SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsai-flavor-detector

ai-flavor-detector

Data & analytics

View repository

Updated 10-15 22:11

About this skill

AI Flavor Detector

Deeply analyzes the AI-generated characteristics of text, evaluating the degree of "human flavor" of the text from multiple dimensions.

When to Use This Skill

Trigger this skill when the user:

  • Asks to check whether the text looks AI-written
  • Asks "Does this article have a heavy AI flavor?"
  • Needs de-AI-ification rewriting
  • Wants to bypass AI detectors (such as Zhuque)
  • Wants to make the text more human-flavored and more natural
  • Mentions trigger words such as "check AI flavor", "detect AI flavor"

Core Workflow

  1. Obtain the text to be detected - The user pastes text or specifies a file path
  2. Multi-dimensional analysis - Detect AI characteristics from 6 dimensions
  3. Output score - Give an overall AI flavor score (0-10)
  4. Paragraph-by-paragraph annotation - Annotate problem locations and keywords
  5. Provide rewriting suggestions - Give specific de-AI-ification rewriting examples

Output Format

Overall score

AI 味评分：X/10
- 0-3 分：人味很浓，几乎看不出 AI 痕迹
- 4-6 分：人味为主，偶有 AI 特征
- 7-10 分：AI 味明显，需要改写

Paragraph-by-paragraph analysis

For each paragraph of text, perform:

  • AI flavor score (0-10)
  • Problem location annotation (line numbers, keywords)
  • Specific problem description
  • Revision suggestions

Problem summary

List all AI flavor problems by priority:

  • High priority (must change)
  • Medium priority (recommended to change)
  • Low priority (optional)

Zhuque-specific detection (if applicable)

If the user mentions Zhuque detection, additionally output:

  • Conjunction density analysis
  • Structuredness detection
  • Parenthetical usage explanation annotation
  • Paragraph regularity analysis

Detection Dimensions

1. Sentence structure analysis

  • Three-part formula: Detect the standardized "definition - explanation - summary" structure
  • Over-structuring: Analyze whether "first, second, finally" is overused
  • Paragraph length: AI tends to use paragraphs of uniform length
  • Sentence complexity: AI sentence structures are often too neat

2. Vocabulary feature analysis

  • Conjunction frequency: Detect "however", "in addition", "it is worth noting that", "in conclusion", "at the same time"
  • Quantifying vocabulary: AI likes "greatly", "significantly", "widely", "comprehensively"
  • AI-specific vocabulary: "empower", "lever", "closed loop", "track", "matrix", "underlying logic"
  • Prohibited word detection: Xiaohongshu prohibited words (juejuezi, yyds, invincible, super useful)

3. Subject and tone analysis

  • Subject usage: "we", "let us" vs "I", "my"
  • Tone judgment: objective and rational vs emotional, self-deprecating, personalized
  • Proportion of imperative sentences: AI tends to use "suggest", "should", "must"
  • Lack of interrogative sentences: AI rarely uses rhetorical questions or self-questioning

4. Specificity and authenticity

  • Time details: Is there a specific time (the evening of the 20th, January 21st)
  • Location details: Is there a specific location or scene
  • Personal experience: Is there a real story or specific case
  • Data precision: AI uses "about, roughly", real people use "1 minute 52 seconds"

5. Internet slang and colloquialism

  • Use of internet slang: Whether internet buzzwords are used
  • Naturalness of use: Trying too hard vs blending in naturally
  • Colloquial expressions: "good grief", "new pit", "amazing" etc.

