Make Chinese text natural, accurate, and marked by human judgment. Preserve the original information and remove the common formulas of machine generation. This version is based on blader's Humanizer, with rules and example sentences rewritten for the Chinese context.
Core task
After receiving the text:
- Look for the following patterns, but do not judge arbitrarily based on one or two words.
- Preserve the information, not necessarily the original shape. Keep both opinions and facts; paragraphs can be split, merged, or reordered.
- Do not add facts, names, numbers, dates, quotations, cases, or sources that are not in the original text or provided by the user. When specific information is needed, ask first; if asking is not possible, use plain expression and do not invent details.
- Match the purpose and tone. Official documents, papers, contracts, and technical documentation should be accurate and restrained; personal articles and social content can have attitude, hesitation, humor, and an uneven rhythm.
- Give priority to the user's explicit requirements and writing samples provided by the user.
First recover the information swallowed by jargon
"Empower, create, drive, build, assist" are not inherently forbidden words. The problem is that they are often used to seem professional, complete, and positive while hiding what the sentence most needs to explain. When encountering jargon, check each item:
- Who is doing it?
- What specifically was done?
- Who or what does it affect?
- What observable result was produced? What is the basis?
When the original text already provides the answers, put this information back into the sentence. When the original text does not provide the answers, do not invent a realistic-looking case; delete effects that cannot hold up, or directly point out that the original text does not explain them. Do not play the synonym-swapping game; changing "empower" to "assist" or "drive" does not count as revision.
"The original text does not explain this" is usually written in review comments, so do not mechanically insert it into every final draft. File mode and embedded mode should directly delete unsupported effects; only explain in the body when the missing information itself matters to the reader.
Example: "AI empowers business growth" does not explain any action. If the context states "customer service uses AI to classify refund tickets, and average handling time dropped from two days to one day," write that fact; if there is no context, at most write "the company uses AI in its business" and point out that the original text does not explain the use case and provides no evidence of growth.
Voice calibration
When the user provides their own sample text, first observe sentence length, word choice, paragraph openings, punctuation, verbal tics, ways of transitioning, and information density, then rewrite according to these habits. Do not upgrade colloquial language into written language, and do not "correct" intentionally preserved dialect, internet slang, or personal quirks. Stable habits in the sample take priority over this skill's default rules.
Common patterns in Chinese AI writing
1. Vague elevation of significance
Watch for: marks, demonstrates, reflects, is of milestone significance, injects new momentum, opens a new chapter, lays the foundation for..., drives ... to a new level.
Problem: Forcing ordinary facts into a grand narrative.
Before: This office relocation marks the company's development entering a brand-new stage and lays a solid foundation for future high-quality development.
After: The company moved offices. The original text does not state the time, reason, or actual impact of the relocation.
2. Replacing information with fame and media lists
Watch for: Coverage by multiple authoritative media outlets, widespread industry attention, recognition from many big names, heated discussion across the internet.
Problem: It only lists media, awards, or celebrities without explaining what was reported or why it is relevant.
Before: The project received widespread attention from CCTV, Xinhua News Agency, and multiple authoritative media outlets.
After: Media outlets such as CCTV and Xinhua News Agency have reported on the project. The original text does not state what was reported.
3. Stacking fake analysis with "thus/thereby/helping" at the end of sentences
Watch for: Thus, thereby, further, effectively promote, help achieve, continuously empower, ending with a long string of result adverbials.
Problem: It appends effects and significance without evidence after the facts.
Before: The platform added a one-click export feature, thereby effectively improving work efficiency and further empowering team collaboration.
After: The platform added a one-click export feature. The original text does not provide data on efficiency or collaboration effects.
4. Promotional copy and advertising tone
Watch for: Crafted with ingenuity, grandly launched, stunning arrival, immersive, all-around, ultimate, excellent, leading, refreshed upgrade, hidden gem, must-visit.
Problem: It uses adjectives to create excitement but does not provide verifiable information.
Before: This hidden-gem homestay is located in the core area of a picturesque ancient town, bringing visitors an ultimate immersive experience.
After: The homestay is located in the central area of the ancient town. The original text does not provide specific information such as facilities, distances, or guest experiences.
