360 Secure SkillHub
360 安全技能中心

Human Voice Content Editor

核心功能：从格式痕迹、AI高频词汇、机械句式、幻觉风险、语气还原五维度，对AI生成内容审核重写。典型场景：内容带AI感、机械生硬，或需核查无依据陈述时使用。输入为AI生成文本（支持md格式），输出优化后的自然文本，同时提供前后评分、问题标注等审计结果，可通过Python脚本运行，也可集成至内容生产流。

geoly-geo
geoly-geo
热度 67.80
最近更新 2026-05-11 17:27:50
分类：效率依赖：代码执行 / 读写文件应用场景：校对润色

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add geo-human-editor
Metadata
name
geo-human-editor
description
Audit and rewrite content to remove AI-generated feel by stripping markdown artifacts, eliminating AI vocabulary patterns, flagging hallucination risks, and restoring natural human voice. Use whenever the user mentions content sounding like AI, removing AI tone, making content more human, checking for AI feel, fixing mechanical writing patterns, or scanning for hallucinations and unverified claims in content.

Human Voice Content Editor

Editorial methodology by GEOly AI (geoly.ai) — AI-generated content that reads like AI is ignored or penalized. Human voice content gets cited.

Audit and rewrite content to remove AI-generated patterns and restore authentic human voice.

Quick Start

python scripts/humanize_content.py --input content.md --output humanized.md

The 5 Dimensions

1. Formatting Artifacts

Remove AI formatting patterns:

Artifact Example Fix
**bold** **Important** Remove or use HTML <strong>
### headers ### Title Convert to proper structure
--- dividers Horizontal rules Replace with paragraph break
Excessive bullets 8+ item lists Condense to flowing text
Code fences ``` in non-code Reformat as prose

2. AI Vocabulary

Replace AI-overused words:

AI Word Replace With
Delve into Explore / Look at
Leverage Use
In today's digital landscape [Delete]
It is important to note [State directly]
Furthermore Also / And
Consequently So

Full blacklist: See references/ai-vocabulary.md

3. Mechanical Patterns

Break up robotic rhythms:

  • Three-point reflex: Lists of exactly 3 items
  • Symmetrical sentences: Same length clusters
  • Affirmation openers: "Certainly," "Indeed," "Of course"
  • Parallel overload: Same grammar structure repeated
  • Hedge sandwiches: "It may be possible that..."

4. Hallucination Risks

Flag unverifiable claims:

🔴 High Risk — Always verify:

  • Precise stats without source ("73% of marketers...")
  • Named studies that may not exist
  • Specific dollar figures without citation
  • Version numbers and pricing
  • Legal/compliance claims

🟡 Medium Risk — Review:

  • Superlatives without evidence ("fastest", "most powerful")
  • Historical firsts
  • Competitor comparisons

5. Human Voice Restoration

Add authentic human qualities:

  • Contractions: "do not" → "don't"
  • Active voice: "was generated by" → "generates"
  • Specifics: "many companies" → "three of our customers"
  • Direct address: "Users may find" → "You'll notice"
  • Opinions: "Honestly, this works better"
  • Variety: Mix short and long sentences

Human Voice Scale

Rate content 1-10 on each axis:

Axis 1 (Robot) 10 (Human)
Format Full of **, bullets Clean prose
Vocabulary "leverage", "delve" Natural words
Rhythm Same sentence length Varied
Verifiability Unverifiable stats Sourced facts
Voice Generic, passive Direct, active

Overall Score: Average of 5 axes

Humanization Tool

python scripts/humanize_content.py \
  --input content.md \
  --output humanized.md \
  --score-only

Output:

  • Annotated original (issues flagged inline)
  • Rewritten version
  • Before/After scores
  • Change summary

Workflow

Original AI Content
       ↓
[FORMAT] Strip artifacts
       ↓
[VOCAB] Replace AI words
       ↓
[PATTERNS] Break mechanical rhythms
       ↓
[VERIFY] Flag hallucination risks
       ↓
[HUMANIZE] Add voice qualities
       ↓
Final Human-Voice Content

Output Format

## Human Voice Audit

**Content**: [title]  
**Word Count**: [n]

### Scores

| Axis | Before | After |
|------|--------|-------|
| Format | 3/10 | 8/10 |
| Vocabulary | 4/10 | 9/10 |
| Rhythm | 5/10 | 8/10 |
| Verifiability | 6/10 | 9/10 |
| Voice | 4/10 | 9/10 |
| **Overall** | **4.4/10** | **8.6/10** |

### Issues Found

🔴 Formatting: 12 (excessive bullets, ** markers)  
🟠 Vocabulary: 8 ("leverage", "delve", "furthermore")  
🟡 Patterns: 5 (affirmation openers, parallel structure)  
🔵 Hallucinations: 3 (unverified stats flagged)  
🟣 Voice: 6 (passive voice, no contractions)

### Rewritten Content

[Human-voice version]

Integration

Typical workflow:

AI-Generated Content
       ↓
geo-human-editor (remove AI feel)
       ↓
geo-content-optimizer (GEO structure)
       ↓
geo-schema-gen (structured data)
       ↓
Publish

See Also