ai-slop-reviewer

AI slop review and humanization skill

AI-generated text is accurate in content, but is often full of mechanical and predictable patterns. This skill systematically finds those patterns and restructures the writing so that readers feel it as a human voice.

Overview

This skill is the final step in every text deliverable workflowand removes AI-specific mechanical patterns. It applies to all text deliverables such as blogs, newsletters, copy, business plans, contracts, proposals, reports, emails, landing copy, business reports, patent drafts, lecture manuscript drafts, etc. all text deliverables.

It does not apply to code, JSON/CSV data, charts and tables, numerical reports, or simple lookup responses.

Trigger keywords

This skill is invoked automatically or manually in the following situations:

1. Explicit request: "Remove AI slop," "Revise it so it looks human-written," "Fix the parts that seem AI-generated"
2. After completing website/content production: Immediately after writing a blog post, landing page, copywriting, or newsletter
3. After completing business document writing: Immediately after drafting a business plan, proposal, report, or email
4. Post-processing AI tool output: When requesting review of text generated with Cowork, Claude, ChatGPT, etc.
5. Quality improvement request: "It feels too mechanical," "Polish it naturally"

Workflow

Step 1: Diagnose AI slop patterns

When text is received, first identify problem patterns based on the checklist.

🔴 Banned words: remove/replace immediately

Overused Korean expressions:

• ~에 대해 살펴보겠습니다 → Get straight to the point
• ~의 중요성은 아무리 강조해도 지나치지 않습니다 → Specifically why it matters
• ~을 통해 알 수 있듯이 → Delete or describe directly
• 다음과 같습니다: + list → Integrate into prose or use a list only when truly necessary
• 이처럼 ~은 매우 중요한 의미를 가집니다 → Empty wrap-up in the conclusion
• 종합적인, 체계적인, 효율적인 (when repeated without meaning)
• 또한, 그러나, 따라서, 더불어, 한편 (when repeated in every paragraph)
• ~라고 할 수 있습니다 → ~입니다 or with a specific description
• ~를 기반으로 한 접근 방식 → Explain directly what approach it is
• 핵심은 + repetition of the same wording

Overused English expressions:

• delve, leverage, utilize, harness, embark → replace with simple verbs
• pivotal, robust, seamless, cutting-edge, comprehensive → with concrete description
• tapestry, landscape, realm (when used metaphorically), synergy, testament
• "It's important to note that", "In today's ever-evolving world"
• "In conclusion / In summary / In essence" (a conclusion appearing at the beginning)
• Overuse of the "Not just X, but Y" structure
• Answer starting with "Certainly!"

🟡 Structural AI patterns

Uniform sentence length:

• If all sentences are of similar length (15-22 words), it is not human-written text
• You need to mix short sentences (3-7 words) and long sentences (25+ words) to create rhythm

Formal paragraph structure:

• If every paragraph follows the pattern "topic sentence → explanation → example → mini-conclusion," it is AI writing
• Vary paragraph length and structure

AI-style introduction:

• Sentences like "In today's rapidly changing world...", "In the digital age..."
• The format "In this article, we will look at X"

AI-style conclusion:

• "As we have seen so far..." + repetition of previous content
• An appeal like "we must strive for a better future" without concrete content

Excessive listing:

• Listing everything that could be written as prose with bullet points (•)
• Three or more nested lists

Overuse of passive voice:

• "It has been confirmed that ~", "It is suggested that ~", "It is judged that ~"
• → Clarify the subject and use active voice

🔴 3 types of structural slop S1 patterns (mandatory detection for headlines and copy)

The 3 most frequently appearing in marketing copy and slide headlines structural AI slop patterns. Detection is at the sentence structure level, not a word dictionary. Even a single standalone occurrence confirms an AI author, so **S1 (decisive, remove immediately upon discovery)**. (Since most commerce and copy skills chain only this skill, these patterns must be in this checklist for the gate to block slop copy.)

#	Pattern	Detection signal	[Bad example]	Revision direction
1	Dash-contrast headline	Dash (em-dash —) structure that splits a sentence to create contrast, the "X — Y" structure (dash-contrast headline)	[Bad example] "From copy-paste to delegation — give only the goal, and the agent handles the rest"	Remove the dash and merge into one sentence, or split into two sentences
2	Particle/noun-ending fragment	A fragmented fragment where the sentence ends with a particle (~은/에/로) or a noun (nominal form) (particle/noun ending)	[Bad example] "The key to success — automation" (noun-ending fragment)	Rewrite as a complete sentence including a predicate
3	"From A to B" transition formula	A cliché that opens the introduction/headline with the "from X to Y" transition formula	[Bad example] "From Excel to Notion, the way we work has changed" (transition formula)	Start with a concrete case or evidence instead of the transition formula

Detection source: plugins/moai-writer/skills/korean-humanize/references/ai-tell-taxonomy.md D category ("from X to Y" conversion formula) + J category (em-dash overuse). If this gate fails to catch these 3 types, slop copy will be distributed after passing the commerce/copy skill chain (which calls only this skill).

