AI flavor is not "bad writing"; it is "the reader senses there is no real person behind it."But removing AI flavor ≠ good content. Spending time removing AI flavor is better spent on getting things clear.
Verbal Neat-Freakism(taxue): AI flavor is not an AI problem; it is using stock phrases to cover up uncertain thinking. The three elements of human speech: take a stance, hold a bias, carry emotion. Every time you delete a stock phrase, ask: what do I actually want to say? If after deleting you find the original had nothing real in it either, then the problem is not AI flavor—it is the content itself.
Execution Principles:
- By default, only diagnose without changing. Identifying AI features helps people see their own writing clearly, not helps people disguise themselves as human. Rewriting must be confirmed by the user.
- Removing AI flavor ≠ good content. If the detection result is very clean, tell the user directly. Do not force-find problems just to output a report.
- Rewriting must be based on the user's preferences. Behind every AI feature is a purpose that was originally intended. Ask about that purpose instead of directly replacing.
- See–Feel–Think–Act is the positive rewriting target(taxue). Removing AI flavor is not just deleting bad things; it is returning writing to the natural structure of "see (observation) → feel (emotion) → think (reflection) → act (action)."
Three rewriting methods (user chooses):
- Automatic replacement: detect → directly revise → quality validation → iterate
- Question-driven: detect → ask about writing intent item by item → give revision directions after the user answers
- Hybrid: automatic replacement + asking about intent for key features
How it works: first judge how heavy the AI flavor is, then choose a strategy—for mild cases change a few spots, for medium cases revise a paragraph, for severe cases rewrite the whole piece, for fatal cases stop it directly. After revising, check quality; if it does not pass, go back and redo, up to 3 rounds.
How to process step by step
First judge the content type (tweet/WeChat official account/academic) to determine the criteria for judging AI flavor. Then quickly scan once and count how many AI-flavor keywords are hit.
- Fewer than 5 hits → directly revise the marked positions
- 5 or more hits → enter full diagnosis, first find root causes (3 root causes can simultaneously trigger multiple symptoms), then verify independent signals
Before revising, first identify the author's voice—verbal tics, emotional passages, imperfect traces; mark these as protected zones and do not change a single word.
When encountering a sentence suspected of AI flavor, go through the decision tree to judge whether it should be deleted: in a protected zone? dense enough? still valid if the topic is changed? is it a breathing sentence? will deleting it lose warmth?
When rewriting, first remove obvious filler, then process paragraph by paragraph, and finally check the three elements of human speech (bias, emotion, stance).
After revising, run checks: is there any fabrication? is the core viewpoint still there? has new AI flavor been introduced? A/B level delivery; C/D level go back and rewrite. Fatal concentration is stopped directly and not allowed to be revised.
First judge the content type
Different genres have different tolerances for "standardized expression." Tweets allow imperfection, WeChat official accounts retain a sense of structure, and academic style should not be mistakenly harmed.
| Genre | AI-flavor sensitivity | Handling principles |
|---|---|---|
| Personal essays/tweets/short video scripts | High | Allow imperfection, allow tangents, allow emotion |
| WeChat official account/column articles | Medium | Keep a sense of structure, but break template-ization |
| Business reports/academic writing | Low | Standard expression is not AI flavor, don't misfire |
Genre judgment is only made when there are obvious features—most content defaults to the WeChat official account/column type. Detailed judgment criteria → references/extended-patterns.md。
Quick scan + full diagnosis
Quick scan
Scan the full text once and count how many of these keywords hit:
- Structural glue: "First… second… finally…", "It is worth noting that", "In summary"
- Empty words: "crucial", "indispensable", "deep empowerment", "significantly improve"
- Template sentences: "In today's… context", "With the development of…"
- Filler phrases: "like a shooting star streaking by", "time waits for no one"
- Conversational tone: "Thank you for your", "look forward to working with you", "let us"
Fewer than 5 hits → directly revise the marked positions. More than 5 hits → proceed to full diagnosis.
Full diagnosis
19 AI-flavor patterns, with 3 root causes that can simultaneously trigger multiple manifestations:
- Tone too neutral/authoritative → leads to symmetrical structure, neat sentence patterns, RLHF templating
- Clichés and empty modifiers → leads to too much generalizing language, empty emotional words, filler phrases
- Translationese → leads to RLHF templating, false scope
Find the root cause first; solving one can make multiple manifestations disappear at the same time. Handle the remaining independent issues (lack of detail, over-elevation, negative sensationalism, etc.) separately. No need to go through all 19 every time.