5. Vague attribution and "it is reported"
Watch for: Experts say, industry insiders believe, research shows, relevant data shows, it is reported, it is widely believed.
Problem: It delegates judgment to authorities without names, institutions, or sources.
Before: Experts say that this technology will have a profound impact on the industry.
After: Delete this sentence; if the original text provides the expert's name, exact words, and source, then state them accurately.
6. The "challenges and outlook" template
Watch for: Despite facing many challenges, opportunities and challenges coexist, the future still requires, looking ahead, forging ahead.
Problem: The ending uses a universal paragraph and avoids the real problems and plans.
Before: Although it still faces many challenges, with the joint efforts of all parties, the project has a promising future.
After: The project still faces some difficulties, but the original text does not explain what they specifically are. If this cannot be supplemented, delete this outlook paragraph.
7. Abstract verbs swallowing concrete actions
Watch for: Deep, comprehensive, continuous, precise, efficient, empower, help, create, build, focus on, center on, devote effort to, dimension, scenario, ecosystem, closed loop, lever, underlying logic.
Problem: A single word is not necessarily a problem; when multiple abstract words appear in succession, they obscure the subject, action, object, result, and evidence.
Before: Centering on core user needs, continuously build a full-scenario service ecosystem and precisely empower business growth.
After: Services will be improved according to user needs. The original text does not say who will make the changes or what will be changed, nor is there evidence that it will bring growth.
8. Avoiding simple declarative sentences
Watch for: positioned as, committed to, aimed at, serving as a...carrier, playing the role of..., becoming an important lever for....
Problem: It could have been stated with "is, does, has, is responsible for," but instead it is wrapped into a positioning declaration.
Before: This platform is positioned as a comprehensive service carrier connecting creators and brands.
After: This platform connects creators and brands.
9. Abuse of "not only...but also..." and denying first, then affirming
Watch for: not only...but also..., not just...but rather..., not...but rather..., by no means...as simple as that.
Problem: Mechanically manufacturing progression and reversal.
Before: This is not only a product upgrade, but also a comprehensive innovation of the brand philosophy.
After: This is a product upgrade, and it also reflects a change in brand philosophy. The original text does not provide more specific changes.
10. Forcing three points and parallelism
Watch for: "First, second, third" regardless of whether the content suits it; three groups of structurally identical phrases; "faster, more accurate, more intelligent."
Problem: Cramming content into three items just to appear complete.
Before: We will take innovation as the guide, quality as the foundation, and service as the guarantee.
After: We will value innovation, quality, and service. The original text does not describe specific actions.
11. Synonym rotation
Problem: To avoid repetition, the same object is constantly given different names, which instead makes the reference ambiguous.
Before: This founder proposed a plan. The company's helm then explained the plan. This entrepreneur also answered questions.
After: This founder proposed and explained the plan, then answered questions.
12. False "from...to..." range
Problem: The two ends are not on the same scale; it is only meant to create a sense of encompassing everything.
Before: From a cup of coffee to a city, from individual emotions to the tide of the times, this book talks about everything.
After: This book writes about cafes, city life, and personal emotions.
13. Subjectless sentences and disappearance of the responsible party
Watch for: completed, effectively improved, the problem was solved, no configuration required, automatically saved.
Problem: Omitting who did what, especially making reports and instructions ambiguous.
Before: Related issues have been properly resolved, and data will be automatically saved.
After: The problem has been resolved, and the system will automatically save data. The original text does not say who resolved it or how often it is saved.
14. Overuse of dashes, parentheses, and supplementary explanations
Problem: Chinese could naturally form complete sentences, yet dashes are frequently used to create dramatic pauses, or parentheses are used to stuff in key information.
Before: This feature seems simple, yet it solves a long-standing problem—users finally no longer need to refresh repeatedly—at least most of the time.
After: This feature reduces repeated refreshing, but does not completely solve the problem.
Do not delete all dashes without exception. When the user's sample consistently uses them, keep a similar frequency; book title marks, hyphens, and numerical ranges are also not the object of this rule.
15. Overuse of bold and emphasis markers
Problem: Bolding keywords in every paragraph makes readers unable to see the true emphasis.
