Help users remove traces of AI-generated text so the content reads more like something a real person wrote.
Use Cases
- The user says "help me get rid of the AI flavor," "make this passage more natural," or "polish it so no one can tell it was written by AI."
- The user wrote an article/email/report with AI and needs a final polish before sending it out.
- The user feels a certain passage is "too AI-like" and wants to make it sound more human.
Execution Method
Use LLM capabilities directly; no additional tools are needed.
Core Principles
- Break AI patterns: Avoid typical AI sentence patterns such as "first... second... finally...," "in conclusion," and "it is worth noting that."
- Add a sense of imperfection: Appropriately use colloquial expressions, irregular sentence structures, and personalized tone particles.
- Inject authenticity: Add specific details, personal perspectives, and emotional color.
- Maintain a natural rhythm: Alternate long and short sentences; avoid every paragraph having a neat three-sentence structure.
Common AI Traces and How to Handle Them
| AI Trace | Handling Method |
|---|---|
| "First... second... finally..." | Change to natural transitions, without numbering |
| "It is worth noting that," "it should be emphasized that" | Delete or replace with colloquial expressions |
| "In today's society," "with the development of..." | Get straight to the point |
| Every paragraph has the same length and symmetrical structure | Break the symmetry; mix long and short paragraphs |
| Overuse of conjunctions | Reduce conjunctions and connect directly with short sentences |
| Lists always have three points | It can be two, four, or no list at all |
Style Levels
- Light de-AI-ification: Keep the structure; only replace typical AI wording
- Moderate de-AI-ification(default): Adjust sentence structures and add colloquial expressions
- Heavy de-AI-ification: Rewrite substantially and inject a strong personal style
Output Specifications
- By default, perform "moderate de-AI-ification"
- Show a before-and-after comparison (mark key changes with strikethrough)
- If the user has a remembered writing style preference, polish according to the preferred style
- Keep the core information of the original text unchanged; only adjust the way it is expressed