you)
```
当然！
```
,
```
一定！
```
,
```
没问题！
```
(Of course! Definitely! No problem!)
```
您说得完全正确！
```
,
```
您说得对！
```
(You are absolutely right! You are right!)
```
您想要…
```
,
```
您是否需要…
```
(You want..., do you need...)
```
请告诉我
```
,
```
您可以问我
```
(Please tell me, you can ask me)
```
这是…的概述
```
,
```
以上是…
```
(This is an overview of, the above is...)

Problem Analysis: Text from chatbot conversations is pasted as formal content; these are polite phrases for chatbots and should not appear in formal articles.

Rewriting Method: Remove all collaboration communication traces, state core content directly.

8. Self-Statements

Expressions to detect:

```
我更愿意
```
,
```
我更倾向于
```
(I prefer, I am more inclined to)
```
我给自己定
```
,
```
我设定了
```
(I set for myself, I have set)
```
我想强调
```
,
```
我想指出
```
(I want to emphasize, I want to point out)
```
我期待
```
,
```
我希望
```
(I look forward to, I hope)

9. Lead-In Tics

Expressions to detect:

```
表面上看
```
(On the surface)
```
更关键的是
```
,
```
更值得写的是
```
,
```
更重要的是
```
(More importantly, more worth writing about, more importantly)
```
从经验看
```
,
```
在实践中
```
(From experience, in practice)
```
这当然成立
```
,
```
毫无疑问
```
(This is certainly true, without a doubt)
```
接下来会
```
,
```
接下来我们
```
(Next will, next we)

Vague Attribution (New):

Detection Keywords:

```
行业报告显示
```
,
```
观察者指出
```
(Industry reports show, observers point out)
```
专家认为
```
,
```
一些批评者认为
```
(Experts believe, some critics believe)
```
多个来源/出版物
```
(Multiple sources/publications - with few actual citations)
```
据.*报道
```
(Reported by... - no specific source)

Problem Analysis: AI attributes views to vague authorities without providing specific sources.

Rewriting Method:

Remove vague attributions, state facts directly
Or provide specific sources: "According to the XX report in XX year"

Knowledge Cutoff Disclaimer (New):

Detection Keywords:

```
截至 [日期]
```
,
```
根据我最后的训练更新
```
(As of [date], based on my last training update)
```
虽然具体细节有限/稀缺…
```
(Although specific details are limited/scare...)
```
基于可用信息…
```
(Based on available information...)
```
在现有资料中…
```
(In existing materials...)

Problem Analysis: AI's knowledge cutoff date disclaimers are retained in the text.

Rewriting Method: State known facts directly, remove disclaimers.

10. Slogan-Style Metaphors

Expressions to detect:

信号, 叙事, 换挡, 战场 (Signal, narrative, shift, battlefield)
组合拳, 闭环, 模板, 生态 (Combination punch, closed loop, template, ecosystem)
把……搬进, 拖进…… (Move...into, drag...into...)

11. Overly Emotional

Vocabulary to detect:

扎心, 硬, 值钱, 最大 (Heart-wrenching, tough, valuable, biggest)
反复翻车 (Repeated failures)

12. Overemphasis on Popularity (New)

Expressions to detect:

```
被.*多家媒体报道
```
,
```
受到.*广泛关注
```
(Reported by multiple media outlets, widely concerned by)
```
独立报道、地方/区域/国家媒体
```
(Independent reports, local/regional/national media)
```
由知名专家撰写
```
,
```
专家团队
```
(Written by well-known experts, expert team)
```
活跃的社交媒体账号
```
(Active social media account)
```
拥有超过.*粉丝
```
,
```
关注者
```
(Has over... followers, followers)

Problem Analysis: AI repeatedly emphasizes popularity claims, usually listing sources without context, making content empty and deliberate.

Rewriting Method:

Remove empty popularity emphasis
Or provide specific, meaningful content: "Stated in the XX interview in 2024..."

