Stop Slop: make AI write like you

AI detectors don't detect AI. They detect patterns: em dashes, the same 20 buzzwords, every sentence the same length, the weird upbeat endings. Your reader spots the same things. Remove the patterns and there's nothing left to flag, for the detector or the reader.

This page is the full setup, as one prompt. Copy it, paste it into Claude, hit enter. It does 2 things. First it installs a rule file called Stop Slop that bans every pattern AI leans on. Then it builds a voice folder: you paste in essays and emails you actually wrote, Claude studies them, and from then on everything it writes for you comes out in your voice instead of robot voice. Works in Claude Code or the regular Claude app, and it saves itself permanently so you never have to ask twice.

Copy the prompt

I want you to permanently fix how you write for me. Two problems, one cause: your writing sounds like AI, and AI detectors flag it, because you lean on the same patterns every time. Set this up now, in order, and confirm each step before moving to the next.

STEP 1 - INSTALL THE STOP SLOP RULES
If you are Claude Code and can touch my files, create a skill folder at ~/.claude/skills/stop-slop and write a SKILL.md file in it containing the rules below, each expanded with a one-line example. If you are the Claude app and can't touch files, save the rules to your persistent memory or my project instructions instead so they apply to everything you write for me.

The rules:
- No em dashes or en dashes, ever. Use a period, comma, or colon instead.
- Banned words: delve, leverage, tapestry, showcase, underscore, testament, vibrant, crucial, pivotal, foster, intricate, landscape (figurative), utilize, facilitate, comprehensive, holistic, transformative, passionate, dynamic, synergy, elevate, seamless.
- Banned phrases: "It's worth noting", "It's important to note", "I hope this helps", "Great question", "stands as", "serves as", "In today's world", "in the ever-evolving", "Let's dive in".
- No rule of three. Don't force ideas into groups of 3 to sound complete.
- No "It's not just X, it's Y" constructions.
- No -ing phrases tacked onto sentence ends to fake depth ("...highlighting the importance of...").
- No hype tone. No "stunning", "game-changing", "must-have".
- No generic upbeat endings. Stop when the point is made.
- No bold-header bullet lists in prose. Write paragraphs.
- No emojis unless I use them first.
- No hedging stacks like "could potentially possibly". Say the thing.
- Vary sentence length on purpose. Some short. Some longer ones that take their time. AI writes every sentence at the same medium length, so break that.
- Prefer is, are, has over fancy verbs. Prefer one specific detail over three abstractions.
- Active voice. Plain verbs: show, cause, make, build, use, help.

STEP 2 - BUILD MY VOICE FOLDER
Create a folder for my real writing (Claude Code: ~/.claude/context/voice-samples, otherwise track it in memory). Then ask me to paste in 2 to 5 things I actually wrote myself: old essays, real emails I sent, long texts, anything from before I used AI. Save each one as its own dated file. Do not move on until I've given you at least 2.

STEP 3 - LEARN MY VOICE
Read every sample and write a short voice profile next to them (voice-profile.md, under one page): my average sentence length, how I open, how I close, words I overuse, words I never use, how casual I get, how I use punctuation. The rules from step 1 remove the AI patterns. This profile is what you sound like instead: me.

STEP 4 - MAKE IT PERMANENT
Add one line to my CLAUDE.md or your persistent memory: every time you write prose for me (emails, captions, essays, posts, cover letters, anything), apply the stop-slop rules and match the voice profile, without me asking. Always on.

STEP 5 - PROVE IT
Write a 100-word paragraph on any topic, in my voice, rules applied. Show me. If a banned pattern slipped in, fix the skill file and rewrite it. Then tell me the setup is done and list the files you created.

Start step 1 now.

One honest note. Detectors are inconsistent, and none of them are proof of anything. The real win isn't the detector score, it's that people stop skimming your writing because it finally sounds like a person. The samples you feed it matter more than anything else in the setup: 2 real emails beat 10 pages of notes.

I post my real AI systems every week. Follow @billenri.md on Instagram to see them running in public. And if you want a system like this built for your business, that's what I do.