deslop-zh

Chinese AI-Tone-Free Writing Guide

1. Core Principles

1. Natural > Stylized

Highest priority:Reads naturally, not like "trying hard to sound human"。

• Don't force in colloquial words just to remove the AI tone ("put it plainly", "surprisingly", "who'd have thought", "so so so")
• Don't rewrite technical text like a spoken-word script just to be "more human"
• If the original sentence is already natural, accurate, and clear, don't change it

2. Subtraction First

Removing the AI tone is subtraction, not addition:

• First delete: explanatory tone, summarizing tone, redundant transitions, boilerplate filler
• Then adjust: sentence rhythm, punctuation, level of colloquialism in word choice
• Don't change: code literals, URLs, API paths, JSON key names, database field names

3. Preserve Semantics > Remove AI Tone

First confirm facts, logic, and causality are unchanged. Breaking the original meaning to be "more colloquial" = failed rewrite.

4. Scenario Determines Strategy

Different writing goals have different standards:

• Technical documentation: restrained, accurate, scannable, no second person
• Blog/technical long-form: natural, steady engineer's voice, neither report tone nor spoken-word tone
• Social media: has personality, has a stance, doesn't list features
• Release notes: one sentence on user effect, item density matching project style
• Scientific papers: maintain academic formality but avoid empty boilerplate

---

2. Quick Checklist (run through once after writing)

• Paragraph ends with a closing summary sentence starting with "this shows", "it can be seen", "at this point"? → Delete
• Has "not... but..." contrast construction? → State only the conclusion
• Has opening boilerplate like "with the development of...", "in today's... era"? → Delete
• Has jargon like "empower", "grip point", "closed loop", "align through", "precipitate"? → Replace
• Used an em dash (—)? → Replace with comma or period
• Four or five short sentences in a row like a telegram? → Merge or vary the rhythm
• Has transition sentences between sections like "the above covered... next we'll look at..."? → Delete
• No space between Chinese and English/numbers? → Add
• Used "very", "extremely", "it is worth noting", "in summary", "for example"? → Replace with "quite", "for instance"
• Has elevated sentences (raising a technical observation to a life lesson)? → Delete
• Used second person "you"? → Technical documentation should avoid it; blogs may use it as appropriate
• Has summary labels like "in one sentence" or "in short"? → State the conclusion directly
• Inanimate things doing human verbs ("decisions emerge", "the data tells us")? → Name the real person
• Overusing triadic parallelism? → Switch to paired or single
• Self-questioning and self-answering? → State directly
• Every list item starts with a bold keyword? → Change the format

---

3. Chinese AI-Tone Patterns

3.1 Paragraph-ending wrap-up summary sentence (highest frequency, check first)

A paragraph has just finished explaining a mechanism, and the ending restates it again.

Signal words: This shows / At this point / It can be seen that / This itself is / Through the above analysis / Thus it can be seen

❌ "At this point, we have already seen the three key optimization points of the memory allocator on the allocation path." ✅ Delete it directly; the previous sentence has already finished the point.

3.2 Elevation sentence

Elevating a specific engineering observation to a universal life lesson.

❌ "This is actually the same as product design: a good abstraction is not for showing off skill, but for making users make fewer mistakes." ✅ Give a concrete suggestion directly: "So when designing an abstraction, first write the calling code once from the caller's perspective, then decide what the interface should look like."

3.3 Contrast sentence pattern ("not... but...") — hard constraint

Do not use negative contrast structures. Any form of "not X, but Y", "not X, it is Y", "X itself has no value, what is truly useful is Y", "X is no longer the bottleneck, Y is" must all be rewritten.

❌ "The true innovator is not a person with ideas, but a person who maxes out five traits at the same time." ✅ "The true innovator is a person who maxes out five traits at the same time."

❌ "This is more like a founder screening framework, not a trading signal." ✅ "This is a founder screening framework."

❌ "The performance bottleneck is no longer I/O, but lock contention." ✅ "Lock contention is now the main performance bottleneck."

3.4 Lecturing tone / preachy opening

❌ "If you really break it down..." "Behind this is the same change..." "The truly key question is..." ❌ "First ask whether it is" "You need to first understand" ✅ Say it directly. "First confirm..." "You can first look at..."

