メニュー
Detect AI-generated writing patterns (slop) in content. Use when auditing drafts, reviewing AI output, or checking content authenticity before publishing.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | slop-detector |
| description | Detect AI-generated writing patterns (slop) in content. Use when auditing drafts, reviewing AI output, or checking content authenticity before publishing. |
| use_when | User wants to detect AI slop in content, audit a draft for AI patterns, check writing authenticity, review AI-generated output, or verify content before publishing. |
| tools | ["Read","Edit","Write"] |
The goal isn't perfect avoidance of any single phrase—it's writing that sounds like a specific human with specific opinions, not a very polite committee trying not to offend anyone.
<quick_start> Paste or point to content you want to audit. The skill will:
<fix_mode> Direct Edit Mode (Default Behavior)
This skill is an EDITOR, not a critic. After detection:
-CLEAN version if requested)Fix Application Order:
Example Transformations:
Output After Fixes:
## Slop Detection Report
**Fixes Applied:** 7
**Slop Score:** 12 → 2 (Medium → Low)
### Changes Made
| Line | Before | After |
|------|--------|-------|
| 3 | "Let's delve into the details" | "Here are the details" |
| 15 | "Game-changing approach" | "Different approach" |
...
### Remaining Considerations
- [Any issues requiring human judgment]
Audit-Only Mode: If user explicitly requests "audit only" or "don't edit", provide the detection report without making changes. </fix_mode>
<the_30_second_test> The fastest way to identify generic AI content:
"Could this exact content be sent to anyone in my industry?"
If yes, it's AI slop. This applies to everything: emails, blog posts, social media, marketing copy, internal communications.
The test works because AI defaults to universal applicability—content that technically works for everyone but resonates with no one. Specificity separates genuine insight from generic templates. </the_30_second_test>
<detection_patterns> <tier_1 name="Almost Always AI - Remove Immediately"> These phrases are so strongly associated with AI that their presence alone suggests unedited AI output:
Artificial Enthusiasm:
Excessive Hedging:
Rhetorical Question Openers:
Corporate Jargon Clusters:
Research Evidence:
<tier_2 name="Suspicious - Check Context"> These are problematic when overused or clustered:
Repetitive Transitions:
Paired Adjectives:
Template Phrases:
<tier_3 name="Watch for Patterns - OK in Isolation"> These are fine individually but problematic in clusters:
Transition Words (when every paragraph starts this way):
Corporate Language:
<structural_tells> Sentence Uniformity: Every sentence 10-15 words. Short. Punchy. Exhausting. Real writing has rhythm—mix 5-word sentences for impact with 25-word sentences that explore implications.
Staccato Fragment Spam (CRITICAL - Often Missed): Three or more consecutive short declarative sentences (under 10 words each) stating facts in parallel structure. This is AI's version of a bulleted list pretending to be prose.
Examples to REJECT:
The FIX: Combine into flowing prose with commas, em-dashes (with spaces), or conjunctions:
Detection Rule: If you see 3+ consecutive sentences that are:
FLAG IT. This is one of the most common AI tells that slips past phrase-based detection.
Over-Balanced Sections: Every section same length. All paragraphs 3-4 sentences. AI doesn't have opinions, so it gives balanced coverage to everything. Real writing reflects priorities.
List Addiction: Everything becomes bullets, even when prose works better. Lists within lists. "Here are the 7 ways to..." opening every section.
Emoji Headers: 🎯 Goal / 💡 Key Insight / ✅ Action Item / 🚀 Next Steps Modern content marketing style that AI overuses.
Flattery Sandwiches: "While traditional methods have merit, modern approaches offer..." "Though conventional wisdom suggests X, evidence indicates..." AI tries to be diplomatic and balanced to avoid controversy.
Comparator Sentences (This Isn't X, It's Y):
Manufactured Personality / Fake Developer Voice:
<clinical_formality> AI defaults to academic formality because training data includes massive amounts of academic papers:
The Fix: If you wouldn't say it in conversation, don't write it. </clinical_formality> </detection_patterns>
<audit_process> Step 1: Run the 30-Second Test Read the opening paragraph. Could it be sent to anyone in the industry? If yes, flag immediately.