6. Sentence opening patterns

  • Repeated sentence openings: AI tends to start with similar sentence patterns
  • Passive voice: Overuse of passive voice
  • Nominalization: Overly turning verbs into nouns

Zhuque AI Detection

Zhuque AI detection is another professional AI detection tool, and the following features require special attention:

Conjunctions Zhuque focuses on detecting (high priority)

The following conjunctions have very low density in normal text. If one appears every 3-4 sentences, it is judged as AI:

  • Structural conjunctions: First, second, third, then, in addition, in short, in conclusion
  • Transitional conjunctions: however, but, actually, in fact
  • Supplementary conjunctions: also, actually, it is worth noting that
  • Temporal conjunctions: with, development, in a certain era, today in a certain
  • Summary conjunctions: in summary, it can be seen that, it can be said that

Zhuque prohibited features (must-check items)

  1. Numbers and lists

    • Prohibited: one, two, three, as well as Arabic numerals 1, 2, 3, etc.
    • Detection: real people rarely use numbered lists when writing opinions

  2. Parentheses and explanatory markers

    • Prohibit any parentheses (round brackets, square brackets, etc.) and explanatory markers
    • Detection: real people rarely use parentheses for supplementary explanation

  3. Dash usage

    • Frequent use of dashes for supplementary explanation
    • Example: "this feature—not merely an extension of functionality—represents..."
    • Detection: dashes are rarely used in natural writing, while AI favors them

  4. Punctuation at the end of paragraphs

    • Real human writing style: no punctuation at the end of paragraphs
    • AI writing feature: every paragraph ends with a period
    • Detection: observe whether paragraphs all end with punctuation

Anti-Zhuque core strategy (human-flavor signals)

The following features are strong human-flavor signals, used to counter Zhuque detection:

  1. Make the thought process authentic

    • Show a real thinking process, including hesitation, turns, sudden thoughts
    • Example: "Oh right, there's one more thing I want to say..."
    • Ask a question somewhere and then answer it yourself
    • Share a small personal story or feeling

  2. Make expression imperfect

    • Paragraphs vary in length; sometimes a single sentence can be a paragraph
    • Colloquial expressions and not very standard sentences
    • Some opinions may be expanded with a few more sentences, rather than covering everything
    • Transition naturally, rather than with obvious structural words

  3. Naturalize emotional fluctuations

    • Add your own emotions and views, sometimes even a bit subjective
    • Interact with readers, ask questions, provoke thought
    • Exclaim when it's time to exclaim, but don't overdo it
    • Overall it looks casual and natural, not like something carefully designed

  4. Naturalize perspective switching

    • Use "I" when telling personal experiences
    • Use "we" when sharing common experiences
    • Use "you" when giving advice directly to readers
    • The third person can be used when discussing universal phenomena

Resources

Detailed detection criteria reference:

  • references/wikipedia-ai-features.md - Wikipedia Comprehensive Guide to AI Writing Characteristics
  • references/zhuque-detection.md - Zhuque AI Detection Detailed Guide and Countermeasures

Troubleshooting

Misjudging technical definitions as AI flavor

  • Problem: Misjudging technical definition introductions as AI characteristics
  • Solution: The introduction of technical definitions is reasonable and should not be judged as AI flavor

Misjudging prompt instructions as main text

  • Problem: Judging instructions to AI such as "I have worked for 20 years..." as AI flavor
  • Solution: Distinguish prompt instructions from the main text of the article

Misjudging code and commands

  • Problem: Judging standardized code and commands as AI characteristics
  • Solution: Technical commands are themselves standardized and should not be judged as AI flavor

Misjudging quoted content

  • Problem: Judging quotations from official documents or standards as AI flavor
  • Solution: Quoted content itself is standardized expression and should not be judged as AI flavor

Detection logic

AI flavor = (standardized sentence patterns × frequency of connectives) + (lack of personal details × lack of emotional expression) - (personal experience × colloquial style × specific details)

Output principles

  1. Don't just give a score - Give specific locations and reasons
  2. Distinguish main text from prompt - Distinguish whether it is the main text of the article or instructions to AI
  3. Provide revision suggestions - Give specific rewriting examples
  4. Avoid misjudgment - Do not misjudge technical definitions or quoted content as AI flavor
  5. Encourage personalized expression - Emphasize the value of personal experience and colloquial expression

Reference standards

  • Wikipedia "AI Writing Characteristics" Comprehensive Guide
  • Anti-Zhuque AI Detection Prompt (Yao Jingang V2.0)