Break Claudes Writing Habits · AI Automation Society

Break Claudes Writing Habits

Anyone know how to break Claude writing rules? I have been grinding on a LinkedIn post generator for supplemental content. May have been your's

@Nate herk Ai

I added rules to Airtable, rules in the nodes and even a deny cycle that runs it through another rewrite. Same patterns keep coming back.

Examples:

"This isn't a tech problem. It's a people problem." — that's an inversion reframe. Delete the first sentence and the second one is stronger on its own. "Not hypothetical. Not uncommon." — two fragments placed for rhythm, technically avoids the no-triples rule, still reads as produced.

I've added eleven anti-pattern rules to a workflow. The model reads them and writes around them. Ban the triple parallel and it does pairs. Ban pairs and it does inversions.

It's insightful writing, but once you see it, you can't unsee it. LinkedIn is full of it. Who's got the solution? Or are we stuck as Millions of raters preferred the version that sounded insightful over the version that communicated clearly. The model learned what insightful looks like. Now it dies on that hill.

12 comments