II. Flexible Detection (Judge Based on Context)

The following patterns need to be judged based on article style, context, and target audience to decide whether to modify:

13. Unordered Lists

Detection Criteria:

Whether lists are used where paragraphs can express the same content
Whether lists are "piling up opinions" instead of "explaining steps"
Whether list density is too high (multiple consecutive lists)
Whether unordered lists appear more than 2-3 times in the entire article

Frequency Limit:

Unordered lists should appear at most 2-3 times in the entire article
Even if list content is reasonable, consider changing to paragraphs if exceeding this frequency
Prioritize retaining necessary lists like operation steps, checklists
For other content that can be converted to paragraphs, use paragraph expressions as much as possible

Acceptable Scenarios:

Operation step instructions
Need to reflect sequential relationships
Minimal checklists
Core framework or standard enumeration

Inline Title Vertical Lists (New):

Detection Pattern:

```
- **用户体验：** 用户体验通过新界面得到显著改善
```
- User Experience: User experience is significantly improved through the new interface
```
- **性能：** 性能通过优化算法得到增强
```
- Performance: Performance is enhanced through algorithm optimization
```
- **安全性：** 安全性通过端到端加密得到加强
```
- Security: Security is strengthened through end-to-end encryption

Problem Analysis: AI outputs lists where items start with bold titles followed by colons. This is a mechanized format.

Rewriting Method: Change to paragraph expressions, or use normal list formats (without bold titles).

Overuse of Emojis and Bold (New):

Detection Pattern:

Emoji decoration:
```
🚀 **启动阶段：**
```
,
```
💡 **关键洞察：**
```
,
```
✅ **下一步：**
```
(🚀 Initiation Phase:, 💡 Key Insight:, ✅ Next Step:)
Mechanical bold emphasis:
```
**OKRs**、**KPIs**、**BMC**
```
(overused)
Bold list titles

Problem Analysis: AI chatbots often decorate titles or bullet points with emojis and bold, making content flashy and unprofessional.

Rewriting Method: Remove emojis and excessive bold, use concise formatting.

14. Quotation Mark Usage

Detect overuse:

High quotation mark density (large number of quotation marks in the entire article)
Adding quotation marks to ordinary words (not terms or citations)
Using quotation marks to express "irony" or "emphasis"
Half-width English quotation marks
```
"..."
```
Half-width single quotation marks
```
'...'
```
Overuse of Chinese quotation marks
```
「」
```

AI-Specific Quotation Misuse Patterns (Must Detect):

The following are common quotation misuses in AI writing, which usually do not require quotation marks:

1. Descriptive Phrases (Ordinary descriptions, no emphasis needed)

```
"可调用的能力"
```
→ 可调用的能力 (Callable capability)
```
"可配置的选项"
```
→ 可配置的选项 (Configurable options)
```
"纯云端 Skill"
```
→ 纯云端 Skill (Cloud-native Skill)
```
"只能调用"
```
→ 只能调用 (Only callable)
```
"前端输入 + 黑盒执行"
```
→ 前端输入加黑盒执行 (Frontend input + black-box execution)

Problem Analysis: These are ordinary function or state descriptions; adding quotation marks seems like forced emphasis and reveals AI traces.

2. Metaphorical Descriptions (Metaphors already have rhetorical effects)

```
"Skill 的外壳"
```
→ Skill 的外壳 (Skill's shell)
```
"散乱的经验"
```
→ 散乱的经验 (Scattered experience)
```
"可运行的结构"
```
→ 可运行的结构 (Runnable structure)
```
"内容消费品"
```
→ 内容消费品 (Content consumer goods)

Problem Analysis: Metaphors already have rhetorical effects; adding quotation marks makes it obvious that "I am using a metaphor", which is not natural.

3. Colloquial Expressions (Colloquial language does not need quotation marks to identify)

```
"能不能一键跑起来"
```
→ 能不能一键跑起来 (Can it run with one click?)
```
"逼它对齐"
```
→ 逼它对齐 (Force it to align)
```
"补复杂度"
```
→ 补复杂度 (Add complexity)
```
"偶尔很神"
```
→ 偶尔很神 (Occasionally amazing)

Problem Analysis: Colloquial expressions do not need quotation marks to mark "this is colloquial"; writing directly is more natural.

