How the Editor Agent Transforms AI Content to Sound Human: A Technical Deep Dive

The Editor agent transforms AI-generated drafts into human-sounding content by scanning for robotic red flags, scoring the text on a multi-dimensional humanity rubric, and systematically applying a seven-step transformation checklist that injects personality and concrete details while preserving SEO value.

The Editor agent is a specialized Claude agent within the TheCraigHewitt/seomachine repository that serves as the final quality gate in an automated content pipeline. Unlike simple paraphrasing tools, this agent performs sophisticated pattern detection to identify how the Editor agent transforms AI content to sound human, applying deterministic rules to eliminate statistical markers of machine generation while maintaining factual accuracy and search optimization.

Pattern Detection: How the Editor Identifies Robotic Content

The agent's first phase involves a comprehensive scan for statistical signatures that distinguish AI-generated text from human writing. This detection framework is defined in .claude/agents/editor.md and cross-referenced against the detection database in .claude/skills/seo-audit/references/ai-writing-detection.md.

Robotic Red Flags vs. Human Green Flags

The agent categorizes textual features into two opposing taxonomies:

Detection Area	Robotic Red Flags	Human Green Flags
Phrasing	Generic openings ("In today's digital landscape"), over-used transition words, corporate buzzwords ("leverage", "synergy")	Specific anecdotes, direct address ("you've probably noticed..."), parenthetical asides
Punctuation	Excessive em-dashes, predictable comma patterns	Varied punctuation, occasional fragments
Syntax	Passive-voice dominance, repetitive sentence structures	Mix of short punchy sentences and longer flowing ones, rhetorical questions
Specificity	Vague temporal markers ("recently"), generic entities ("many businesses")	Concrete dates ("in March 2024"), precise statistics ("73% of SaaS companies")

The AI Writing Detection Reference

The detection logic relies on a structured reference file that catalogs AI-specific phrases and filler words. When the Editor agent scans content, it treats any occurrence of patterns listed in .claude/skills/seo-audit/references/ai-writing-detection.md as a Robotic Red Flag, automatically triggering the transformation checklist for those sections.

The Multi-Dimensional Scoring Model

Before rewriting, the Editor calculates a composite humanity score using five weighted dimensions defined in the "Scoring Dimensions" section of .claude/agents/editor.md (lines 49-58):

Dimension	Weight	Evaluation Criteria
humanity	30%	AI phrase frequency, passive voice ratio, contraction usage, conversational devices
specificity	25%	Concrete examples count, numerical data points, named entities
structure_balance	20%	Prose-to-list ratio, paragraph length variation, heading hierarchy
seo	15%	Keyword placement naturalness, heading optimization, meta description quality
readability	10%	Flesch Reading Ease score, sentence length variety

The agent outputs these scores in a structured JSON block (lines 409-447 of the agent definition) that includes a boolean passed flag and a list of priority_fixes with specific locations and recommended transformations.

The Transformation Checklist: From AI to Human

The Editor applies a deterministic seven-step transformation checklist to rewrite the draft:

1. Show, Don't Tell

Replace abstract statements with vivid anecdotes. For example, instead of "Many businesses struggle with SEO," the agent injects: "Last Tuesday, a SaaS founder told me they'd spent $40K on content that generated exactly zero leads."

2. Inject Personality

Add humor, rhetorical questions, second-person address ("you"), and informal connectors ("Look," "Here's the thing," "Honestly").

3. Kill Corporate Speak

Replace buzzwords using the mapping defined in the AI detection reference:

• "leverage" → "use"
• "synergy" → "working together"
• "moving forward" → "from now on"

4. Add Specific Details

Replace vague temporal markers with concrete dates and generic quantities with precise statistics. The agent scans for words like "recently," "many," "some," and "often" to flag for specificity upgrades.

5. Vary Sentence Structure

Mix short punchy sentences (under 8 words) with longer, flowing compound-complex structures to create rhythmic variation.

6. Use Conversational Devices

Insert parentheticals, sentence fragments, and direct address to mimic spoken language patterns.

7. Make Lists Actionable

Convert generic bullet points into narrative steps with context, transforming "Optimize titles" into "Start by rewriting your title tags—this is often the fastest win because..."

