DeAI Writing Pro · Detection, Scoring & Platform Style Rewrite

Problem

LLM-written prose often sounds smooth, symmetric, and overly objective, but swapping words only replaces one template with another. Engineers and editors need to identify where text sounds AI-generated, adapt tone to the publishing platform, and preserve facts, data, citations, and proper nouns. This skill treats de-AI-ification as a set of inspectable writing issues rather than a vague prompt.

How It Works

It runs a detection → targeted rewrite → re-scoring loop:

• 12-dimension scoring: rates AI-flavored vocabulary, mechanical connectors, over-structuring, false objectivity, uniform sentence length, and weak viewpoints, with a higher total meaning more AI-like and direct quotes for problem sentences.
• Platform tone matching: loads rules for WeChat Official Accounts, Xiaohongshu, Zhihu, academic writing, or general use so articles do not share one identical voice.
• Minimal edits: changes only the problem sentences identified by scoring, preserves the author's stance, avoids adding new facts, and keeps length within ±15%.
• Deliverables: outputs rewritten text, a before/after dimension score table, and the first 10 before-to-after edits; batch mode summarizes file name, before score, after score, and reduction.

Boundaries

It is suitable for technical articles, operational copy, Zhihu answers, WeChat articles, and academic fluency polishing. It is not for faking human originality to evade academic integrity review, nor for political or illegal content. If facts, data, and citations must remain unchanged, treat it as an expression-layer rewrite tool, not a fact-checking tool.