Step 2: Humanization Revision Strategy

Once the diagnosis is complete, apply the strategy below to rewrite the text.

Strategy 1: Specify the Voice

• Reveal who the writer is. It must contain opinions, judgments, and experiences
• "This feature is efficient" → "Thanks to this feature, our team reduced weekly meetings by 30 minutes"
• Discard vague generalizations and use specific examples, dates, and numbers

Strategy 2: Vary the Rhythm

• Create emphasis with a short sentence. Just one line
• The next sentence fills in the context. This makes the reader's breathing natural
• Split long, complex sentences into two, or sometimes deliberately leave them long to give a rich explanatory feel

Strategy 3: Transition Word Diet

• If transition words such as "also," "moreover," and "meanwhile" appear in succession, try deleting them one by one
• Check whether it flows naturally without transitions. Most of the time it does
• Use them only for transitions that are truly necessary. That is how transition words gain meaning

Strategy 4: Convert Lists → Prose

• For three or fewer items, use prose: "The three key points are A, B, and C"
• When a list is truly necessary: information that is actually being compared or whose order matters

Strategy 5: Inject Specificity

• "Many people" → "63% of survey respondents" or "4 out of 5 people on our team"
• "Significant improvement" → "Processing speed doubled"
• "According to recent research" → "According to a 2024 MIT study"

Strategy 6: Rewrite the Introduction

• The introduction of AI writing should almost always be discarded. Write the first 1-2 sentences yourself from scratch
• Start with a question the reader might be curious about or a surprising fact
• Alternatively, the inverted pyramid approach of presenting the conclusion first is also effective

Strategy 7: Rewrite the Ending

• Do not simply repeat the previous content
• Finish with a next action (CTA), an open question, or a line addressed directly to the reader

Step 3: Revision Output Format

When providing the revised text, follow this format:

### 🔍 진단 요약
발견된 패턴: (예: AI식 도입부, 수동태 남용, 금지어 12개, 획일적 문장 길이)
심각도: 낮음 / 보통 / 높음

### ✍️ 수정된 텍스트
(수정된 전체 텍스트)

### 📝 주요 변경사항
(중요한 수정 포인트를 3~5개 간략히 설명)

If the original is long (500 characters or more), show the diagnostic summary and major changes first, then provide the full revised text afterward.

Usage Examples

Example 1: Blog Post (Korean)

Input:

오늘날의 급변하는 비즈니스 환경에서 효율적인 팀 커뮤니케이션의 중요성은 아무리 강조해도 지나치지 않습니다. 또한, 원활한 협업을 통해 조직의 생산성을 극대화하는 것이 중요합니다. 다음과 같은 방법을 통해 커뮤니케이션을 개선할 수 있습니다:
• 정기적인 미팅 실시
• 명확한 역할 분담
• 피드백 문화 조성

Output:

### 🔍 진단 요약
발견된 패턴: AI식 도입부, 금지어 3개(~의 중요성은 아무리 강조해도 지나치지 않습니다, 또한, 다음과 같습니다), 과도한 목록화, 획일적 문장 길이
심각도: 높음

### ✍️ 수정된 텍스트
우리 팀은 작년에 미팅을 줄였더니 오히려 일이 더 빨리 됐습니다. 역설적이지만 사실입니다. 비결은 세 가지였는데, 생각보다 단순합니다. 매주 월요일 15분 스탠드업 미팅만 유지하고, 각자의 담당 범위를 노션 문서로 공유했습니다. 그리고 결과물에 대해 솔직한 피드백을 주고받는 분위기를 만들었습니다—처음에는 어색했지만, 한 달 후엔 팀원들이 먼저 피드백을 요청하기 시작했습니다.

### 📝 주요 변경사항
- "오늘날의 급변하는 비즈니스 환경에서..." 도입부 삭제 후 구체적 경험으로 시작
- 금지어(~의 중요성, 또한, 다음과 같습니다) 제거
- 목록(•)을 산문으로 변환하여 리듬감 추가
- "피드백 문화 조성" → "한 달 후엔 팀원들이 먼저 피드백을 요청하기 시작했습니다"로 구체화

Example 2: Business Email (English)

Input:

In today's rapidly evolving landscape, it is crucial to leverage cutting-edge technologies to streamline your workflow. Furthermore, it is important to note that comprehensive integration of these tools can significantly enhance productivity. Let us delve deeper into the multifaceted approaches available.

Output:

### 🔍 진단 요약
발견된 패턴: AI식 도입부(In today's rapidly evolving landscape), 금지어 5개(leverage, cutting-edge, furthermore, important to note, comprehensive, delve), 획일적 문장 길이, 수동형 느낌
심각도: 높음

### ✍️ 수정된 텍스트
Most productivity tools promise to save you time. Most of them don't. But after testing 12 different apps over three months, I found two that actually changed how our team works—and neither of them cost more than $20/month.

