Humanize AI Text#

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written.

Based on Wikipedia's comprehensive "Signs of AI writing" guide.

Features#

• AI Detection: Identifies 16 pattern categories of AI-generated text

• Pattern Categories:

  • Citation bugs
  • Knowledge-cutoff phrases
  • Chatbot artifacts
  • Markdown/code remnants
  • AI-specific vocabulary
  • Filler phrases
  • Punctuation quirks
  • Stylistic parallelisms
  • Inflated symbolism
  • Promotional language
  • Superficial -ing analyses
  • Vague attributions
  • Em dash overuse
  • Rule of three
  • Negative parallelisms
  • Excessive conjunctive phrases

• Diagnostic Reports: Generate detailed AI detection reports

• Text Transformation: Rewrite text to sound more human

• Comparison Mode: Compare original vs humanized versions

Usage#

Detect AI Writing#

bash

# Basic detection
python skills/humanize-ai-text/scripts/detect.py text.txt

# JSON output
python skills/humanize-ai-text/scripts/detect.py text.txt -j

# Score only
python skills/humanize-ai-text/scripts/detect.py text.txt -s

# Pipe text directly
echo "This text might be AI-generated" | python skills/humanize-ai-text/scripts/detect.py

Transform Text#

bash

# Basic transformation
python skills/humanize-ai-text/scripts/transform.py text.txt

# Output to file
python skills/humanize-ai-text/scripts/transform.py text.txt -o clean.txt

# Aggressive mode (more changes)
python skills/humanize-ai-text/scripts/transform.py text.txt -a

# Quiet mode (no progress output)
python skills/humanize-ai-text/scripts/transform.py text.txt -q

Compare Versions#

bash

# Compare original and transformed
python skills/humanize-ai-text/scripts/compare.py text.txt

# Compare and save result
python skills/humanize-ai-text/scripts/compare.py text.txt -o document_v2.txt

Installation#

The skill is already installed. To use it, ensure you have Python 3 installed.

Dependencies#

No external dependencies required - uses Python standard library.

Pattern Detection#

The detection script identifies these AI writing patterns:

1. Citation Bugs#

• Fake or unverifiable sources
• "According to experts" without specifics
• Generic study references

2. Knowledge Cutoff#

• "As of my last update"
• "I don't have real-time information"
• Date-stamped limitations

3. Chatbot Artifacts#

• "I'd be happy to help"
• "Great question!"
• "Let me break this down"
• Excessive hedging

4. AI Vocabulary#

• "delve into"
• "testament to"
• "in conclusion"
• "it's worth noting"
• "important to understand"

5. Filler Phrases#

• Unnecessary introductions
• Redundant summaries
• Padding content

6. Stylistic Issues#

• Rule of three overuse
• Negative parallelisms
• Em dash overuse
• Excessive conjunctive phrases

Best Practices#

1. Detect First: Run detection before transformation
2. Review Changes: Always review transformed text
3. Use Compare: Compare versions to understand changes
4. Iterate: May need multiple passes for best results
5. Custom Patterns: Edit patterns.json to add custom detections

Customization#

Add Custom Patterns#

Edit skills/humanize-ai-text/scripts/patterns.json:

json

{
  "custom_patterns": {
    "my_pattern": {
      "regex": "pattern_here",
      "replacement": "replacement_text",
      "description": "What this pattern detects"
    }
  }
}

Batch Processing#

Process multiple files:

bash

for f in *.md; do
  python skills/humanize-ai-text/scripts/detect.py "$f" -s
  python skills/humanize-ai-text/scripts/transform.py "$f" -a -o "${f%.md}_clean.md" -q
done

Output Formats#

Detection Output#

text

AI Detection Report
===================
File: text.txt
AI Probability: 73%

Patterns Detected:
- AI Vocabulary: 12 instances
- Filler Phrases: 8 instances
- Stylistic Parallelisms: 5 instances
...

Transformation Output#

text

Original: "I'd be happy to delve into this important topic."
Humanized: "Let's explore this topic."

Security Notes#

• ✅ No external network calls
• ✅ All processing done locally
• ✅ No credential access
• ✅ Safe file operations only
• ✅ No data exfiltration

Integration#

With Writing Workflow#

1. Write draft (AI-assisted or not)
2. Run detection to check AI score
3. Transform if score is high
4. Compare versions
5. Manual review and final edits

With Content Pipeline#

bash

# In CI/CD or pre-commit hook
python skills/humanize-ai-text/scripts/detect.py content.md -s
# Fail if AI score > threshold

Troubleshooting#

False Positives#

Some legitimate writing may trigger patterns. Review manually.

False Negatives#

AI models evolve. Update patterns.json regularly.

Encoding Issues#

Ensure files are UTF-8 encoded:

bash

python skills/humanize-ai-text/scripts/detect.py file.txt --encoding utf-8

Examples#

Example 1: Blog Post#

bash

# Check AI score
python skills/humanize-ai-text/scripts/detect.py blog_post.md -s
# Output: AI Probability: 68%

# Transform
python skills/humanize-ai-text/scripts/transform.py blog_post.md -o blog_post_clean.md -a

# Compare
python skills/humanize-ai-text/scripts/compare.py blog_post.md -o blog_post_clean.md

Example 2: Email#

bash

echo "I hope this email finds you well. I wanted to reach out regarding..." | \
  python skills/humanize-ai-text/scripts/detect.py

Example 3: Academic Writing#

bash

# Detect AI patterns in paper
python skills/humanize-ai-text/scripts/detect.py paper.md -j > report.json

# Transform with conservative settings
python skills/humanize-ai-text/scripts/transform.py paper.md -o paper_clean.md

Resources#

• Wikipedia: "Signs of AI writing"
• Original research on AI text detection
• Community pattern contributions

---

For more info, visit: https://clawhub.ai/biostartechnology/humanizer