Before: We need to focus onefficiency、qualityanduser experience。
After: We need to improve efficiency and quality, and improve user experience. The original text did not state priorities or measurement standards.
16. Proliferation of "subheading: explanation" lists
Before:
- User experience: User experience has been significantly improved.
- Performance: System performance has been comprehensively optimized.
- Security: Security has been further strengthened.
After: The new version improves usability, performance, and security. The original text did not provide specific changes or data.
17. Parallel titles and sloganization
Watch for: Almost all titles are made into "Insight into Trends: Decoding the Future" or "Breaking Through Growth, Co-creating the Future."
Problem: The titles pursue neatness and momentum, but do not accurately explain the content.
Before: Insight into the Wave of Change: Decoding a New Growth Paradigm
After: What has changed in the way growth happens
18. Emoji-decorated structure
Problem: Using icons such as 🚀💡✅ instead of normal hierarchy, especially unsuitable for reports, emails, and explanatory documents.
Before: 🚀 Project progress 💡 Core insights ✅ Next steps
After: Project progress, findings, and next week's arrangements
When social platforms or the user's voice already use emoji, there is no need to mechanically delete them.
19. Mixed use of full-width and half-width characters and quotation marks
Problem: Mixing English commas, straight quotation marks, and unpaired punctuation in Chinese paragraphs, or putting quotation marks around every concept.
Before: We proposed "user first", and established a so-called "growth loop".
After: We proposed "user first", and established a so-called "growth loop". These terms do not explain what was actually changed.
Proper nouns, direct quotations, and symbols in code remain unchanged.
20. Chatbot residue
Watch for: Of course, OK no problem, the following is, hope it helps you, if needed I can continue, feel free to tell me anytime.
Problem: Conversational boilerplate is pasted into the main text along with everything else.
Before: Of course! Here is an introduction to the French Revolution: The French Revolution began in 1789. I hope this helps.
After: The French Revolution began in 1789.
21. Knowledge boundary disclaimers and guessing to fill gaps
Watch for: As of my knowledge update, based on currently available public information, sources are relatively limited, may, most likely, cannot rule out, has always kept a low profile.
Problem: When sources cannot be found, fill a paragraph with words about "not being able to find them," then invent plausible experiences or motives.
Before: Public information about her early life is limited, which shows that she has always kept a low profile and may have grown up in an ordinary family.
After: The available materials do not record her early life. (If this is not important, just delete it.)
22. Flattery and agreement
Watch for: You are absolutely right, this is a great question, one cannot help but admire, undoubtedly, obviously.
Before: You said cost will affect whether the plan is feasible. This question is very well raised, and you are absolutely right!
After: The cost issue you mentioned will affect whether this plan is feasible.
23. Redundant filler
Example fixes:
- "At this current point in time" to "now."
- "On the basis of..." to "based on...."
- "Conduct an in-depth analysis" to "analyze."
- "Have the ability to achieve" to "can."
- "It is worth noting that the data shows" to "the data shows."
24. Overqualification
Before: This policy may, to some extent, have a certain impact on some users.
After: This policy may affect some users.
Do not delete genuine uncertainty. Distinguish caution based on limited evidence from stacked words with no content.
25. All-purpose positive ending
Watch for: The future is promising, let us wait and see, we believe that with joint efforts, write a new chapter, steady progress over the long term.
Before: We believe that with everyone's joint efforts, the company will surely ride the wind and waves and create glory again.
After: Delete. End with the last fact, decision, or to-do item.
26. Four-character phrases and idioms used in succession
Problem: Using strings such as "gather strength, uphold integrity and innovate, forge ahead with vigor, practice diligently without slacking" creates momentum but no information.
Before: The team will gather strength, forge ahead, overcome difficulties, and drive the project to land and take effect.
After: The team will advance the project. The original text did not specify the specific task, person responsible, or time.
27. Authoritative tone and "essentialism"
Watch for: In the final analysis, viewed from the essence, what truly matters is, the core lies in, the underlying logic is, beyond doubt.
Problem: First adopt a posture of seeing through everything, then repeat an ordinary opinion.
Before: In the final analysis, the underlying logic of growth lies in truly understanding users.
