Back to Docs
Humanizationv10.0

Humanization Pipeline

Transform AI patterns into natural academic prose while preserving scholarly integrity

Overview

The Humanization Pipeline detects and transforms AI-generated writing patterns into natural, human-sounding academic prose. It analyzes 28 pattern categories across 4 transformation layers (vocabulary, phrase, structure, discourse) while maintaining 100% citation accuracy, statistical precision, and scholarly rigor. Powered by Humanizer MCP v3.0 with 13 quantitative stylometric metrics. Orchestrated by the /diverga:humanize skill with mandatory human checkpoints at every stage.

100% Citation Preserved
100% Statistics Preserved
Academic Rigor Maintained

3-Agent + MCP Pipeline

G5

Academic Style Auditor

AI Pattern DetectionSonnet
  • •Detects 28 AI writing pattern categories across 7 domains
  • •AI probability scoring (0-100%)
  • •Risk classification (High/Medium/Low)
  • •Section-specific detection
  • •13 quantitative stylometric metrics via Humanizer MCP
  • •v3.0 composite scoring (6-component formula)
G6

Academic Style Humanizer

Pattern TransformationOpus
  • •Pattern-by-pattern transformation
  • •100% citation preservation
  • •100% statistical value preservation
  • •Three transformation modes
  • •4-layer transformation (vocabulary → phrase → structure → discourse)
  • •Discourse-level DT1-DT4 strategies
F5

Humanization Verifier

Quality AssuranceHaiku
  • •Citation integrity verification
  • •Statistical accuracy checking
  • •Meaning preservation validation
  • •AI pattern reduction measurement
  • •8 verification domains including discourse naturalness
  • •v3.0 composite score regression checking

28 AI Pattern Categories

Organized into 7 major domains for comprehensive detection

Content Patterns (C1-C6)

  • C1: Significance Inflation - "groundbreaking", "revolutionary"
  • C4: Promotional Language - "impressive", "remarkable"
  • C5: Vague Attributions - "some researchers suggest"

Language Patterns (L1-L6)

  • L1: AI Vocabulary - "delve", "crucial", "leverage", "foster"
  • L2: Copula Avoidance - "serves as" instead of "is"
  • L6: False Ranges - "challenges and opportunities"

Style Patterns (S1-S6)

  • S1: Em Dash Overuse - Multiple — dashes — per paragraph
  • S3: Inline-Header Lists - "First, ... Second, ... Third, ..."
  • S5: Emoji Usage - Academic text with emojis

Communication (M1-M3)

  • M1: Meta-Commentary - "As an AI language model..."
  • M2: Excessive Affirmation - "Great question!", "Absolutely!"
  • M3: Apology Hedging - "I apologize if this..."

Hedging (H1-H3)

  • H1: Realm Language - "in the realm of", "landscape of"
  • H2: Excessive Hedging - "could potentially perhaps suggest"
  • H3: Redundant Intensifiers - "very unique", "extremely essential"

Academic-Specific (A1-A6)

  • A1: Overclaiming - Causal language without evidence
  • A4: Methods Boilerplate - Generic methodology descriptions
  • A6: Implications Inflation - "profound implications for society"

Structural Patterns (S7-S10)

  • S7: Enumeration as Prose - Lists hidden in paragraph form
  • S8: Repetitive Paragraph Openers - Same opening pattern across paragraphs
  • S9: Formulaic Section Structure - Predictable IMRaD layout
  • S10: Hypothesis Checklist - Systematic hypothesis-by-hypothesis reporting

Three Transformation Modes

Choose the right level of transformation for your context

Conservative

Target:High-risk patterns only
Reduction:20-35%
Text Change:5-15%
Layers:Layer 1-2 (vocabulary + phrase)

Best for: Journal submissions, formal publications

Balanced

⭐
Target:High + medium-risk patterns
Reduction:35-50%
Text Change:15-30%
Layers:Layer 1-3 (vocabulary + phrase + structure)

Best for: Most academic writing (Recommended)

Aggressive

Target:All patterns
Reduction:50-70%
Text Change:30-50%
Layers:Layer 1-4 (vocabulary + phrase + structure + discourse)

Best for: Blog posts, informal writing

May affect scholarly tone

Pipeline Flow

G5 Pattern Analysis + MCP Metrics
Pass 1: Lexical (Layer 1-2)
🟠 CP_PASS1_REVIEW
Pass 2: Structural (Layer 3)
🟡 CP_PASS2_REVIEW
Pass 3: Discourse (Layer 4)
🟠 CP_PASS3_REVIEW
F5 Final Verification
🟢 CP_FINAL_REVIEW

Absolute Preservation Rules

These elements are NEVER transformed

  • Citations: (Author, year) format preserved exactly
  • Statistics: p-values, effect sizes, CIs, N values unchanged
  • Direct quotes: Verbatim quoted text maintained
  • Technical terms: Field-specific terminology kept
  • Mathematical expressions: Formulas and equations preserved
  • Proper nouns: Names, places, organizations unchanged

Ethics Framework

"Humanization improves expression, not deception."

✅ DOES

  • •Help express ideas naturally
  • •Remove robotic phrasing
  • •Maintain scholarly tone while improving readability
  • •Remove obvious AI artifacts

❌ DOES NOT

  • •Make AI use "undetectable"
  • •Replace need for AI disclosure
  • •Generate original ideas
  • •Substitute for human judgment

Quick Commands

"Check AI patterns"Run G5 analysis only
"Humanize my draft"Full pipeline (balanced mode)
"Humanize (conservative)"Minimal changes
"Humanize (aggressive)"Maximum naturalness
"Export with humanization"Pipeline before export

Transform Your Academic Writing

The Humanization Pipeline is built into Diverga's G5-AcademicStyleAuditor, G6-AcademicStyleHumanizer, and F5-HumanizationVerifier agents.

Explore Agents