Step 2: Search for Tier 1 Phrases Scan for these exact terms:
If more than two Tier 1 phrases found: High probability of unedited AI output.
Step 3: Check Patterns
Step 4: Scan for Staccato Fragment Spam This is CRITICAL and often missed. Look for 3+ consecutive short sentences stating facts:
If found: FLAG IMMEDIATELY. This pattern is as telling as "delve" but flies under phrase-based radar.
Step 5: Assess Sentence Diversity Measure sentence length variation. AI tends toward uniformity (10-15 words per sentence). Human writing varies naturally (5 to 40+ words).
Step 6: Calculate Slop Score
Scoring:
<output_format> When reporting findings, use this structure:
## Slop Detection Report
**30-Second Test:** [PASS/FAIL] - [reason]
**Overall Slop Score:** [X] - [Low/Medium/High Risk]
### Tier 1 Findings (Remove Immediately)
- [phrase] at line/position [X] - Replace with: [suggestion]
### Tier 2 Findings (Check Context)
- [phrase] appears [N] times - [keep one / remove all / rephrase]
### Tier 3 Clusters
- [section name]: Found [list of phrases] - Consider varying transitions
### Structural Issues
- Sentence uniformity: [analysis]
- Section balance: [analysis]
- List overuse: [yes/no]
### Recommended Fixes
1. [Most critical fix]
2. [Second priority]
3. [Third priority]
### The Core Principle
Your voice is in the specificity, the opinions, the rough edges, and the rhythm. Protect those.
</output_format>
**Input:** "In today's digital landscape, it's worth noting that AI has become a game-changer for content creation. Let's delve into how you can leverage these cutting-edge solutions to unlock your potential. Moreover, by implementing these strategies, you'll discover that the results are truly revolutionary."Analysis:
Verdict: Unedited AI output. Rewrite from scratch focusing on specific claims and evidence.
**Input:** "Here's the thing about productivity tools: they only work if you use them consistently. The bottom line is that the best system is one you'll actually stick with. Here's the thing though—most people give up after a week."Analysis:
Fixes:
Analysis:
Verdict: Reads as authentic human writing. No changes needed.
**Input:** "The new model is impressive. Complex code ships in a single session. Documentation writes itself. Problems that would have taken a weekend now take an afternoon."Analysis:
The Problem: This pattern is AI's attempt to sound punchy and modern. But it's unmistakably artificial. Real prose flows with commas, conjunctions, and em-dashes. This is a bulleted list pretending to be a paragraph.
The Fix: "The new model is impressive — complex code ships in a single session, documentation practically writes itself, and problems that would have taken a weekend now take an afternoon."
One sentence. Same information. Actually reads like human writing.
Note: Em-dashes should always have spaces on either side — like this — not—like this.
Verdict: Combine fragments into flowing prose using commas, em-dashes, or conjunctions.
**Input:** "I run five different media management services on my home server. Five different web interfaces. Five different API endpoints. Five different browser tabs open whenever I want to check what's downloading.That got old fast.
So I built a tool that unifies all of them."
Analysis:
The Problem: This looks human because it avoids obvious AI phrases. But the rhythm is artificial—it's trying to sound like a developer blog post. Compare to more natural writing:
"I run my systems on a home server. It's a solid setup, but like most self-hosted stacks, it means another dashboard, another set of menus to navigate, another context switch when I'm in the middle of something else."
No manufactured punch. No snark. Just describes the situation.
Verdict: Rewrite without the performance. State facts, show examples, move on.
<advanced_metrics> For quantitative analysis, these patterns can be measured:
Lexical Diversity (Type-Token Ratio):
Hedging Frequency:
Sentence Length Variance:
Perplexity Score:
<success_criteria> The audit is complete when:
<core_principle> AI slop isn't about individual words—it's about patterns.
One "moreover" doesn't make content AI-generated. But "moreover" + "it's worth noting" + "delve into" + uniform sentence length + emoji headers + flattery sandwiches = obvious AI slop.
The goal is writing that sounds like a specific human with specific opinions, not a very polite committee trying not to offend anyone. </core_principle>