3.5 Summary-tone labels

Do not add summary labels before the ending.

❌ In one sentence / In short / To put it broadly / In conclusion / To sum up / To summarize ❌ One-sentence takeaway / One-sentence explanation / One-sentence summary ✅ Give the punchline directly, without a label.

3.6 Article opening clichés

❌ "With the continuous development of AI technology..." ❌ "In today's digital wave..." ❌ "In recent years, XXX has received more and more attention..." ✅ Start directly. "After using it for less than a month..." "First, the conclusion..." "This time I want to explain the X chain clearly..."

3.7 Chinese jargon / internet jargon

The following words are not used by default (unless quoting the original text or fixed industry terminology):

• Absolutely prohibited: empower, lever, closed loop, precipitate, align, benchmark, connect through, open up, synergize, coordinate, insight, track, mindshare, tonality, campaign, chain, potential energy, backstop
• Use with caution: scenario, ecosystem, system, path, touchpoint, blocker, layout, matrix, granularity, retrospective, sort out, output, distill

❌ "Empower the business closed loop through data precipitation" ✅ "After accumulating data, the business can form a complete process"

Extra vigilance:

• Overemphasizing importance: For ordinary topics, using "crucial", "of great significance", "key moment", "turning point" → delete
• Over-listing sources: "According to multiple media reports..." "showing that it has a certain visibility in the industry" → delete
• Surface analysis attachment: Use participial phrases to append hollow commentary, e.g. "reflects the changes in contemporary Chinese society" → delete or provide specific evidence
• Promotional tone: "cultural heritage", "cultural richness", "natural beauty" → delete
• Vague attribution: "according to industry insiders", "some commentators believe" → either make clear who said it, or don't say it
• Outlook section boilerplate: "although (the topic) has its significance, the challenges it faces include..." → state the problems concretely, don't give a vacuous outlook

3.8 De-formalizing word choice

Don't use	Use
very / extremely	really
It is worth noting that	(state the conclusion directly)
In summary	(wrap up directly)
for example	like
purchase / use	buy / use
many students / quite a few students	many people / quite a few people
these few things	these few things

3.9 Punctuation rules

• Use corner brackets 「」, not curly quotes ""
• Em dashes (—) are prohibited; replace with commas or semicolons
• Don't use exclamation marks (unless it's a tweet/social context and there's genuine emotion)
• Chinese punctuation: ，。；：！？、「」『』【】
• Four or five periods in a row = sending a telegram → merge into a long sentence or connect with commas
• Don't use em dash (—); replace with a comma or period

3.10 Spacing in mixed Chinese-English typesetting

• Add a space between Chinese and English:CN 文字 EN → CN 文字 EN
• Add a space between Chinese and numbers:版本 2.0
• Objects that don't get spaces: code blocks, JSON key names, URLs, inline code

3.11 Second person

• Technical documentation: "you" (你/您) is prohibited. Use role words like "developer", "technical lead", or don't address the reader directly
• Blog: "you" (你) is allowed, but don't use "you'll find", "you need", "you can see" throughout

3.12 Translationese clichés

❌ Physical action verbs like "catch", "break through", "sharp", "won't collapse", "won't explode", "pierce", "hold up" used for abstract cognition ✅ Replace with everyday Chinese: "I've received all of these points from you", "this assumption doesn't hold"

❌ Abstract nouns as subjects:工程上的现实比这些数字难看 ✅ Rewrite, with a person/action as the subject.

3.13 Vague adjective pre-judgments

❌ "Cleaner: xxx" "The logic is very clear: xxx" ✅ Delete the adjective pre-judgment and just say xxx.