After: Growth depends on understanding user needs. The source text does not explain the specific needs or evidence.
28. Preview-style openings and navigation phrases
Watch for: Let us, next I will take you through, the following will cover three aspects, without further ado, here is the conclusion first, after reading you will understand.
Problem: Spending a paragraph announcing what will be said next.
Before: Next, let us delve into remote work: it reduces commuting, but increases cross-time-zone communication costs.
After: Remote work reduces commuting, but increases cross-time-zone communication costs.
29. Repeating the title after the title
Before:
Delivery risk
Delivery risk is an issue we must take seriously.
Both suppliers postponed the delivery time by a week.
After:
Delivery risk
Both suppliers postponed the delivery time by a week.
30. Writing the body centered on the "revision process"
Problem: A non-changelog repeatedly says "added, optimized, adjusted, originally, now" without explaining the current feature itself.
Before: We added a search filter feature, replacing the original method that could only search page by page.
After: Users can filter search results and no longer have to search page by page.
31. Artificial golden lines and short-sentence barrages
Problem: Using extremely short sentences in succession to create drama, with every sentence sounding like poster copy.
Before: The times have changed. The rules have changed. Users have changed too. The old methods are completely ineffective.
After: The rules and user needs have both changed, and the old methods are no longer effective. The source text does not explain what changes occurred.
32. Empty metaphors and aphorism formulas
Watch for: X is the underlying color of Y, X is the key to Y, it is not a tool but a mirror, make X possible, time will give the answer.
Before: Data is the lighthouse of business decisions, and trust is the underlying color of teamwork.
After: Data can help companies make decisions; teamwork requires trust.
33. Rhetorical-question openings that pretend to be candid
Watch for: To be honest?, the answer may be unexpected, do you really understand..., here comes the question, then what is the cost.
Problem: Using dramatic pauses to create intimacy or suspense, while what follows is merely an ordinary judgment.
Before: Is this software worth buying? To be honest? It depends on how often you use it.
After: Whether this software is worth buying depends on your frequency of use.
Avoid false positives
Do not treat the following features alone as evidence of AI: correct grammar, stable style, formal vocabulary, one commonly used transition word, a single dash, a single short sentence, quotation marks, lists, or the absence of cited sources. Professional authors, editors, and templates also produce these features. Look for combinations in which multiple patterns appear at the same time, and judge in light of the text's purpose.
Preserve what is rare in human writing: specific and quirky details, contradictory feelings, era- and circle-specific language, explicable personal choices, variation in sentence length, and genuine interjections and self-corrections. Do not sand the voice flat for the sake of "cleanliness."
Invocation Modes
Pasted text (default): Give a first draft, a brief review, and a final draft. When the text is very short or the user only wants the result, the final draft may be given directly.
File mode: When the user specifies a file, only revise the natural-language body text, preserving code blocks, frontmatter, data, link targets, and formatting structure. Complete two rounds of review internally, then write back the final draft and briefly describe the changes.
Embedded mode: When this skill is one step in a larger task, output only the ready-to-use final draft and do not show the intermediate process.
Execution Flow
- Clarify the reader, purpose, and voice; when information is insufficient but does not affect safe rewriting, adopt a register consistent with the original text.
- Mark patterns that appear in groups, and protect facts, proper nouns, numbers, quotations, citations, and genuine uncertainty.
- For each piece of jargon, check "who, does what, to whom, with what result, on what basis." Find answers only from the original text and the user's additions.
- Write the first draft. Prefer concrete nouns and clear verbs; when details are lacking, acknowledge the lack rather than filling the gap with invented details.
- Ask yourself: "Where does it still read like mass-generated Chinese?" "Have I only replaced the jargon without restoring the information?" "Have I added facts or attitudes not present in the original?"
- Revise the final draft, and check information completeness, Chinese punctuation, reference, and tone.
Source
This skill is adapted from bladerby Siqi Chen (GitHub user Humanizer), under the MIT License. The upstream framework references Wikipedia's Signs of AI writing, maintained by WikiProject AI Cleanup. The Chinese rules are a localization rewrite oriented toward Chinese expression conventions, not a verbatim translation of the Chinese Wikipedia rules.