Chinese Anti-AI Rewrite with 58 Quality Checks

The Problem: "Uncanny Valley" of AI-Generated Chinese Text

In technical documentation, business reports, and content creation, AI-generated Chinese text often carries a recognizable "machine flavor." This isn't about grammar errors, but a pattern of "correctness" devoid of personal perspective. Symptoms include:
- Formulaic Phrases: Crutch openers and connectors like "值得注意的是" (It is worth noting that) or "在当今...背景下" (In today's context of...) that add no information.
- Structural Repetition: Uniform paragraph structures, such as points consistently listed with "首先/其次/最后" (First/Second/Finally) or overuse of binary contrasts like "不是A，而是B" (Not A, but B).
- Inflated Vocabulary: Prolific use of abstract terms like "赋能" (empower), "抓手" (leverage), "底层逻辑" (underlying logic), or hyperbolic modifiers like "革命性" (revolutionary), "颠覆性" (disruptive).
- Lack of Humanity: Text that feels unnaturally calm and objective, lacking the tonal rhythm, opinionated reactions, and acknowledgment of complexity expected from a real author.

The Solution: Systematic Detection and Humanizing Rewrite

中文去AI味·58项质检 is not a simple thesaurus tool. It is a systematic "de-mechanization" process for text. Based on analyzing patterns from massive AI outputs, it offers structured detection and deep rewriting.

Core Capabilities

1. A Library of 58 Trace Patterns: It deconstructs AI flavor into 58 enumerable patterns across eight categories (Filler Phrases, Structural Templates, Lexical Alerts, Formatting Abuse, Narrative Tone, Business Document Specifics, Conversational Residue, Register Appropriateness). For example, P1 marks formulaic openers, W1 marks high-frequency AI buzzwords, and S3 marks the classic three-point list.
2. A Five-Dimension Scoring System: After rewriting, text is scored (1-10 per item) on Directness (getting to the point), Rhythm (variation in sentence length), Trust (respecting the reader's intelligence), Authenticity (sounding like a real person), and Concision (information density). The total score is 50; text scoring below 35 is revised further.

Key Process Steps

The workflow forms a closed loop to ensure quality:
- Detection and Tagging: The text is read through, and all instances matching the 58 patterns are flagged.
- Rule-Based Rewrite: Five core rules are applied for initial revision: cut filler phrases, break formulaic structures, vary rhythm, trust the reader, and delete fake "golden sentences."
- Injecting Soul: After removing traces, the text is actively humanized using six techniques (e.g., having an opinion, acknowledging complexity, using "I" appropriately), preventing it from becoming bland.
- Anti-AI Review Loop: This is the crucial step. The rewritten text is re-examined to simulate a reader's perspective, scanning for any newly introduced AI patterns, unnatural uniformity in tone, or other residual issues, followed by a second pass of revision.
- Checklist & Scoring: A final verification is done using a checklist of 20+ items (e.g., "Are the lengths of three consecutive sentences identical?", "Is the ending a forced elevation?"), followed by the five-dimension scoring.

Scope and Considerations

This skill is designed for scenarios requiring highly credible, naturally flowing formal Chinese text, including but not limited to:
- Strategic reports, consulting proposals, business plans
- Research summaries, literature reviews, academic papers
- Social media articles, product introductions, brand copy
- Feasibility studies, tender documents, engineering reports

Boundaries and Notes:
1. Goal is "Modern Standard Vernacular": For formal contexts, the rewrite direction is towards written, standardized language, not colloquialism. It also addresses issues like informal words needing formalization (R1) or excessive classical style (R7).
2. Preserving Factual Claims: The skill only rewrites style, not facts. It flags placeholders like [client name] or XX to be filled for user attention.
3. De-flavoring, Not De-formalizing: It explicitly distinguishes between AI flavor and professional, formal style. Perfect grammar and formal vocabulary alone are not grounds for flagging; the judgment is based on multiple patterns clustering.
4. Iteration Based on Feedback: If the five-dimension score is below the threshold, further modification is triggered to ensure output quality standards are met.