AI Slop Is a Real Problem
Every first draft from the AI had em dashes, 'key insights', and rule-of-three lists. I had to build rules to stop it.
The first time Claude wrote a project description for this site, I counted seven em dashes in four paragraphs. Every section ended with a punchy one-liner. There were three instances of "the key insight was" and two "this meant I could" transitions. It read like every AI-generated LinkedIn post I've ever scrolled past.
The same problem showed up in Qumio, my Telegram assistant. The bot's responses had "Let me know if you need anything else!" and "Here's your..." prefixed to every message. It sounded like a customer service chatbot, not a personal tool.
The fix
I built an explicit ban list into the content generation rules. No em dashes. No "delve", "leverage", "seamlessly", or "game-changer". No rule-of-three lists that exist for rhythm instead of content. No performative self-deprecation. No punchy paragraph endings.
The list keeps growing. Every time I read output that sounds like AI, I add the pattern to the rules. Some recent additions:
- "It's important to note"
- "In an ever-changing"
- Starting sections with "So"
- Rhetorical questions as transitions
For Qumio, the rules went directly into the skill prompts. The assistant should sound like a brief from an executive assistant, not a chatbot trying to be friendly. No filler, no sign-offs, no emoji.
Why it matters
The problem isn't that AI writing is bad. It's that it's recognizably AI. There's a specific cadence, a specific set of transitions, a specific way it structures parallel phrases. Once you see the pattern, you can't unsee it. And your readers can't either.
The fix isn't "write better prompts." It's a concrete list of things to never do, checked after every draft.