humanizer

name

humanizer

description

The skill for "is this AI-written?" or "make this sound more human." Use whenever a specific text sample is in question and the core concern is AI authenticity — student submissions suspected of AI authorship, blog/wiki drafts that feel off, PR commit messages that smell like boilerplate, cover letters that sound robotic, paragraphs an editor flagged. Detects copula avoidance, ChatGPT vocabulary, significance inflation, and 30+ other documented AI tells. Also rewrites flagged sections. Not for grammar checks, summarization, or style edits where AI detection is not the concern.

Humanizer: AI Writing Pattern Detection

Detect AI-generated text using Wikipedia's documented patterns from "Signs of AI writing". Provides systematic analysis across 35+ specific pattern categories with confidence levels and concrete suggestions. Patterns are backed by academic research (Kobak et al., Reinhart et al., Juzek & Ward, Russell et al.).

Before analyzing: Read references/patterns.md — it contains the full pattern database organized by category.

Caveats

• Patterns are indicators, not proof. Use confidence levels.
• Each AI model has a distinctive idiolect; patterns typical for ChatGPT may not apply to Gemini or Claude.
• Expert LLM users correctly identify AI text ~90% of the time, meaning ~10% false positive rate even for experts.
• Do not rely solely on automated AI detection tools (GPTZero, etc.) — they have non-trivial error rates and can be fooled by paraphrasing.

When to Use

ALWAYS use when:

• User invokes /humanizer command (even for short/simple text - user explicitly requested analysis)
• User asks to check for AI patterns or analyze text for AI writing

Don't use for:

• Grammar checking, summarization, or style improvements where AI detection is not the concern
• Obviously human text when NOT explicitly requested (e.g., short factual announcements)

Detection Strategy

1. Read references/patterns.md — load the full pattern database before scanning.

2. Scan high-confidence patterns first:

   • Content: significance inflation, notability/attribution emphasis (incl. named-source misattribution), superficial analysis, promotional language, vague attributions, challenges/prospects formula, title-as-proper-noun leads
   • Language: AI vocabulary (check era clustering), copula avoidance, elegant variation
   • Markup: chatbot citation artifacts, ChatGPT attribution markup, hallucinated citations/references
   • Communication: collaborative chatbot phrases, knowledge-cutoff disclaimers, placeholder text

3. Check for pattern clusters — multiple patterns in the same sentence = higher confidence. "Additionally" + "serves as" + "pivotal" in one sentence is very likely AI. Look for present participle ("-ing") chains at sentence ends.

4. Verify with context — is the pattern inappropriate for this text type? Would a human expert in this domain write this way? Consider the specific AI model's idiolect.

5. Give concrete suggestions — don't just flag. Show the specific rewrite. "The building is a community center" not "serves as a community center."

Output Format

## Analysis Results

**Text analyzed**: [word count] words

### High Confidence Patterns (N)
1. **Pattern Name** - "trigger phrase" (location)
   Text: "...exact quote..."
   Issue: [Why it's flagged — specific to this instance]
   Suggestion: [Concrete alternative for this specific case]

### Medium Confidence Patterns (N)
[Same format, note these may be stylistic choices]

### Low Confidence / Context-Dependent (N)
[Same format, explicitly note these need human judgment]

### Summary
- **Total findings**: N patterns across M categories
- **Confidence breakdown**: X high, Y medium, Z low
- **Recommendation**: [Prioritization guidance]

Ineffective Indicators

Do NOT flag these (false positives):

• Perfect grammar or spelling
• Bland or robotic prose style (not specific to AI)
• Academic or formal vocabulary (domain-appropriate in many contexts)
• Conjunctions at sentence starts (normal English)

Common Mistakes

• Flagging every instance: Consider density and context — one "additionally" may be fine, five in a paragraph is suspicious.
• Overclaiming: Patterns are indicators, not proof. Never say "definitely AI."
• Generic suggestions: Show the actual rewrite, not "be more specific."
• Missing clusters: When you find one pattern, check the surrounding sentences for others.
• Ignoring context: "Vibrant" in marketing copy is fine; in an encyclopedia entry it's suspicious.
• Over-weighting historical patterns: "It's important to note" was a strong tell in 2023; newer models use it rarely. Downweight accordingly.

Updating Patterns

Wikipedia's "Signs of AI writing" article evolves as AI writing changes. To update:

1. Browse: https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
2. Run update-patterns.sh to fetch latest wikitext (curl required — WebFetch gets 403'd by Wikipedia)
3. Review changes and update references/patterns.md
4. Update last_updated below

Last updated: 2026-04-15 (Wikipedia revision: 2026-04-15T23:42:51Z)