Avoid AI-Generated Tone

This skill loads the appropriate language guide based on the language of the content being written or edited.

Language Routing

Detect the target language from:

1. The language of the copy being edited or generated
2. The language the user is writing in
3. An explicit instruction ("write in English / 中文で / 日本語で")

Then load the corresponding instruction file:

If the copy mixes languages (e.g., English product name in Chinese copy), apply the primary-language guide and spot-check vocabulary from the secondary language guide.

How to Apply This Skill

1. Read the instruction file for the detected language
2. Scan the copy against that language's vocabulary blacklist and structural patterns
3. Rewrite flagged sections — produce the actual rewritten copy, not just annotations
4. Apply the rewrite checklist from the instruction file
5. Self-test using the three questions at the end of the instruction file

Output format: Always produce the rewritten version. Show the original and rewrite side-by-side only when explaining specific moves. Never return a list of annotations without the fixed copy.

What Each Guide Covers

All three language guides share the same structure:

• Vocabulary blacklist (words/phrases that signal AI generation)
• Sentence structure problems (rhythm, parallelism, list addiction)
• Tone problems (hedging, no stance, emotional flatness)
• Formatting problems (bold-first bullets, em-dash overuse, fractal summaries)
• Content problems (vague without specifics, one-point dilution, fake attribution)
• Rewrite checklist (step-by-step scan)
• Before/after rewrite examples
• Three self-test questions

Core Principle (All Languages)

The problem isn't a list of bad words. It's a posture: AI writing covers everything, commits to nothing, varies nothing. Good writing makes cuts, takes sides, and sounds like a specific person said it.

> Chinese: AI文章像预制菜——成分齐全、摆盘漂亮、没有锅气。

> English: AI writing is the verbal equivalent of stock photography.

> Japanese: どこを読んでも同じ温度で、感情の起伏がない。