Humanizer: AI Pattern Detection & Voice Injection
Transform AI-generated text into human writing by detecting patterns and injecting authentic voice.
Before You Edit: Diagnostic Framework
Ask yourself these 3 questions BEFORE applying any patterns:
1. Voice Assessment
- Does this text have a distinct voice? Or is it neutral/corporate?
- What personality should come through? (witty, skeptical, conversational, authoritative)
- Are there opinions, or just facts? Human writing has stakes and perspective.
2. Pattern Prioritization
- Which 3-5 patterns dominate this text? (Don't fix everything at once)
- What's the writer's intent? (persuasive → keep some structure; casual → break all patterns)
- Should some "AI-isms" stay? (Formal technical docs may keep certain structures)
3. Rewrite Philosophy
- Am I removing patterns OR injecting personality? (Must do both)
- Does my rewrite sound like a specific human wrote it? (Not just "less AI")
- Have I varied sentence rhythm? (Short. Longer flowing sentences. Mix it up.)
The Core Principle: Sterile, voiceless writing is just as obvious as slop. Don't just remove bad patterns—add soul.
4. Pattern Detection Procedure (Domain-Specific)
Run these checks BEFORE editing:
Statistical Density Check:
- Count AI vocabulary words per 100 words: >3 = heavy AI signature
- Count em dashes per paragraph: >2 = structural tell
- Count "However" paragraph starts: >20% = AI transition overuse
Structural Signature Check:
- All paragraphs same length? = AI rhythm uniformity
- Every list has exactly 3 items? = rule of three addiction
- Conclusions use passive voice? = AI hedging pattern
Context-Specific Preservation:
- Academic: Keep formal structure, remove only vocabulary slop
- Technical: Preserve precision terminology, remove promotional language
- Marketing: Full humanization except brand voice requirements
Critical Anti-Patterns (NEVER Do This)
❌ Pattern #1: Mechanical Pattern Removal
Problem: Just deleting AI phrases without adding human voice produces sterile text.
markdown
❌ BAD EDIT:
"The framework serves as a testament to modern development practices."
→ "The framework is modern."
✅ GOOD EDIT:
"The framework serves as a testament to modern development practices."
→ "This framework gets it right. Clean APIs, sensible defaults, actual documentation."
Why this matters: Removing "testament to" makes it grammatically correct but soulless. The good edit has opinion, rhythm, and personality.
❌ Pattern #2: Over-Correction
Problem: Making every sentence "unpredictable" creates chaos, not humanity.
markdown
❌ BAD EDIT (too chaotic):
"Results. Interesting ones! The experiment? It generated code—lots of it.
3 million lines worth. Developers (some of them) were impressed!!!!"
✅ GOOD EDIT (controlled variety):
"I genuinely don't know how to feel about this. 3 million lines of code,
generated overnight. Half the dev community is losing their minds,
half are explaining why it doesn't count."
Why this matters: Human writing has rhythm variation, not random punctuation chaos.
❌ Pattern #3: Removing ALL Structure
Problem: Not all AI patterns are bad—some formal writing needs structure.
markdown
Context: Academic paper abstract
❌ BAD EDIT:
"Our study looked at machine learning. We found some stuff.
It's interesting. Check out our results."
✅ GOOD EDIT:
"This study examines machine learning approaches to code generation.
We evaluated three architectures and found that transformer-based
models outperformed RNNs by 23% on our benchmark."
Why this matters: Formal contexts need clarity over personality. Know your audience.
❌ Pattern #4: Batch-Replacing AI Words Without Context
Problem: Blindly replacing "delve" or "landscape" breaks legitimate usage.
markdown
Context: Computer vision paper
❌ BAD EDIT:
"Our model examines the feature landscape" → "Our model examines the feature terrain"
✅ GOOD EDIT:
"Our model examines the feature landscape" → "Our model analyzes feature space"
OR keep "landscape" if it's established terminology in CV papers
Why this matters: Not every AI word is wrong—check if it's domain-appropriate first. "Landscape" in data science ≠ "business landscape" slop.