3.14 Sentence rhythm

• Using too many triple parallel structures? → Switch to double or single. One triple is fine; back-to-back triples are a pattern-recognition failure
• Using "first... second... finally"? → Switch to "on the one hand... on the other hand..."
• Do all paragraphs end with punchy short sentences? → Change the rhythm
• bold subheading + period pattern (**xxx**。content) → change to **xxx**，content, so that bold becomes a load-bearing word rather than a standalone heading

---

IV. Scenario-specific strategies

4.1 Technical documentation / API documentation / product copy

Trigger: "technical documentation", "API documentation", "product copy", "interface copy", "help documentation"

• Tone: restrained, direct, scannable
• Do not use second person ("you")
• Do not use promotional adjectives such as "leading", "powerful", "blockbuster", "disruptive", "shocking", "explosive"
• Do not use jargon such as "empower", "lever", "closed loop"
• Do not use exclamation marks
• Corner brackets 「」
• Add spacing between Chinese and English
• Do not change code literals, JSON key names, URLs, API paths
• Annotate a term the first time it appears; do not explain it again afterward
• Do not open with greetings such as "Hello" or "Hi"
• Goal: make clear "what it is, what it is used for, where to look next"

4.2 Technical blog / long-form article

Trigger: "blog", "technical article", "technical long-form article", "write one", "popular science"

• Default mode: the tone of a senior engineer explaining a system to a colleague
• Can be colloquial, but not slangy
• Reduce emotional words, interjections, and verbal tics
• Use fewer phrases such as "to put it plainly / actually / you'll find / who would have thought" to move the text forward
• Prioritize: clear, steady, specific, natural rhythm
• Prioritize subtraction: delete explanatory tone, summarizing tone, and unnecessary transitions
• Make polishing a subtraction, not a rewrite
• If the original sentence is already natural, do not change it
• For non-engineer audiences: first remove condescending words such as "classmates who do not understand technology", remove preachy tones such as "you must", and remove overly deep terminology. Take the path of "how I use it, you try it", not "systematic overview + framework table + long analytical paragraphs"

4.3 Social media / tweets

Trigger: "tweet", "X tweet", "twitter", "social post", "post"

Five rules:

1. Community first: open with a community anchor point (star count, thanking users, whose feedback drove it), put the change list later
2. Do not include all highlights: pick the 2-4 most interesting ones; readers want a story, not a changelog
3. User-feeling frame: write it as "when you use it..." "there is a feeling of...", not "this tool does..."
4. One stance: at least one opinion sentence stating the reason for the decision
5. Chinese rhythm: use idiomatic expression; avoid translationese and formal words

• End with an invitation, not a CTA
• Redact identity and sensitive signals
• Do not step on competitors
• Before publishing, scan for three things: redaction, not stepping on competitors, user feeling before feature list

4.4 Release notes / Release Notes

Trigger: "release", "changelog", "version", "release notes"

• Structure: Breaking Changes -> New Features -> Fixes & Improvements -> Deprecations
• Format: prioritize matching the target project's most recent release style
• Entries: 5-8 items, each about one sentence, only state user impact
• title: version number + the most core change or theme, no more than 10 words
• Do not add emoji (unless the project itself has a celebratory style)
• Do not add termination signals such as "no longer updated", "final release", "maintenance stopped" (unless it is the real situation)
• Chinese-English bilingual: entries correspond one to one, do not mix languages, English first

4.5 Scientific Papers / Academic Writing

Triggers: "paper", "abstract", "cover letter", "grant", "manuscript"

• Maintain academic formality, but remove empty filler
• Use "we" to refer to your own work (don't dodge with passive voice)
• Cite specific authors, don't use "some researchers have pointed out"
• Don't use "for a long time people have recognized..."
• Give specific numbers rather than vague conclusions
• Don't use "in today's rapidly changing field"
• Don't use "this paradigm shift has profound significance"
• Don't use "the results show" or "it was observed"

4.6 Bilingual Content

Triggers: "bilingual", "Chinese-English", "bilingual", "translation"

• Add a space between Chinese and English characters
• Don't mix punctuation: use Chinese punctuation for Chinese, English punctuation for English
• Keep terminology consistent throughout the text
• First occurrence of a term: keep the English + Chinese annotation (or Chinese first + English in parentheses)
• Afterward use only Chinese, don't mix Chinese and English
• Common replacements: context→上下文, state→状态, cache→缓存, claim→断言, runtime→运行时
• For terms that haven't been settled (prompt, embedding, tokenizer), keeping English is reasonable
• Bilingual version: item numbers correspond one-to-one, confirm EN/CN meanings match, mark translation loss
• Em dashes prohibited

---

Five, Vocabulary Replacement Table (Formal → Colloquial)