4. Action or State Descriptions (Action descriptions do not need emphasis)

```
"换一个 Skill"
```
→ 换一个 Skill (Change a Skill)
```
"换一种用法"
```
→ 换一种用法 (Change a usage)
```
"如何调用它"
```
→ 如何调用它 (How to call it)
```
"它为什么这么做"
```
→ 它为什么这么做 (Why does it do this)
```
"共同编辑同一个对象"
```
→ 共同编辑同一个对象 (Collaboratively edit the same object)

Problem Analysis: Action or state descriptions do not need quotation mark emphasis; direct statements are more natural.

5. Attribute Enumeration or Feature Description

```
"可组织、可迭代、可协作、可分发"
```
→ 可组织、可迭代、可协作、可分发 (Organizable, iterable, collaborative, distributable)
```
"稳定可解释"
```
→ 稳定可解释 (Stable and interpretable)
```
"调用的便利"
```
→ 调用的便利 (Convenience of calling)
```
"结构的可控"
```
→ 结构的可控 (Controllability of structure)

Problem Analysis: Attribute or feature enumerations do not need quotation marks; these are ordinary adjectives.

6. Question or Query Expressions

```
"Skill 很火"
```
→ Skill 很火 (Skill is popular)
```
"Skill 很强"
```
→ Skill 很强 (Skill is powerful)
```
"工具好不好玩"
```
→ 工具好不好玩 (Is the tool fun?)

Problem Analysis: Questions or evaluation expressions do not need quotation marks to mark "this is someone else's opinion".

7. Value Judgment or Goal Description

```
"组织和个人能不能把能力沉淀下来"
```
→ 组织和个人能不能把能力沉淀下来 (Can organizations and individuals沉淀 their capabilities?)
```
"工具好不好玩"
```
→ 工具好不好玩 (Is the tool fun?)
```
"工作标准"
```
→ 工作标准 (Work standards)

Problem Analysis: Value judgments or goal descriptions are ordinary expressions; no quotation mark emphasis is needed.

Correct Quotation Mark Usage Scenarios:

Quoting Original Text: Quoting the exact words of others
Core Term Definition: Technical terms that appear for the first time and need special marking
- Example:
```
"可运行的上下文配置"
```
  ("Runnable context configuration" - core concept defined by the author)
- Example:
```
"上下文挂载能力"
```
  ("Context mounting capability" - may be a technical term)
Avoid Ambiguity: When not adding quotation marks will cause understanding ambiguity
Special Concept Distinction: When needing to distinguish concept levels specifically

Detection Criteria:

If the meaning is still clear after removing quotation marks, quotation marks are usually not needed
If it is a descriptive use of ordinary words (nouns, verbs, adjectives), quotation marks are not needed
If it is a metaphor, colloquialism, or action description, quotation marks are not needed
Minimize quotation mark density in the entire article
Prioritize retaining: core term definitions, original text citations

15. Evaluative Statements and Overgeneralization

Patterns to detect:

```
把.*讲得很清楚
```
,
```
把.*说得很明白
```
(Explain...very clearly, state...very clearly)
```
更尖
```
,
```
更锐利
```
,
```
更深刻
```
(Sharper, more acute, more profound - evaluative adjectives modifying abstract concepts)
```
更硬的定义
```
,
```
更.*的.*
```
(Harder definition, more... - over-modified adjective structures)
```
被.*卡住
```
,
```
被.*阻挡
```
(Stuck by..., blocked by... - passive voice + overgeneralization)
```
一旦把视角从.*就更容易看清
```
(Once shifting perspective from..., it's easier to see - stylized perspective shift)
```
更具体一点
```
,
```
更准确地说
```
(To be more specific, more precisely - AI transition phrases)

Problem Analysis:

