ai-writing-avoidance git:20260728.80a7ab6A

Immutable. This exact content never changes and is served at /api/v1/blob/5a01ad4ed6d1db19.

---
name: ai-writing-avoidance
description: "Help writers produce content that sounds genuinely human by avoiding telltale AI-generated text patterns"
lastReviewed: 2026-04-30
---

# AI Writing Avoidance


**Domain**: Document quality assurance, policy writing, authentic voice preservation

## Purpose

Help writers produce content that sounds genuinely human by avoiding the telltale patterns that make AI-generated text identifiable. Essential for policy documents, professional communications, and any writing that needs to carry authentic voice and credibility.

## Quick Reference

### The Big Five Categories

1. **Vocabulary Tells** — Overused "sophisticated" words
2. **Structure Tells** — Predictable organization patterns
3. **Tone Tells** — Hedging and over-qualification
4. **Content Tells** — Generic examples, missing specificity
5. **Format Tells** — Mechanical use of bullets and headers

---

## Level 1: Vocabulary Tells (The Biggest Giveaway)

### Red Flag Words

AI models consistently overuse these words regardless of context:

| Category | Words to Avoid |
|----------|----------------|
| **Academic Flair** | delve, myriad, plethora, multifaceted, nuanced, paradigm |
| **Poetic Overreach** | tapestry, beacon, mosaic, symphony, landscape, realm |
| **Trendy Jargon** | liminal, holistic, synergy, leverage (as verb), optimize |
| **Empty Intensifiers** | crucial, vital, essential, pivotal, paramount, indispensable |
| **False Precision** | specifically, notably, particularly, fundamentally, inherently |
| **Corporate Speak** | cutting-edge, state-of-the-art, innovative, robust, seamless |
| **Transition Padding** | furthermore, moreover, additionally, consequently, thus |

### Replacement Strategy

Instead of: *"It's crucial to delve into the multifaceted tapestry of..."*

Write: *"We need to examine the different aspects of..."*

**Rule**: If a word sounds like it's trying too hard, it probably is.

---

## Level 2: Structure Tells

### The Preamble Problem

AI loves to announce what it's about to say:

❌ **AI Pattern**:
> "In this document, we will explore the key aspects of our new policy, examining its implications and providing guidance for implementation."

✅ **Human Pattern**:
> "Here's how the new policy works and what you need to do."

### The Enumeration Addiction

AI compulsively numbers and lists things even when prose would work better:

❌ **AI Pattern**:
> "There are three main benefits: 1) improved efficiency, 2) cost savings, and 3) better outcomes."

✅ **Human Pattern**:
> "It's faster, cheaper, and actually works."

### The Sandwich Structure

AI follows predictable patterns:

- Introduction that announces the topic
- Body with exactly 3-5 points
- Conclusion that restates everything

**Fix**: Vary your structure. Start in the middle. Skip the intro. End abruptly when you're done.

---

## Level 3: Tone Tells

### Hedging Language

AI is terrified of being wrong, so it qualifies everything:

| AI Hedging | Human Directness |
|------------|------------------|
| "It's worth noting that..." | (just say the thing) |
| "One might argue that..." | "Some people think..." |
| "It could be said that..." | (delete and state directly) |
| "In some cases, it may be..." | "Sometimes..." |
| "It's important to consider..." | (just consider it) |

### Over-Qualification

❌ **AI Pattern**:
> "While there are certainly many valid perspectives on this issue, and it would be presumptuous to claim definitive answers, it seems reasonable to suggest that..."

✅ **Human Pattern**:
> "I think..."

### The Neutral Voice Problem

AI writes like a diplomatic robot trying not to offend anyone:

- No strong opinions
- No personality
- No humor or frustration
- No admission of uncertainty

**Fix**: Have a point of view. Disagree with something. Acknowledge what you don't know.

---

## Level 4: Phrase Tells

### Overused Transitions

| Phrase | Frequency | Human Alternative |
|--------|-----------|-------------------|
| "It's important to note that" | Extremely high | (delete entirely) |
| "In conclusion" | Very high | (just conclude) |
| "Let's explore" | Very high | (just explore) |
| "Moving forward" | High | "Next" or nothing |
| "At the end of the day" | High | "Ultimately" or nothing |
| "When it comes to" | High | (delete, start with topic) |
| "In terms of" | High | "For" or "About" |
| "In today's world" | Very high | (delete—it's always today) |
| "In the realm of" | High | "In" |
| "A wide range of" | High | "Many" or "Various" |

### AI's Favorite Sentence Starters

- "It's worth mentioning..."
- "One of the key aspects..."
- "When we look at..."
- "Let's take a closer look at..."
- "It goes without saying..." (then why say it?)

### Dead Giveaway Phrases

These almost never appear in natural human writing:

- "In the ever-evolving landscape of..."
- "Navigating the complexities of..."
- "Unlock the potential of..."
- "A testament to..."
- "Serves as a beacon..."
- "Resonates deeply with..."