Machine-Readable Output: The JSON Report

The Editor returns both a human-readable editorial report and a machine-readable JSON block that enables automated quality loops. The JSON structure (defined in lines 409-447 of .claude/agents/editor.md) includes:

{
  "scores": {
    "humanity": 87,
    "specificity": 92,
    "structure_balance": 78,
    "seo": 95,
    "readability": 88
  },
  "composite_score": 87.4,
  "passed": true,
  "priority_fixes": [
    {
      "location": "Introduction, paragraph 2",
      "dimension": "humanity",
      "issue": "Passive voice dominance",
      "fix": "Change 'The tool was used by marketers' to 'Marketers used the tool'"
    }
  ]
}

You can parse this output programmatically to trigger conditional workflows:

import json

# Assume `response` is the raw text returned by the editor agent

json_start = response.find('{')
json_block = response[json_start:]

report = json.loads(json_block)

humanity = report["scores"]["humanity"]
passed   = report["passed"]
fixes    = report["priority_fixes"]

print(f"Humanity score: {humanity} – {'✅ Pass' if passed else '❌ Fail'}")
for f in fixes:
    print(f"- {f['location']} ({f['dimension']}): {f['issue']} → {f['fix']}")

Integration with the Content Pipeline

The Editor agent operates as the final stage of the /write command pipeline in the seomachine repository. According to the README's "Specialized Agents" section, the workflow chains together:

1. Research Agent – gathers topical authority sources
2. SEO Optimizer – structures headings and keywords
3. Writer Agent – produces the initial draft
4. Editor Agent – performs the humanity transformation described in this article

When the Editor returns a passed: false flag in its JSON output, the pipeline can automatically route the draft back to the Writer Agent with specific fix instructions, creating an automated quality loop without manual intervention.

Summary

• The Editor agent transforms AI content to sound human by scanning for robotic red flags (generic openings, passive voice, buzzwords) and human green flags (specific anecdotes, conversational devices).
• It scores content on five weighted dimensions: humanity (30%), specificity (25%), structure_balance (20%), seo (15%), and readability (10%).
• The transformation checklist includes seven deterministic steps: show don't tell, inject personality, kill corporate speak, add specific details, vary sentence structure, use conversational devices, and make lists actionable.
• Output includes both a human-readable editorial report and a machine-readable JSON block with scores, pass/fail flags, and prioritized fixes for automated pipeline integration.
• All logic is defined in .claude/agents/editor.md and cross-referenced with .claude/skills/seo-audit/references/ai-writing-detection.md.

Frequently Asked Questions

How does the Editor agent detect AI-generated patterns in content?

The Editor agent detects AI-generated patterns by scanning for robotic red flags defined in the AI writing detection reference file. It specifically looks for generic openings like "In today's digital landscape," over-used transition words, excessive em-dashes, passive-voice dominance, corporate buzzwords like "leverage" or "synergy," and repetitive sentence structures. These patterns are statistical markers commonly found in LLM output.

What is the scoring rubric used to evaluate content humanity?

The Editor uses a five-dimensional weighted rubric: humanity (30%) measures AI phrase frequency and conversational devices; specificity (25%) counts concrete examples and data points; structure_balance (20%) evaluates prose-to-list ratios; seo (15%) checks keyword naturalness; and readability (10%) calculates Flesch scores and sentence variety. The composite score determines whether the content passes or requires revision.

Can the Editor agent's output be integrated into automated workflows?

Yes, the Editor returns a structured JSON block containing numerical scores, a boolean passed flag, and a priority_fixes array with specific locations and recommended corrections. This machine-readable format allows the /write command pipeline to automatically route failing drafts back to the Writer Agent with specific instructions, creating a closed-loop quality system without manual intervention.

Where are the Editor agent's rules and detection patterns defined?

The Editor's core logic, transformation checklist, and scoring dimensions are defined in .claude/agents/editor.md. The supplementary detection patterns for AI-specific phrases and robotic markers are cataloged in .claude/skills/seo-audit/references/ai-writing-detection.md. These files work together to provide the Editor with both the strategic framework and the tactical pattern library needed to humanize content.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}