"把...讲得很清楚" is an evaluative statement of external content by AI, which seems deliberate
Evaluative adjectives like "更尖" are AI "scoring" concepts, which is not natural
"被...卡住" is passive voice plus overgeneralization, lacking specificity
Stylized perspective shift expressions reveal AI traces

Rewriting Directions:

"把...讲得很清楚" → State content directly, remove evaluation
"更尖" → Remove evaluative modifiers, or change to more specific descriptions
"被两件事卡住" → "Two things stand in the way" or list directly
"一旦把视角从...就更容易看清" → Remove transition phrases, enter the topic directly
"更具体一点" → "Once this critical point arrives" or state directly

Rewriting Examples:

Original Sentence	Rewritten Sentence
Notion 那篇文章把个人层面的跃迁讲得很清楚	The Notion article uses the "bicycle to car" comparison
组织层面的隐喻更尖	(Delete, enter content directly)
它被两件事卡住	Two things stand in the way
一旦把视角从职业挪到时代与经济，就更容易看清这一点	(Delete, enter content directly)
也可以用一句更硬的定义来收束	Define it in one sentence
更具体一点，它意味着三层变化同时发生	Once this critical point arrives, three layers of changes will occur simultaneously

16. Comparative and Assertive Expressions

Patterns to detect:

```
它更像一种
```
,
```
更像一个.*
```
(It is more like a..., more like a)
```
它其实是
```
,
```
这里其实
```
,
```
其实是第二类
```
(It is actually..., actually here, actually the second type)
```
天然会
```
,
```
天生就
```
,
```
自然而然就
```
(Naturally will, inherently will, spontaneously will)
```
会更清楚
```
,
```
会更.*
```
(Will be clearer, will be more... - overused "will be more" pattern)
```
就.*了
```
,
```
就.*能
```
(Just..., just can... - overly absolute assertions)

Problem Analysis:

"更像" creates unnecessary comparisons, seems deliberate
"其实" seems artificial, like correcting readers' cognition
"天然会", "天生就" are overly absolute
Frequent use of "会更" is repetitive and has AI traces
Common feature of these expressions: "deny first then affirm" or "establish comparison", revealing AI traces

Rewriting Directions:

"它更像一种" → Directly state what it is, no need for "更像"
"它其实是" → State facts directly, remove "其实"
"天然会" → Use flexible expressions like "usually will", "tend to"
"会更清楚" → State effects directly, no need for "会更"
Overall strategy: Remove these prefixes, state core content directly

Workflow

Step 1: Reading Comprehension

First read the entire article completely, understand:

Core viewpoints and argument logic
Article style and expression habits
Contextual context

Step 2: Identify AI-Generated Expressions

Refer to relevant content in the core detection rules above to identify various AI-generated expression patterns.

Step 3: Judge Whether to Modify

Not all matches need modification; judgment criteria:

Whether it really affects the naturalness of expression
Whether it conforms to the overall style of the article
Whether it has a specific rhetorical purpose

Step 4: Restate Sentences

Core Principle

Restate the entire sentence, do not just delete words.

Wrong Way (Leads to Stiff Language)

Original: "它更像一种可维护的模块" (It is more like a maintainable module)
Deletion-based modification: "这是一种可维护的模块" (This is a maintainable module)
Problem: Only deleted "更像", sentence structure remains, still stiff

Correct Way (Restatement)

Original: "它更像一种可维护的模块" (It is more like a maintainable module)
Restated: "这样的结构具有可维护性" / "这种设计便于后续维护" (This structure is maintainable / This design facilitates subsequent maintenance)
Advantage: Completely restructures the sentence, making expression natural and smooth

More Comparison Examples

Original Sentence	Deletion-based Modification (×)	Restated (✓)
它其实是第二类问题	这是第二类问题	This belongs to the second type of problem
这里其实有三种可能	这里有三种可能	There are three possibilities here
天然会形成这种结构	会形成这种结构	This structure usually forms naturally
会更清楚这一点	更清楚这一点	This helps to better understand this point
它更像一种框架	这是一种框架	This fits the characteristics of a framework

Modification Principles

Restate sentence structure, do not simply delete marked words
Understand the core meaning the original sentence wants to express
Reorganize language in a natural way
Maintain logical coherence of context
Avoid using fixed replacement templates
Choose appropriate expressions based on specific context
Maintain language diversity and personality

Step 5: Inject Personal Style (Optional)

Overview

After removing AI-generated expressions, optionally inject personal style features to make the article more "human-like".

