A Claude skill that audits text for patterns commonly found in AI-generated writing, based on Wikipedia's "Signs of AI Writing" detection guide.

This skill systematically audits text for linguistic and structural patterns that frequently appear in LLM-generated content, including:

• Inflated language ("stands as," "serves as," "testament to")
• AI-lexicon overuse ("delve," "landscape," "multifaceted," "transformative")
• Formulaic structures (not-only-but, rule-of-three parallelism)
• Superficial qualifiers ("-ing" phrases tacked onto facts)
• Structural tells (em/en dash usage, title case headings, inline bolding)
• Communication artifacts (knowledge cutoff references, collaborative phrases)
• Citation problems (vague attribution, malformed references)
• Abstract metaphor nouns ("flywheel," "north star," "scaffolding")
• Portable sentences that could appear in any other post or project's docs

The skill provides:

1. Detailed audit with tagged patterns and severity markers
2. Corrected text with flagged patterns removed while preserving meaning
3. Changelog documenting each correction
4. Voice check so a clean audit does not leave sterile, voiceless prose

This skill does not detect whether text was written by AI.

Here's why:

• Humans write this way too. Many flagged patterns appear in human writing, especially in academic, marketing, or formal contexts.
• AI can avoid these patterns. Well-prompted LLMs can produce text without any of these tells.
• Patterns ≠ Authorship. Finding these patterns means the text has characteristics common in AI writing—nothing more.

This is a style audit tool, not an authorship detection tool.

Use it to improve writing quality by removing formulaic patterns, not to determine if something was AI-generated.

1. Download the current ai-writing-audit.skill package
2. In Claude (claude.ai), go to your account settings
3. Upload the .skill file in the Skills section

For Codex or another agent that supports folder-based skills, use the repository's SKILL.md and references/ directory together.

Provide text and ask Claude to audit it:

Audit this text for AI writing patterns:

[your text here]

Claude first returns a detailed audit with tagged patterns:

## AUDIT

1. "stands as a testament" [INFLATED] [AI-LEX +H]
2. "delve into the intricacies" [AI-LEX +H]
3. "Not only does it provide clarity, but also" [NOT-ONLY-BUT]
...

— END AUDIT: 12 issues found —

When issues are found, you'll receive:

• Complete audit with tagged issues
• Corrected text with patterns removed
• Changelog of all changes made

The skill checks for patterns across multiple categories:

Content Tells

• Inflated language and symbolism
• Promotional tone
• Superficial qualifiers
• Generic future outlook sections
• Portable prose (sentences that could appear in anyone else's post)

Language Tells

• AI-lexicon overuse (40+ flagged terms)
• Formulaic structures
• Elegant variation
• Vague attribution
• Abstract metaphor nouns (flywheel, north star, scaffolding, substrate)

Structural Tells

• Formatting patterns (title case, inline bold)
• Em/en dash usage
• Markdown/formatting artifacts

Communication Artifacts

• Direct address to readers
• Collaborative phrases ("Let me know if...")
• Knowledge cutoff references
• Letter-style formatting

Citation Issues

• Source emphasis patterns
• Malformed references
• Citation artifacts (oaicite, etc.)

See references/checklist.md for the complete detection checklist.

The implementation of AI technologies stands as a testament to innovation. 
Delving into the intricate landscape of machine learning, we can see how 
these transformative tools are not only enhancing productivity, but also 
fostering unprecedented collaboration—ultimately reshaping the future of work.

AI technologies show significant innovation. Machine learning tools are 
increasing productivity and enabling new forms of collaboration, changing 
how people work.

Flagged patterns removed:

• "stands as a testament" → deleted puffery
• "Delving into" → removed AI-lex
• "intricate landscape" → simplified
• "not only...but also" → broke formula
• "transformative," "fostering," "unprecedented" → replaced inflated language
• Em-dash → restructured sentence
• "ultimately reshaping the future" → cut vague claim

1. False positives happen. Some flagged patterns are perfectly fine in context—use judgment.
2. Not comprehensive. Patterns evolve faster than any checklist can track.
3. Context matters. Academic writing, legal documents, and formal reports may legitimately use some of these patterns.
4. Style preferences vary. What's flagged as "AI writing" might be appropriate for your audience or domain.

Always review audit results critically. This tool suggests improvements; you decide what to keep.

✅ Good use cases:

• Editing marketing copy for authenticity
• Reviewing documentation for formulaic language
• Improving academic writing clarity
• Catching Wikipedia-style AI tells before publication
• Training yourself to recognize these patterns

❌ Bad use cases:

• "Proving" a student's essay was AI-generated
• Content moderation decisions
• Automated AI detection systems
• Making accusations about authorship

Found a pattern that should be added? Have suggestions for the checklist? Open an issue or submit a pull request.

Detection patterns based on Wikipedia's Signs of AI Writing guide.

MIT License - See LICENSE file for details.

Remember: This tool identifies patterns, not authors. Use it to improve writing, not to judge writers.