AI Flavor Proofreading

When to Use This Skill

• User says "proofread this", "reduce AI flavor", "too AI-like"
• After the first draft is completed
• User mentions "high AI detection rate", "lack of human touch"
• The article needs final polish

Core Objectives

• Reduce AI Detection Rate: Target to keep it below 30%
• Add Human Touch: Make the article experienced, warm, and sound like a real person speaking
• Maintain Peanut Style: Practice-oriented, technically in-depth, easy to understand, practical and actionable

Three-Pass Proofreading Process

Pass 1: Content Proofreading (Logic, Facts, Structure)

• Are facts accurate? (Data, time, product names)
• Is logic clear? (No contradictions before and after)
• Is structure reasonable? (No off-topic content)
• Is information complete? (All key points are covered)
• No fabrication? (All data and cases are real)

• Read through the entire text and mark questionable points
• Search to verify uncertain information
• Check the accuracy of numbers and proper nouns

Pass 2: Style Proofreading (AI Flavor Reduction)

1️⃣ Clichéd Phrases

• "In today's era", "Against the backdrop of..."
• "In summary", "All in all"
• "It is worth noting that", "It should be emphasized that"
• "With the development of...", "Along with..."

❌ "In today's era of rapid AI technology development, programming tools are constantly evolving..."
✅ "Claude Code is out. I've used it for two weeks, and it's definitely better than Cursor."

2️⃣ AI Sentence Patterns (High-Frequency Repetitive Templates)

• "Not...but..." (used more than 3 times consecutively)
• "Not only...but also..." (stacked use)
• "On one hand...on the other hand..."
• "Both...and..." (overused)

❌ "Claude Code is not just a simple tool, but an intelligent assistant. It can not only write code, but also understand intentions."
✅ "Claude Code is more than just code completion. It understands your intentions and knows what you want to do. That's the key."

3️⃣ Formal Vocabulary (Strong Sense of Distance)

• "Significantly improve", "Greatly enhance" → Use specific numbers
• "Make full use of" → "Make good use of"
• "Perform operations" → Use verbs directly ("click", "input")
• "Realize functions" → "Achieve"
• "Promote development" → "Help..."

❌ "By making full use of Claude Code's capabilities, you can significantly improve development efficiency..."
✅ "Make good use of Claude Code, and your development speed can be much faster. From my tests, it saves 3-5 hours per project on average."

4️⃣ Mechanical Structure (AI's Obsession with Lists)

• Overuse of "First, Second, Finally"
• Dense meaningless subheadings (one every 200 words)
• Insisting on using lists when prose can clearly explain
• "The following points", "Mainly reflected in the following aspects"

❌ "The advantages of Claude Code are mainly reflected in the following aspects: First, it has strong context understanding ability; Second, it supports multiple programming languages; Finally, its response speed is fast."

✅ "What's the strongest part of Claude Code? Context understanding. If you ask it to modify a feature, it knows which files to touch and won't miss anything. It also supports over 20 languages, and the speed is fast."

5️⃣ Neutral Attitude (No Stance or Opinion)

• Always objectively balanced: "Has both advantages and disadvantages"
• Afraid to make judgments: "Depends on the actual situation"
• Adds qualifiers to every sentence: "May", "Perhaps", "In some cases"

❌ "Claude Code may perform well in some scenarios, but it also has some limitations. The specific effect depends on the actual usage situation."

✅ "Claude Code is really good. It's much more stable than Cursor when handling large projects. But it's expensive, and individual developers may not afford it."

6️⃣ Lack of Details (Abstract and Not Actionable)

• Many big words, few specific details
• "Many", "Some", "Significant", "A large number" (no numbers provided)
• Lack of real feelings and personal experiences
• No timeline, no scenarios, no conflicts

❌ "After using it for a period of time, I found that Claude Code can effectively improve development efficiency."

✅ "I've used it for two weeks. The most obvious change is that a project that used to take 3 days can now be done in 1.5 days. Especially when refactoring old code, Claude Code can find all related files at once, so I don't have to go through them one by one."

Pass 2: Specific Methods

1. Check paragraph by paragraph, mark AI-flavored expressions
2. Rewrite against the 6 categories of AI jargon list
3. Add personal experiences and real details (check personal material library)
4. Add uncertain expressions (moderately use "I think", "may", "perhaps")
5. Check if it sounds like Peanut speaking

• Split long sentences into short ones (split sentences over 30 words into 2-3 sentences)
• Change abstract to specific ("many" → "3", "significantly improve" → "save 3-5 hours")
• Change formal to colloquial ("perform operations" → "click")
• Add real feelings ("When I used it before...", "To be honest...")