19-dimension quick reference table
Detailed identification signals, fixes, and boundary notes for each dimension →
references/extended-patterns.md
| # | Source | Identification signal | Fix |
|---|---|---|---|
| 1 | Structural symmetry | Three-part parallelism, uniform paragraph lengths, a summary sentence in every paragraph | Break symmetry, vary lengths |
| 2 | Too much generalizing language | "Overall", "in summary", abstract words replacing scenes | Replace with specific numbers/people |
| 3 | Empty emotional words | "Thought-provoking", "crucial" with no details | Delete emotional words, replace with specific details |
| 4 | Neat sentence patterns | Heavy repetition of "not only… but also…", uniform sentence lengths | Follow long sentences with very short ones, switch to colloquial |
| 5 | Lack of detail | No time, place, or people; "research shows" without saying which study | Add time, numbers, specific descriptions |
| 6 | Authoritative tone | No "I'm not sure" anywhere in the piece, like a textbook | Add uncertainty, bias, hesitation |
| 7 | Over-elevation | "Historic moment", "watershed", "milestone" | Keep things at their original scale |
| 8 | Negative sensationalism | "Not just X, but Y" forced elevation | State the facts directly |
| 9 | False scope | "From X to Y" forced pairing, "covers multiple fields" | Delete scope words, say what was done |
| 10 | Translationese | Too much passive voice, mixed Chinese and English, stacked nominalizations | Change passive to active, shorten sentences |
| 11 | Platform套路 | Xiaohongshu style/Douyin style/Zhihu style copied wholesale | Like that person, but not like a template |
| 12 | RLHF templating | Abnormal function-word density, too low sentence-length variance | Deliberately break fluency, add pauses |
| 13 | Formatting traces | Em dashes/bold in every paragraph, emoji, curly quotes | Restrain formatting, use corner brackets |
| 14 | Conversational traces | Sycophantic tone, disclaimers, filler phrases, over-qualification | Equal tone, state the content directly |
| 15 | Empty rhetoric | Endless metaphors, hollow calls to action, clichés, piled-up adjectives | Replace metaphors with numbers, delete clichés |
| 16 | Cliché modifiers | Dense "crucial", "in today's..." openings, universal sentence patterns | Delete or replace with data |
| 17 | Confused perspective | Person drifts between "I", "we", "you" | Fix a stable perspective, keep it throughout |
| 18 | False logical connections | "Therefore"/"so" connecting clauses with no causal relationship | Check causality, delete false causal words |
| 19 | Fabricated numbers | Precise to decimals but no source, no context | Keep if sourced, make vague if not |
The more AI flavor there is, the more aggressive the strategy:
- Mild (1-2 spots) → only fix those marked spots
- Moderate (3-4 spots) → rewrite key paragraphs, keep core points
- Severe (5-6 spots) → rewrite as a whole, keep core points
- Fatal (all AI flavor, no real person speaking) → stop, don't rewrite. Have the user first think through what they want to say
Special cases: AI flavor very light but structure verbose → recommend editor-revisor. Zero AI flavor → recommend editor-revisor or chinese-write-checker.
Identify the author's voice (protection zone)
Rewriting paragraph by paragraph without doing this first will delete human flavor as if it were AI flavor.
Receive voice list (if any)
If the user provides a voice list from voice-dissolver, cite it directly as the protection zone, don't re-judge. Sentences in the list stay untouched, word for word.
Judge on your own
Scan the whole text and use the following criteria to mark which sentences are "only this person would write it this way":
| Type | Operational criteria | Example |
|---|---|---|
| Catchphrase/signature expression | Same word/sentence pattern appears ≥ 3 times in the whole text, and in a non-AI-template context | The author likes to say "to be honest", not the clichéd "it is worth noting that" |
| Elevation at emotional moments | Emotional intensity clearly higher than surrounding text (≥ 2 levels), and supported by concrete details/experience | "That day when I saw the news, my hand stopped midair" |
| Imperfect traces | Awkward word order, self-correction, going off-topic, grammatical errors—these are all evidence of real human writing | "No wait, let me think... actually it's another matter entirely" |
Don't touch content within protected zones.
Genuine emotion vs. platitudes
Elevated moments of genuine emotion are anchored to specific context; deleting them loses the warmth. Platitudes work anywhere. How to tell: the topic-swap test—move the sentence into a completely unrelated topic and see if it still holds.
Judgment: should it be deleted
After detecting suspected AI flavor, don't act directly. Follow the decision tree.
1. 在保护区内?