Most Common AI Patterns (Quick Reference)
Content-Level Patterns
Undue Emphasis on Significance
- Words: stands as, serves as, testament to, pivotal, crucial, underscores, broader trends
- Fix: Remove inflated symbolism, state facts directly
Promotional Language
- Words: boasts, nestled, vibrant, rich heritage, breathtaking, stunning
- Fix: Replace adjectives with specific details
Vague Attributions
- Words: Industry reports, Observers note, Experts argue, Some critics
- Fix: Name specific sources or remove the claim
Language-Level Patterns
AI Vocabulary Words (post-2023 frequency spike)
- Words: delve, crucial, enhance, foster, garner, intricate, landscape (abstract), pivotal, showcase, tapestry (abstract), underscore
- Fix: Use plain synonyms or restructure
Copula Avoidance (avoiding "is/are")
- Pattern: "serves as", "stands as", "represents", "boasts", "features"
- Fix: Use simple "is/are/has"
Negative Parallelisms
- Pattern: "Not only... but...", "It's not just about X, it's Y"
- Fix: State directly without forced contrast
Style-Level Patterns
Em Dash Overuse
- Pattern: Multiple em dashes in one paragraph (—)
- Fix: Replace with commas, periods, or parentheses
Rule of Three Overuse
- Pattern: "innovation, inspiration, and industry insights"
- Fix: Break groups of three, vary list sizes
Title Case Headings
- Pattern: "Strategic Negotiations And Global Partnerships"
- Fix: Sentence case: "Strategic negotiations and global partnerships"
Humanization Strategy: When to Preserve vs Remove
The Decision Framework: Not all contexts need full humanization.
| Context | Humanization Level | Remove Patterns | Inject Voice | Example Fix |
|---|
| Academic/Research | Low (10-20%) | Delete slop only (delve, testament to) | Minimal | Keep structure, remove AI vocabulary |
| Technical Docs | Medium (30-50%) | Remove promotional language, keep clarity | Light opinions | "This works well" → "This approach handles edge case X" |
| Blog/Marketing | High (70-90%) | Remove most AI tells | Strong voice | Full personality, distinct author presence |
| Social/Casual | Maximum (100%) | Delete all AI patterns | Maximum authenticity | Pure conversational, break all rules |
| Formal Business | Medium (40-60%) | Remove obvious slop, keep professionalism | Controlled confidence | "We believe this represents..." → "This delivers X" |
Critical Non-Obvious AI Tells (beyond the common list):
- Paragraph-starting "However": AI overuses this transition (appears 3x more in GPT text)
- Passive voice in conclusions: "It can be concluded that..." (AI hedges at the end)
- Symmetric sentence structure: Every paragraph follows same length/rhythm pattern
- "Importantly" mid-sentence: AI uses this more than humans (statistical quirk)
- Abstract "landscape" metaphors: "the technology landscape", "the business landscape"
When to Load Full Pattern References
For comprehensive pattern catalogs, use mandatory loading:
MANDATORY - READ ENTIRE FILE: references/content-patterns.md
- Text contains 5+ promotional adjectives (stunning, breathtaking, vibrant, rich)
- Significance inflation detected ("serves as testament", "stands as pivotal")
- Need complete "symbolism removal" examples
- Do NOT load for casual blog posts or social media text
MANDATORY - READ ENTIRE FILE: references/language-patterns.md
- Text uses 8+ AI vocabulary words (delve, showcase, intricate, foster, garner)
- Heavy copula avoidance patterns ("serves as" instead of "is")
- Need elegant variation catalog for substitutions
- Do NOT load for technical documentation where precision matters
MANDATORY - READ ENTIRE FILE: references/style-patterns.md
- Text has 6+ em dashes in single paragraph
- Rule of three appears 4+ times
- Title case headings throughout document
- Do NOT load for academic papers (formatting may be required)
Never load references for simple opinion injection or rhythm fixes—handle with decision framework above.
Process
- Read input text - Identify 3-5 dominant patterns
- Apply diagnostic framework - Answer the 3 questions above
- Make strategic edits - Fix patterns + inject voice simultaneously
- Verify rhythm - Read aloud test (does it sound natural?)
- Present result - Show rewritten text with brief summary if helpful
Quick Example
Before (AI-sounding):
The new software update serves as a testament to the company's commitment to innovation. Moreover, it provides a seamless, intuitive, and powerful user experience—ensuring that users can accomplish their goals efficiently. It's not just an update, it's a revolution in how we think about productivity.
After (Humanized):
The software update adds batch processing, keyboard shortcuts, and offline mode. Early beta feedback has been positive—most testers report finishing tasks faster.
What changed:
- Removed inflated symbolism ("serves as a testament")
- Removed AI vocabulary ("Moreover", "seamless, intuitive, powerful")
- Removed negative parallelism ("It's not just...it's...")
- Removed vague claims ("commitment to innovation")
- Added specific features (batch processing, shortcuts, offline)
- Added concrete evidence (beta feedback, faster completion)
- Kept neutral tone appropriate for feature announcement
Reference Materials
Based on
Wikipedia:Signs of AI writing, maintained by WikiProject AI Cleanup.
Key insight: "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases."
Translation: AI writing is optimized for average acceptability, not authentic voice.