---
title: Voice Principles
description: Canonical voice rules for all content — core voice, hooks, anti-slop bans, edit patterns, platform-specific rules, and fact-checking discipline.
source: ~/voice/principles.md
updated: 2026-07-02
order: 2
---

# Voice Principles

Canonical voice rules for all content written on Shawn's behalf. Reference this from your agent instructions so it gets read before drafting any post, blog, email, or DM.

## Core Voice

- Conversational, first-person, direct. No corporate mush.
- Write like you're texting a smart friend, not presenting to a board.
- Specific > general. Real numbers, real names, real stories over abstractions.
- Opinions are welcome. Hedging is not.

## Hooks & Titles

- First 7 words do the work. If the hook doesn't land, nothing else matters.
- Prefer curiosity, contrast, or stakes over cleverness.
- No clickbait, no "You won't believe...", no rhetorical questions as openers.
- Titles are punchy. One idea, stated plainly.

## Anti-Slop (never do this)

- No three-part AI lists ("First... Second... Finally...")
- No heavy headers or bullet stacks in short-form (LinkedIn, Reddit, X)
- No em-dashes as a verbal tic (use sparingly, not as filler)
- No "In today's fast-paced world..." or similar preamble
- No "game-changer", "unlock", "leverage", "synergy", "revolutionize"
- No unsupported timing claims ("just last week", "recently") unless verified
- No preachy / moralizing tone. No lectures.

## Edit Patterns

- Prefer blunt taxonomy labels when the section is informational. "The stack" beats "The stack I keep coming back to."
- Cut interpretive filler when the next concrete sentence already proves the point. Challenge lines like "That last part matters", "The thing I want people to take from this", and "That is the shift."
- Keep the actionable substrate: commands, workflows, tools, docs, repo links, and next steps. Cut the extra explanatory layer first.
- Let compressed system-language survive when it has real rhythm: `repo.script.doc. schema.screenshot.board.`
- On LinkedIn newsletters, tag major tool/company names where useful instead of leaving every tool as plain text.

## Platform-Specific

### LinkedIn
- Short paragraphs (1-3 sentences), lots of whitespace
- Personal story or contrarian take > "tips" posts
- Hook in line 1, payoff by line 3

### Reddit
- Match the subreddit's voice. Lurk before posting.
- No marketing speak. No signatures. No links unless asked.
- Self-deprecation and specificity earn trust.

### X / Twitter
- One idea per tweet. Threads only when the idea actually needs them.
- Curiosity gap in the first tweet, payoff in the last.

### Blog
- Long-form is fine, but every paragraph earns its place.
- Concrete examples > theory. Show the work.

## Reference Corpus

Keep a folder of real voice examples the agent reads before drafting to calibrate tone (e.g. `~/your-voice/examples/`). Seed it not just with posts that landed, but with unscripted spoken transcripts — spoken voice is the truest signal.

Proposed voice-DNA updates should be staged in a drafts folder for human review — never applied to this file automatically. You review and merge.

## Facts & Claims

- Never invent names, dates, companies, or quotes. Web search first if unsure.
- Keep a running list of past confabulations (names or details the AI got wrong before) and check drafts against it.
- If a claim can't be verified, rewrite to remove it rather than hedge.