是 → 留(不进入后续判断)
否 → 继续
2. 200 字内出现 ≥ 3 次?
否 → 不算 AI 味,留
是 → 继续
3. 换话题还成立?
(把这句话放到另一个完全无关的话题中,看是否仍然成立)
不成立 → 这句话和具体内容绑定,是作者自己的话 → 留
成立 → 继续
4. 是两段之间的喘息句?
(位于两个长段落/高强度段落之间的短句或低强度句)
是 → 留
否 → 继续
5. 删了之后会失去温度/锋利/犹豫吗?
是 → 留
否 → 删
Core detection principle: look at density, not single occurrences. Parallelism, parallelism of phrases, and idioms are already used in human writing. Repeated stacking, no connection to the surrounding content, and still holding up when moved to another topic—that's what AI flavor is.
Avoid over-correction: removing AI tone ≠ removing all standardized expressions. Common human phrasings (like "supports platforms such as X" or "used for X") are not AI tone in themselves; don't force-delete them just to be "more colloquial."
Rewrite execution
First clear out the obvious
Mark the following for direct deletion:
- Structured filler: firstly/secondly/in addition/in summary/it is worth noting/taken together/overall/it is not hard to find/obviously/as everyone knows/undoubtedly/deep empowerment
- Piled-up adjectives: important/crucial/significant/effective/vital (degree modifiers without data support)
- Slogan endings: let us together/the future is bright/join hands and move forward
- Formatting abuse: excess bolding, dash overuse, divider lines, numbered subheadings (1, 2, 3)
Pre-cleaning is not rewriting; it's just quickly cutting out obvious filler.
Rewrite section by section
Address the diagnosed problems section by section, skipping the marked protected zones.
Rewriting principles:
- Preserve the original's core points and key information
- Break symmetric structures; vary paragraph lengths
- Replace empty emotional words with specific details
- Add specific times, numbers, people (but anchor only to clues already in the original text; don't fabricate)
- Allow imperfect tone, allow hesitation—the perfect one is a god; the one who talks a blue streak is human
- Avoid over-correction—don't force colloquialism just to remove AI flavor
- Important points should be explained like you're talking to someone familiar: lay out the logical chain, background, and why you think so—don't just drop conclusions
- When rewriting, anchor viewpoints to specific experiences, scenes, or details that appear in the text
- Check rhythm and conversational feel: where it should be long, fully explain the causes and consequences; where it should stop, land with a short sentence
- After rewriting, go over it yourself: did you round off the sharp, hesitant, or imperfect parts of the original? If so, restore them
- Any change must be traceable to the author's preferences already revealed in the text; don't use "more natural" or "more human" as a default preference to override the stiffness, hesitation, or sharpness the author actually wants to keep
Step three: the three elements of human speech check
After deleting platitudes, check whether the whole text has these three things:
- At least one genuine bias: "I know this view is controversial, but I really think..."—the bias must be rooted in something specific: tie the judgment to a fragment the author has experienced, seen, or tried
- At least one moment of genuine emotion: not an empty phrase like "it was deeply moving," but specific to the reaction at that moment—"when I saw this message, my hand stopped in midair." The emotion should unfold the way you'd recount that day's scene to a friend
- At least one genuine stance: take a clear side, don't straddle the fence. A stance can have back-and-forth thinking: "At first I also thought... later when I ran into this, I realized..."
Quality control
After rewriting is complete, go through the checklist. Anything that doesn't pass goes back to the judgment layer or the rewriting layer to be redone.
Hard indicators (veto)
| Check item | Standard | Fail |
|---|---|---|
| Hard boundary | Are there any fabricated characters, stories, details, or sample sizes that aren't in the original text? | Yes → rewrite entirely |
| Alignment | Can every core judgment, key data point, and author stance in the rewritten version be found in the original text? | Can't find it → rewrite the problematic paragraph |
| Information fidelity | Have the core viewpoints, key data, or author stance been mistakenly deleted or altered? | Mistakenly deleted → restore the original text |
| Newly introduced clichés | Has the rewrite introduced new AI flavor? | Yes → delete it |
Soft indicators
- Person consistency: the perspective is unified throughout
- Causal chain authenticity: "therefore" and "so" have real causation before and after
- Rhythm variation: long and short sentences alternate, not every sentence the same length
- Ending quality: not a slogan ending, but with concrete action or lingering resonance
Quality grading: A (all pass) deliver, B (deliver after minor adjustments), C (go back and rewrite the problematic paragraph), D (re-judge and rewrite entirely). Maximum 3 rounds; if still not up to standard on the 3rd round, deliver the current version and note the unresolved issues.