Pass 3: Detail Polishing (Punctuation, Formatting, Rhythm)

• Are sentence lengths appropriate? (15-25 words mainly, no more than 30 words)
• Are paragraphs not too long? (3-5 lines on mobile screen)
• Is punctuation natural? (Use more periods, fewer commas to connect long sentences)
• Is there rhythm variation? (Combine fast and slow pacing)
• Are subheadings not too dense? (One every 300-500 words)
• Are lists used moderately? (Use prose instead of lists when possible)
• Is bolding used moderately? (1-2 times every 200-300 words)

• Read aloud to feel the rhythm
• Identify long sentences over 30 words and split them
• Check paragraph length, split overly long paragraphs
• Adjust formatting, remove excessive formatting

Final Check

• Read through the entire text, rewrite awkward parts
• It sounds like Peanut himself speaking
• (Optional) AI detection rate is below 30%

Peanut Style Characteristics (Must Maintain)

• Practice-oriented: With specific cases and executable steps
• Technical depth + easy to understand: No jargon, but in-depth
• Practical and actionable: Readers can take immediate action after reading
• Strong human touch: Experienced, warm, avoid AI jargon
• Real data: All tests and cases must be real
• Moderate uncertainty: Say "I think" when appropriate, don't pretend to know everything

Personal Material Library Usage

• ✅ The article needs real case support
• ✅ Needs real evaluation of tools/products
• ✅ Needs personal experience as an opening
• ✅ The article is too "AI-flavored" and needs to add human touch

Method A: Directly Search Original Data (Recommended Priority)

/写作参考/个人素材库/全部即刻动态.csv

pattern: "keyword1|keyword2"
path: /Users/alchain/Documents/写作/写作参考/个人素材库/全部即刻动态.csv
output_mode: content
-n: true

• Writing about Amap Street Sweeping List → Search "Amap|Street Sweeping List"
• Writing about AI programming tools → Search "Cursor|Claude Code|Copilot"
• Writing about product growth → Search "Growth|Conversion|Users"

Method B: View Refined Materials

1. Open the general index:

   /写作参考/个人素材库/README.md

2. Find the theme: Open

   _即刻动态主题索引.md

3. Select materials: View refined files (e.g.,

   AI编程工具使用心得.md

   )
4. Rewrite and use: Expand into long-text logic, maintain real attitude

• All materials are real, cannot be fabricated or exaggerated
• Pay attention to the time span (2025.05-2025.10), avoid outdated information
• Must rewrite into long-text logic, cannot directly copy and paste

Proofreading Tools (Optional)

Automatic Detection Tools

• Keyword Detection: Search for clichés like "In today's", "In summary", "It is worth noting that"
• Sentence Length Check: Mark sentences over 30 words
• AI Detection: Use tools like GPTZero, Originality.AI (for reference only)

Manual Detection Methods

• Read aloud: Awkward parts need revision
• Compare with historical articles: Check if the style is consistent with Peanut's previous excellent articles
• Ask yourself: Would Peanut say this?

Common Mistakes and Rewriting Examples

Mistake 1: Circumlocutory Opening

❌ "Against the backdrop of rapid AI technology development, the programming tool field is also ushering in new changes. As an emerging AI programming assistant, Claude Code is receiving increasing attention from developers..."

✅ "Claude Code is out. I've used it for two weeks, and it's definitely better than Cursor. Let's talk about why today."

Mistake 2: Overuse of AI Sentence Patterns

❌ "Claude Code is not just a simple code completion tool, but an intelligent programming assistant. It can not only generate code, but also understand developers' intentions."

✅ "Claude Code is more than just code completion. It understands what you want to do. For example, if you say 'refactor this module', it knows how to do it."

Mistake 3: Lack of Specific Details

❌ "After a period of testing, I found that Claude Code performs well in many aspects."

✅ "I've used it for two weeks. The most obvious change is that a project that used to take 3 days can now be done in 1.5 days."

Mistake 4: Overly Neutral Attitude

❌ "Claude Code has its advantages and also some shortcomings. Whether it is suitable for use depends on the actual situation."

✅ "Claude Code is really good. But it's expensive, and individual developers may not afford it. If your company reimburses it, highly recommended."

Reference Resources

• /写作参考/AI味审校指南.md

  - Complete proofreading guide (includes more examples)

• /写作参考/个人素材库/README.md

  - Personal material library general index

• /写作参考/个人素材库/全部即刻动态.csv

  - Original material data

• /写作参考/个人素材库/_即刻动态使用规范.md

  - Material usage specifications

Success Cases

• 《Why the Best Content is Useless》: After three-pass proofreading, AI detection rate dropped to 28%
• 《A Team Burning Billions of Tokens a Month: How They Use Claude Code》: Added a large number of real cases, strong human touch