### 📝 주요 변경사항
- "In today's rapidly evolving landscape..." 삭제 후 직관적인 관찰로 시작
- "leverage", "cutting-edge", "delve" 등 AI 흔한 단어 제거
- 3문장 모두 비슷한 길이(20~25단어) → 8단어, 7단어, 32단어로 리듬 변화
- 능동태와 구체적 숫자(12 apps, three months, $20/month) 추가

Example 3: Report (Mixed)

Input:

본 보고서에서는 2024년 고객 만족도 조사 결과를 분석하고자 합니다. 조사 결과에 따르면, 전반적인 만족도가 전년 대비 15% 향상된 것으로 확인되었습니다. 또한, 제품 품질과 서비스 응대 속도가 주요 요인으로 작용한 것으로 판단됩니다. 다음과 같은 개선안을 제안합니다:
• 제품 품질 테스트 강화
• 고객 센터 인력 증원
• 피드백 시스템 도입

Output:

### 🔍 진단 요약
발견된 패턴: AI식 도입부(~에서는 ~하고자 합니다), 수동태(~확인되었습니다, ~판단됩니다), 금지어(또한, 다음과 같은), 과도한 목록화
심각도: 보통

### ✍️ 수정된 텍스트
고객 만족도가 전년 대비 15% 올랐습니다. 숫자만 보면 좋아 보이지만, 원인을 찾아야 이 개선을 유지할 수 있습니다. 우리 팀이 분석한 결과, 두 가지가 핵심이었습니다: 제품 불량률을 3%에서 1.2%로 낮춘 점, 그리고 평균 응대 시간을 7분에서 4분으로 줄인 점입니다. 이제 세부 대책입니다. 첫째, 품질 테스트를 매주에서 매일로 바꿉니다. 둘째, 고객 센터 인력을 3명 추가합니다. 셋째, 실시간 피드백 시스템을 도입해 문제를 즉시 잡습니다.

### 📝 주요 변경사항
- "본 보고서에서는 ~하고자 합니다" 삭제 후 결과(15% 향상)부터 시작
- "~확인되었습니다", "~판단됩니다" 수동태 → 능동태로 변환
- "제품 품질과 서비스 응대 속도" → 구체적 수치(불량률 3%→1.2%, 응대 시간 7분→4분)
- 목록(•)을 "첫째, 둘째, 셋째" 산문으로 변환하여 연결성 강화

Output Format

This skill outputs results in the following structure:

1. Diagnosis Summary: List of detected AI patterns and severity (low/medium/high)
2. Revised Text: The full text reconstructed in a human voice while preserving the original meaning
3. Key Changes: Explanation of 3-5 key revision points

The Diagnosis Summary lets the user quickly grasp "which patterns were detected" before revision. If the severity is 'high', it suggests there are many AI patterns and the difference before and after revision will be large.

Precautions

✅ Principle of Preserving Meaning

• Change the form, but always preserve the core information and message
• Be careful not to change factual relationships (numbers, dates, proper nouns, etc.)

✅ Maintain Tone and Manner

• Official documents, reports, and technical documents require formality
• Maintain a tone suited to the target reader and purpose
• Do not make it excessively colloquial

✅ Context-Based Judgment

• Do not mechanically remove all AI expressions
• Expressions like "utilize" and "it is important" can also be natural depending on the context
• The problem is the repetition of patterns, not the words themselves

✅ Confirm User Intent

• If there are stylistic characteristics the user wants to keep, confirm them
• Example: "I'll keep the professional tone, shall I just remove the AI patterns?"

❌ Excluded from Application

This skill does not apply in the following cases:

• Code (programming language source)
• JSON/CSV data (structured data)
• Charts and tables (visualization materials)
• Numeric reports (statistics, metrics)
• Simple lookup responses (information provision)

Related Skills

Preceding Skills (input generation)

• moai-marketer:content-blog — Blog post generation
• moai-marketer:marketing-landing-page — Landing page copywriting
• moai-marketer:content-newsletter — Newsletter writing
• moai-marketer:content-copywriting — Marketing copy writing
• moai-officer:doc-docx — DOCX document generation
• moai-officer:doc-pptx — PPTX presentation generation

Following Skills (optional)

• moai-media:media-higgsfield-image — (Optional) Image generation step of the blog publishing chain

Integrated Workflow

This skill is invoked as the final stepof the following chain:

블로그 발행:
  content-blog → ai-slop-reviewer → (선택) media-higgsfield-image

사업계획서(PPT):
  consult-strategy → doc-pptx → ai-slop-reviewer

제품 랜딩:
  content-copywriting → marketing-landing-page → ai-slop-reviewer

It is a required post-processing step in all text output workflows.

References

File

Load Conditions

references/kr-pii-masking.md

Personal Information Masking Sharing Standard (physically owned by this skill). Referenced by all skills that process outputs containing resident registration numbers, account numbers, cards, contact information, addresses, etc. (career, recruiter, cs, accountant, lawyer lines) — masking digit rules, the principle of prohibiting re-output of outputs, and notice conditions for overseas transfer under the Personal Information Protection Act

---

Classification: Post-processing Skill (Text Output Quality)