When the content itself is found to have problems
Content quality (information density, logic chain, clarity of stance) is not the job of removing AI flavor.
If the content itself is found to have problems (low information density, logical jumps, vague stance), it's recommended that the user first use chinese-write-checker for a checkup, complete the content skeleton, and then come back to remove the flavor.
Diagnostic report template
【AI 味有多重】{轻微/中等/严重/致命}
【体裁】{类型}
【作者声音保护区】
{标记了哪些句子——口头禅、动情处的拔高、不完美痕迹。这些不碰。}
【问题定位】
来源{X}({类型})——第{X}段/第{X}句
原文:「{原文}」
200字内出现:{X次}
决策树结果:{留/删,原因:动情处/喘息句/换话题不成立/密度不足/…}
问题:{一句话}
改法:「{建议改法}」
【人话三要素检查】
偏见:{有/缺——补的话扎在原文已有线索上}
情绪:{有/缺——补的话展开到具体瞬间}
立场:{有/缺——补的话让它有来有回的过程}
【改写版本】
{直接输出改写后的完整文本}
【改写说明】{改了什么,为什么;重点说明锚定了什么具体经历、节奏调整、对话感}
【过度修正检查】{有没有磨平了锋利、犹豫、不完美的地方}
【节奏检查】{句子长短变化够不够、读起来有没有呼吸感}
【下一步建议】
{条件触发,只在结果明确指向下一步时推荐}
Self-check only, no rewriting
If the user only requests a self-check, follow this process separately. The check standards and quality grading are the same as above.
Notes
- AI flavor diagnosis must be pinpointed to specific sentences or paragraphs; don't say "it has an overall AI feel"
- For each source of AI flavor, give an "original text → revision" comparison
- Don't say "add personal feelings"—just add them directly in the rewritten version
- Eliminating AI flavor is not "making the text look better," it's "making the reader feel a real person is present"
- After rewriting, the article may become shorter and "less complete"—this is correct, not a problem
- You must output the complete rewritten version, you can't just give a diagnosis without a rewrite
- If the user has already provided a voice list but you re-determine the protected zone → the list is already confirmed upstream, quote it directly
DO NOT
- Diagnosis only without rewriting → diagnosis + rewriting are one integrated whole
- Deleting the author's genuine hesitation, sharpness, or imperfection as AI flavor → first determine the protected zone before taking action
- Forcing colloquialism to remove the AI flavor → removing the AI tone ≠ removing all standardized expression
- Don't look only at a single occurrence instead of density → one instance of parallelism/antithesis is not an AI flavor; 3 or more occurrences within 200 characters is
- Writing experiences or details for the user that they don't have → the specific scenes added must be anchored to clues already present in the original text
- Fabricating sample size or sources → if the original says "73.2% of users are satisfied," it cannot be changed to "200 questionnaires were sent out"
- The follow-up question itself must not exhibit AI characteristics(dbs self-check rules):
- Don't start with "you might feel" (this is putting stupid words in the reader's mouth)
- Don't use the "it's not X, it's Y" structure for follow-up questions
- Don't use "essentially" or "in the final analysis" to elevate the follow-up question
- Don't use a multiple-choice structure (A or B?) — this is thinking up the answer for the user
- Directly state the phenomenon you observed, then ask an open-ended question
- Removing the AI flavor ≠ good content → if the detection result is clean, just say so; don't force yourself to find problems
Next-step suggestions (condition-triggered)
| Result condition | Recommendation |
|---|---|
| AI flavor removed, but the structure is loose/verbose | Recommend using editor-revisor to streamline |
| AI flavor removed, but the title/opening/layout/punctuation need polishing | Recommend using packaging-workshop for a presentation-layer check |
| During de-flavoring, found that the personal voice was covered up | Recommend using voice-dissolver first to clarify the voice |
| After rewriting, the user is unsure about the quality | Recommend using chinese-write-checker for a checkup |
| Fatal concentration — no real person is speaking throughout the text | Don't recommend a next step. Stop it directly and have the user first think clearly about what they want to say |
Example
→ references/examples.md
Shared specifications
This skill shares writing specifications with chinese-write-checker, voice-dissolver, editor-revisor, and packaging-workshop (cliché list, format specifications, protected-zone logic, topic-switch test, density principle). See ~/.agents/skills/chinese-write-checker/references/writing-standards.md(cross-skill reference). Read as needed; do not preload the full text.