Don't use	Use
very / extremely	very
It is worth noting that	State the conclusion directly
In summary / in conclusion	Wrap up directly
for example	like
purchase	buy
utilize	use
possess	have
present	is / appears
carry out (+verb)	Delete "carry out" and use the verb itself
empower	provide / help
grip	key measure
closed loop	complete process
precipitate	form / accumulate
align	unify
pull through / connect through	connect / link through
insight	analysis conclusion
fallback	guarantee mechanism
crucial / of great significance	important
boast	have
is located in	in
is committed to	strive / focus on

---

Six, Execution Process

1. Identify the scenario first: Blog? Technical documentation? Tweet? Paper? Determine the tone standard.
2. Preserve semantics first: Confirm it hasn't been broken.
3. Subtract：
   • First scan the concluding summary sentence at the end of the paragraph (the most frequent AI trace in Chinese)
   • Then scan for contrastive sentence patterns ("not... but...")
   • Then scan for throat-clearing openings, meta-commentary, and emphasis crutches
   • Then scan for adverbs, passive voice, and false subjects
   • Then scan for jargon and empty adjectives
4. Then smooth the sentences: rhythm, punctuation, Chinese-English spacing, em dashes
5. Finally, read it back: prioritize fixing what is unnatural; do not force changes to what is natural
6. Output: provide only the rewritten text. Do not add a change list, explanation, or closing remarks. Unless the user explicitly requests them.

---

Seven, Rewriting Examples

Example 1: Chinese Technical Blog

Before rewriting:

As is well known, with the continuous development of cloud-native technology, Kubernetes has become the de facto standard in the field of container orchestration. It is worth noting that although K8s itself is powerful, many people find in actual implementation that it is not a silver bullet. This article will explore in depth from multiple dimensions the challenges and response strategies that enterprises face when implementing Kubernetes, hoping to bring some inspiration to those who are considering or have already embarked on the cloud-native journey.

After rewriting:

Our team has run Kubernetes in production for three years. The biggest takeaway: K8s solves the deployment problem, but shifts the complexity elsewhere. This article records five pitfalls we hit, and the stop-loss plan for each.

Changes: removed the opening boilerplate, removed "It is worth noting" and "many people", removed the contrastive sentence pattern "not a silver bullet", removed "explore in depth from multiple dimensions", removed the uplifting ending, and replaced them with concrete facts and personal experience.

Example 2: Technical Documentation

Before rewriting:

Hello! Welcome to our API service! Our platform provides leading, powerful data processing capabilities, able to empower your business and help you achieve overtaking on the curve in the track of digital transformation. Before use, you need to apply for an API Key first.

After rewriting:

The API provides the following data processing capabilities: batch queries, real-time push, and data cleaning. Before calling, you need to apply for an API Key; see the "Authentication" section for details.

Changes: removed the greeting, removed the exclamation marks, removed the promotional adjectives plus jargon chain, removed the second person, and replaced the filler with a feature list.

Example 3: Fixing Contrastive Sentence Patterns

Before rewriting:

The value of microservice architecture lies not in "micro", but in clear boundaries.

After rewriting:

The value of microservice architecture lies in clear boundaries.

Example 4: Purifying Empty Boilerplate

Before rewriting:

From multiple dimensions, Redis has crucial significance in high-concurrency scenarios. It can not only improve system response speed, but also has a profound impact on the scalability of the overall architecture.

After rewriting:

Redis uses a single-threaded model to handle reads and writes, eliminating lock overhead. Our stress tests found that on the same hardware, Redis's QPS is two orders of magnitude higher than MySQL's.

---

Eight, Quality Scoring (Optional)

Five dimensions, each scored 1-10; below 35/50 requires revision:

Dimension	Issue
Directness	Is it stating facts or previewing/summarizing?
Rhythm	Do sentence lengths vary, or are they all the same?
Trust	Does it underestimate the reader's intelligence (over-explaining, preachy tone)?
Authenticity	Does it read like something a human wrote or something AI-generated?
Density	Is there anything else that can be cut?

---

9. Related Skills

• tech-doc-style-chinese: Chinese technical documentation formatting specialization (corner brackets, spacing between Chinese and English, jargon avoidance)
• deslop-en: English de-AI-flavor writing (throat-clearing openings, adverbs, false subjects, contrastive sentence patterns, AI vocabulary ban list, etc.)