Style Balance Principle

Target Ratio: Technical Accuracy 60% / Personalization 40%

Technical accuracy part (60%): Accurate core concepts, clear technical logic, appropriate professional terms, rigorous argumentation
Personalization part (40%): Flexible expressions, colloquial metaphors, display of thinking processes, conversational sentence patterns

Positive Features of "Human-Like" Expression

Based on analysis of real human writing samples, the following features make articles more "human-like":

1. Authentic Opening

AI-style openings (should be removed):

"This article will discuss..."
"With the development of..."
"In recent years... has received widespread attention"
"Against this background..."

Real human opening features:

✅ Start with personal feelings/surprises: "I didn't expect either..."
✅ Start with specific scenarios: "Just send a public account article to the dialog box"
✅ Start with plain judgments: "To be honest, both of these methods are quite cumbersome for everyone to use"
✅ Use dashes to create unexpected turns

Core Principle: Openings should have "authenticity", like chatting with a friend instead of writing a report

2. Moderate Use of Flexible Expressions

Overly absolute (should be removed):

"必然", "绝对", "必须" (Inevitably, absolutely, must)

Flexible expressions (can retain/add):

✅ "可能", "或许", "某种程度" (May, perhaps, to some extent)
✅ "往往", "通常", "一般情况下" (Tend to, usually, generally)
✅ "当然...可能..." (Concession + flexibility)
✅ "可以说" (Prudential judgment)

Purpose:

Keep room for interpretation, do not over-commit to conclusions
Show the exploratory process of thinking
Let readers feel "this is a warm article"

3. Correct Boundary of Colloquialism

Overly casual (should be removed):

❌ "搞定", "没问题", "很稳" (Get it done, no problem, steady)
❌ "跑偏", "挺顺", "肯定行" (Deviate, quite smooth, definitely work)
❌ "迟早得拆", "必须得" (Sooner or later it needs to be split, must)

Measured colloquialism (can retain/add):

✅ "说实话", "也许" (To be honest, maybe)
✅ "在我看来", "通常情况下" (In my opinion, usually)
✅ "往往", "多半", "或许" (Tend to, mostly, perhaps)

Core Principle: Colloquialism should be "measured" - sincere but not casual, plain but not rough

4. Natural Integration of Humor

Humor types that can be retained/added:

✅ Self-deprecating: "After reading, you will have a preliminary understanding of what Skill is... or be more confused"
✅ Unexpected turns: Use dashes to create effects
✅ Counter-intuitive: Break readers' expectations

Humor to avoid:

❌ Showy self-praise
❌ Overuse of internet buzzwords
❌ Forced humor

Core Principle: Humor should be "self-deprecating" rather than "showy"; "unexpected" rather than "forced"

5. Natural Expression of Contrast

AI-style contrast (should be removed):

❌ "并非...而是..." (Not...but...)
❌ "不是...而是..." (Not...but...)
❌ "与...是两回事" (X is not the same as Y)
❌ "与...不等同于" (X is not equivalent to Y)

Natural transition/progression:

✅ "当然...但是..." (Concession + transition)
✅ "Skill 出来之后，这样的问题可以..." (State directly: After Skill came out, such problems can be...)
✅ Split into two independent declarative sentences

Core Principle: Do not force contrasts to create反差, let logic unfold naturally

Personal Style Features

Based on

reference/personal-style-guide.md

, injectable style features include:

