Human-Like Chinese Text Rewriter

Specific Problem

In Chinese content scenarios requiring originality proof, AI-generated text is often flagged by detection tools due to stylized patterns. Typical features include sentence-length variance below 6, consecutive 12+ character sequences, and high-density explicit logical connectors (like "therefore" or "however"). This predictable output pattern not only undermines content credibility but also challenges platforms' originality audits.

Core Method: The Five-Layer Detonation Approach

The skill employs the Five-Layer Detonation Approach to systematically restructure text, aiming to simulate human cognitive traces of "thinking while writing, occasionally digressing." The workflow begins with a preliminary CT diagnosis, quantitatively analyzing the original text to identify high-risk features such as "zombie verbs" (conduct, implement), high-density function words ("的" >8%), or "bureaucratic template zones" (prepositional phrases >30%).

Subsequently, the following five layers of transformation are executed strictly in sequence:

• Layer 1: Global Skeleton Surgery – Reverses argument order (e.g., to "countermeasure dilemma → anomalous phenomenon → causal inquiry"), moves core conclusions to the opening as rhetorical questions, forces paragraph-initial sentences to start from details or challenges, and creates irregular paragraph length ratios (>3:1).
• Layer 2: Syntactic Variation – For at least two out of every three sentences, applies adverbial relocation, subject omission (replacing 30% of explicit subjects with subjectless sentences), infinitive splitting, and zero-anaphora (using different pronouns across three consecutive sentences to refer to the same target).
• Layer 3: Lexical Gene Reassembly – Eliminates zombie verbs (conduct, lead to, produce), replacing them with concrete verbs (e.g., "breed" for "catalyze", "invite" for "cause"); concretizes abstract nouns ("problem" to "lesion"), and implants at least one uncertainty word (e.g., "roughly", "in a sense") every 200 characters.
• Layer 4: Covert Logic Replaces Overt Logic – Forcefully deletes ≥60% of explicit logical words, substituting them with dashes for unexpected consequences, colons to replace causal clauses, or independent exclamatory short sentences for implicit cohesion.
• Layer 5: Cognitive Friction Trace Implantation – Actively adds 1-2 instances of self-negation correction, one redundant analogy, emotive punctuation (e.g., "?!"), and personalized catchphrases (e.g., "simply put") to create human "imperfections" rarely simulated by AI.

Additionally, numerical and citation reconstruction (e.g., changing "2023" to "the past three years") is required, with outputs formatted in specific Markdown: the full rewritten text, transformation log (at least 8 rows), and quantitative self-scores, ensuring global statistical divergence from AI distributions.

Applicable Boundaries and Considerations

This skill specializes in Chinese discourse, with tuning strategies for academic papers, marketing copy, news releases, and platforms like Zhihu or WeChat Official Accounts. Key considerations include:

• Strict adherence to the red-line checklist: Modifying only beginning/end is forbidden; synonym replacement without syntax adjustment (plagiarism-style rewriting) is prohibited; retaining ChatGPT openers (e.g., "in the current context") or forcibly replacing technical terms (e.g., "quantum entanglement" must not become "quantum twist") invalidates the output.
• Model affinity patches are available for different LLMs (Claude, GPT, DeepSeek, etc.) to mitigate specific word biases at the start.
• If initial rewriting still shows high AI risk, second-round (fine-tuning) or third-round (coloring) iterations can be executed by sending specified commands without resending the full text. Ultimately, the goal is to use engineered methods to achieve a "de-AIification" of text at both statistical and semantic levels.