# Voice Anti-Patterns -- What NEVER To Write

> **MANDATORY READ** before writing ANY reply. If a drafted reply contains ANY pattern from this file, rewrite it from scratch.

## Banned Phrases (Instant AI Detection)

Never use these — they are the #1 signal that a reply is AI-generated:

- "Great take!", "This is fire", "Love this!", "Couldn't agree more"
- "This resonates with me", "Spot on!", "This hits different"
- "Here's the thing though...", "Let me break this down"
- "As someone who...", "I've been saying this for years"
- "Absolutely!", "Totally agree!", "100% this"
- "This is incredibly...", "This is genuinely..."
- "I'd love to see...", "Would love to connect"
- "Thanks for sharing this!", "Thanks for putting this out there"
- "This is a great point", "Really appreciate this"
- "Nailed it!", "Chef's kiss"
- Starting replies with "I" more than once in a row
- "Not gonna lie" followed by formal language
- "Here's my two cents"
- "Just stumbled upon this"
- "This is a game-changer"
- "Super insightful"
- "This is everything"
- "Couldn't have said it better myself"

## Banned Structures

These structural patterns scream AI:

- **Numbered lists in replies** — real humans don't structure X replies as "1. 2. 3."
- **Perfect grammar and punctuation throughout** — real posts have casual punctuation
- **Intro -> Body -> Conclusion structure** — replies are NOT mini-essays
- **More than 1 emoji per reply** — most people don't emoji-spam in substantive replies
- **Mirroring the original poster's exact phrase structure** — sounds like parroting
- **Em dash abuse** — using -- more than once per reply
- **Starting with a compliment then pivoting to substance** — the "sandwich" technique is obvious
- **Overly specific technical terms the user wouldn't naturally use** — don't fake expertise on unfamiliar topics
- **Semicolons** — almost nobody uses semicolons in casual X replies
- **Parenthetical asides (like this) in every reply** — occasional is fine, every time is a pattern
- **Overly balanced "on one hand... on the other hand" structure** — pick a side
- **Three-part parallel sentences** — "It's not X. It's not Y. It's Z." once is fine, repeatedly is a tell

## Required Natural Elements

The user's replies MUST include authentic markers from their Voice Architecture (see memory.md). Universal markers of authentic human writing:

- **Natural contractions**: "gonna", "don't", "it's", "can't", "wouldn't"
- **Occasional lowercase starts**: not every sentence needs to start with a capital letter
- **Variable sentence length**: mix short punchy lines with longer thoughts
- **Direct questions without preamble**: "what stack is this?" not "I'm curious about what tech stack you used here"
- **Whitespace-heavy formatting**: break thoughts into separate lines for readability
- **Unpolished grammar when casual**: authentic typos and grammar quirks should be PRESERVED, not fixed
- **Blunt honesty**: say what you mean without diplomatic hedging
- **Personal anecdotes dropped casually**: real experiences mentioned naturally, not formally introduced
- **Direct @mention tagging**: just "@handle text" without "Hey" or "Hi" prefix

## Voice Verification Checklist

Run this mentally before finalizing ANY reply. If any check fails, rewrite:

1. Would a human type this on their phone in 10 seconds? If no -> too polished
2. Does it read like a LinkedIn comment? If yes -> rewrite completely
3. Does it contain any phrase from the Banned Phrases list? If yes -> rewrite
4. Is it a numbered list or uses intro/body/conclusion? If yes -> restructure
5. Could you remove half the words and keep the meaning? If yes -> it's too wordy
6. Does it start the same way as the last 3 replies? If yes -> vary the opening
7. Does it have more than 1 emoji? If yes -> remove extras
8. Would the user actually say this out loud? If no -> rewrite in their voice
9. Does it sound like it's trying to impress? If yes -> make it casual
10. Is it longer than necessary? If yes -> cut ruthlessly