Display of Thinking Process
- Flexible expressions: "往往", "某种程度", "或许" (Tend to, to some extent, perhaps)
- Acknowledge limitations: "坦白说，还是存在一定差距" (To be honest, there is still a certain gap)
- Observational summaries: "很多...都经历过类似的" (Many... have experienced similar situations)
Colloquial Expressions
- Life metaphors: "算力只是柴火，你还得有锅" (Computing power is just firewood, you still need a pot)
- Conversational sentence patterns: "很多人会迅速得出结论..." (Many people will quickly draw conclusions...)
- Directly address readers: "你可以..." (You can...)
Sentence Rhythm
- Mix long and short sentences: Long sentences for discussion (technical accuracy) + short sentences for emphasis (personalization)
- Avoid consecutive long or short sentences
- Natural sentence breaks, do not deliberately pursue neatness
Natural Transitions
- Do not use stylized conjunctions (first, second, however)
- Let logic connect naturally

Injection Method

Read Reference Files
- ```
reference/personal-style-guide.md
```
  : Personal style feature guide
- ```
reference/expression-transformations.md
```
  : Expression transformation comparison table
- ```
reference/sentence-rhythm-guide.md
```
  : Sentence rhythm guide
- ```
reference/quality-scoring.md
```
  : Quality scoring system (new)
Identify Entry Points
- Which sentences can add flexible expressions
- Which concepts can use life metaphors
- Where sentence rhythm needs adjustment
Apply Transformations
- Use comparison tables to replace AI-generated expressions
- Add exploratory expressions and colloquial elements
- Adjust sentence rhythm to create natural changes
Check Balance
- Is technical accuracy maintained (60%)?
- Is personalization expression moderate (40%)?
- Is over-stylization avoided?

Notes

Avoid over-stylization: Do not lose professionalism for the sake of colloquialism
Serve the content: Style should serve the core viewpoint to be expressed
Keep authenticity: Let content serve the core viewpoint, adjust appropriately according to specific topics and readers

Continuous Iteration

Personal style definitions will be continuously optimized in actual use:

Adjust reference files based on user feedback after each use
Record effective style features and expression methods
Distinguish between general features and features suitable for specific topics

de-ai-polish

NPX Install

Tags

SKILL.md Content (Chinese)

De AI Polish

Quick Start

Core Detection Rules

I. Absolute Prohibitions (Must Remove)

1. Contrast Structures

2. Stylized Conjunctions

3. Concluding Gesture Sentences

4. Colloquial and Casual Expressions

5. Absolute, Philosophical and Dramatic Expressions

6. Rigid Parallel and Symmetric Structures

7. AI Transition Phrases

8. Self-Statements

9. Lead-In Tics

10. Slogan-Style Metaphors

11. Overly Emotional

12. Overemphasis on Popularity (New)

II. Flexible Detection (Judge Based on Context)

13. Unordered Lists

14. Quotation Mark Usage

1. Descriptive Phrases (Ordinary descriptions, no emphasis needed)

2. Metaphorical Descriptions (Metaphors already have rhetorical effects)

3. Colloquial Expressions (Colloquial language does not need quotation marks to identify)

4. Action or State Descriptions (Action descriptions do not need emphasis)

5. Attribute Enumeration or Feature Description

6. Question or Query Expressions

7. Value Judgment or Goal Description

15. Evaluative Statements and Overgeneralization

16. Comparative and Assertive Expressions

Workflow

Step 1: Reading Comprehension

Step 2: Identify AI-Generated Expressions

Step 3: Judge Whether to Modify

Step 4: Restate Sentences

Core Principle

Wrong Way (Leads to Stiff Language)

Correct Way (Restatement)

More Comparison Examples

Modification Principles

Step 5: Inject Personal Style (Optional)

Overview

Style Balance Principle

Positive Features of "Human-Like" Expression

1. Authentic Opening

2. Moderate Use of Flexible Expressions

3. Correct Boundary of Colloquialism

4. Natural Integration of Humor

5. Natural Expression of Contrast

Personal Style Features

Injection Method

Notes

Continuous Iteration