---

## Level 5: Content Tells

### Missing Specificity

AI gives generic examples because it can't verify real ones:

❌ **AI Pattern**:
> "Many companies have found success with this approach."

✅ **Human Pattern**:
> "Acme Corp tried this last quarter and cut costs by 12%."

### No Personal Experience

AI can't draw from lived experience:

❌ **AI Pattern**:
> "Users often report feeling frustrated when..."

✅ **Human Pattern**:
> "I've been stuck on hold with them three times this month."

### Balanced to a Fault

AI presents perfectly balanced arguments on everything:

❌ **AI Pattern**:
> "While some argue X, others contend Y. Both perspectives have merit."

✅ **Human Pattern**:
> "X is correct. Here's why the Y argument doesn't hold up."

### The "Comprehensive" Trap

AI tries to cover everything superficially rather than anything deeply:

- Lists every possible consideration
- Touches on all angles
- Never commits to priorities

**Fix**: Make choices. Prioritize ruthlessly. Leave things out intentionally.

---

## Level 6: Format Tells

### Bullet Point Addiction

AI reaches for bullets instantly:

❌ **AI Pattern**:
> The benefits include:
>
> - Improved efficiency
> - Cost savings
> - Better outcomes
> - Enhanced collaboration
> - Streamlined processes

✅ **Human Pattern**:
> It saves time and money. That's really the main thing.

### Predictable Headers

AI uses generic, parallel headers:

- Understanding the Challenge
- Exploring Solutions
- Implementing Changes
- Measuring Success

**Fix**: Use specific, interesting headers that could only apply to THIS content.

### The Rule of Three (Overused)

AI loves groups of three:

- Three main points
- Three examples
- Three recommendations

**Fix**: Sometimes there are two things. Sometimes seven. Use the real number.

---

## Document Audit Checklist

Run this checklist on any document to audit for AI tells:

### Quick Scan (2 minutes)

- [ ] **Ctrl+F test**: Search for "delve", "myriad", "tapestry", "crucial", "landscape"
- [ ] **Opening check**: Does it announce what it will do instead of doing it?
- [ ] **Closing check**: Does it just restate the opening?
- [ ] **Bullet count**: More than 3 bulleted lists in a single page?

### Deep Scan (10 minutes)

- [ ] **Voice check**: Could you identify the author by voice alone?
- [ ] **Specificity check**: Are examples real and verifiable?
- [ ] **Opinion check**: Does it take any actual positions?
- [ ] **Structure check**: Is the organization obvious/predictable?
- [ ] **Transition check**: Count instances of "moreover", "furthermore", "additionally"
- [ ] **Hedge check**: Count "it's important to", "it's worth noting", "one might argue"

### Red Flag Scoring

| Red Flags Found | Assessment |
|-----------------|------------|
| 0-2 | Likely human or well-edited |
| 3-5 | Needs revision |
| 6-10 | Significant AI patterns |
| 11+ | Almost certainly AI-generated |

---

## Policy Writing Guidelines

### For Official Documents

1. **Lead with action**: State what people must do, not what the document will cover
2. **Use concrete language**: Replace "stakeholders" with actual role names
3. **Be direct**: "You must" not "It is required that personnel shall"
4. **Include real examples**: Reference actual incidents, dates, names
5. **Vary sentence length**: Mix short punchy sentences with longer ones
6. **Show personality**: Policy can be clear without being robotic

### Before/After Examples

❌ **AI-Style Policy**:
> "It is essential for all employees to familiarize themselves with the multifaceted aspects of our comprehensive data security protocols. These guidelines serve as a beacon for maintaining robust protection of sensitive information in today's ever-evolving digital landscape."

✅ **Human-Style Policy**:
> "Everyone must follow these data security rules. We've had two breaches this year—both from people ignoring password policies. Don't be the third."

---

## Writing Tips to Sound Human

### Do This

- Start sentences with "But", "And", "So"
- Use contractions (don't, won't, can't)
- Include personal observations
- Admit what you don't know
- Use real numbers and dates
- Reference specific people or events
- Break "rules" occasionally
- End some sentences with prepositions
- Use fragments. For emphasis.

### Don't Do This

- Announce what you're about to say
- Use five words when two will do
- Hedge every claim
- Balance every argument perfectly
- Reach for impressive vocabulary
- Start every paragraph the same way
- Put everything in bullets
- End with a summary of what you just said

---

## Resources

- AI-WRITING-TELLS.md — Background research on AI detection
- [Plain Language Guidelines](https://www.plainlanguage.gov/) — Federal guidelines for clear writing
- Style guides: Strunk & White, AP Style, Chicago Manual

## Related Skills

- **When writing policies** → Load this skill + documentation-quality-assurance
- **When reviewing documents** → Load this skill + code-review (for systematic review patterns)
- **When teaching writing** → Load this skill + bootstrap-learning

---

*Skill Version: 1.0.0 | Created: 2026-03-05 | Category: Writing Quality*