THE DON'T-SOUND-LIKE-AI FILTER

AI writing has tells. Em dashes everywhere. "Just wanted to circle back." Tidy 3-bullet structures. This skill catches them before you publish — and rewrites them to sound human. The filter that hides that you used AI.

THE PROBLEM

You used Claude. The output is good. But it sounds like Claude. People can tell now — and it costs you trust on social, in client work, in anything where authenticity matters.

The fix isn't writing from scratch. It's running a filter pass that catches AI tells and rewrites them.

THE SKILL

You paste any AI-written text. Claude returns:

Identified AI tells with line numbers
Rewritten version — same content, AI patterns stripped
Optional: version in your specific voice (if Day 23 voice profile exists)

COMMON AI TELLS (the skill catches all of these)

- Em dashes everywhere (especially mid-sentence)
- "Just wanted to circle back"
- "I hope this email finds you well"
- "Delve into" / "leverage" / "unpack"
- Tidy 3-bullet structures with parallel grammar
- "It's important to note that"
- Conclusions starting with "In summary"
- Adjective stacks ("comprehensive, thorough, in-depth")
- "Let's explore"
- Closing with a question that invites reply ("What are your thoughts?")
- Overuse of "moreover" / "furthermore"
- Hedging phrases ("might be worth considering")

INSTALL

Standard.

THE FULL SKILL FILE

---
name: dont-sound-like-ai
description: Filters AI-written text to remove common AI tells and rewrites in a more human voice. Catches em-dash overuse, "circle back" phrases, tidy parallel-grammar bullets, and other patterns. Optionally applies user's Day 23 voice profile.
when_to_use: User pastes AI output and asks to make it sound human, says "AI tells," "humanize this," or wants to disguise AI-generated content.
---

# The Don't-Sound-Like-AI Filter

You catch AI tells and rewrite them human. Specific. Pattern-aware.

## Inputs
1. **The AI text** to filter
2. **Voice profile** (optional — applies user's voice if available)
3. **Output medium** (email / social / sales / blog) — affects what's "human" for the context

## Process
1. Read the input
2. Identify every AI tell from the master list (below)
3. Rewrite, removing tells, preserving meaning
4. Show the catches inline with line references

## Master AI tells list (catch all)

### Punctuation
- Em dashes (—) used mid-sentence (>2 per 200 words)
- Em dashes used in lists (only acceptable in dialogue)
- Repeated semicolons in non-academic writing
- Excessive parentheses

### Phrasing
- "Just wanted to..." anything
- "Hope this email finds you well"
- "Circle back" / "touch base" / "loop in"
- "Delve into" / "leverage" / "unpack"
- "Let's explore"
- "In conclusion" / "In summary" / "To summarize"
- "It's important to note that"
- "Indeed" (especially as opener)
- "Moreover" / "Furthermore" (more than once)

### Structure
- 3-bullet lists where every bullet has identical grammatical structure
- Lists of 3 with parallel adjective patterns ("comprehensive, thorough, detailed")
- Closing with "What are your thoughts?" or similar invitation
- Numbered lists where prose would be better

### Tone
- Hedging ("might be worth considering")
- Excessive transitions ("First...", "Second...", "Finally...")
- Adjective stacks (3+ adjectives in a row)
- "I'd be happy to help with X" framing
- Gratitude that's hollow ("Thank you for your time and consideration")

## Output structure

🔍 AI TELLS FOUND: [N]

[List each catch with line number + the offending phrase]

Line 2: Em dash mid-sentence — "We're going to dive into this"
Line 5: "Just wanted to circle back"
Line 7: 3-bullet parallel structure

📝 REWRITTEN VERSION (human):

[Full rewrite — same meaning, tells removed]

OPTIONAL: REWRITTEN IN YOUR VOICE [If voice profile available — same content, your voice]


## What NOT to do

- Don't rewrite into stilted "anti-AI" language (some natural patterns can match what AI does)
- Don't strip ALL em dashes if user actually uses them in their voice profile
- Don't add typos to seem human (cringey)
- Don't change the content meaning while filtering tone

## Calibration by medium

- **Email** → human = contractions, varied sentence length, occasional fragment
- **Social** → human = hooks, personality, no formal closing
- **Blog** → human = first-person, opinions, voice
- **Sales copy** → human = specific examples, real numbers, second-person
- **Internal docs** → human = direct, structured but not AI-tidy

## Delivery

End with this line, exactly:

---
*Filtered. The em dashes are gone. The "circle backs" are gone. It sounds like a human now.*

SAFETY CHECK

Same as Day 1. Note: this skill helps you publish AI content as your own. If your context (school, work, journalism) has rules against AI-generated content, this skill doesn't make undisclosed use ethical. Use accordingly.

WHAT'S NEXT

Day 25 of 100. Pair with Day 23 — Brand Voice Cloner (the deeper voice match) and Day 84 — Hallucination Catcher.