Skip to content
learnthevibe

modul 5quality and anti-slop

anti-slop copy: prose that survives a suspicious reader

Readers now scan for machine-prose tells, and the tells are documented. A voice document plus corpus-level checks keep agent-drafted text worth your name.

~3 min read · ~9 min with the exercise · v1.1 · verified: 2026-07-15

the reader arrived suspicious

Your product's text meets people who have read three years of machine prose and learned its taste. Wikipedia's editors turned that taste into a document: Signs of AI writing, a field guide built from thousands of flagged articles. It catalogs an "AI vocabulary" (delve, pivotal, testament, landscape, robust), the inflation of ordinary facts into significance ("stands as a testament to"), the rule-of-three reflex, the negative parallelism, the em-dash cadence. Their key observation explains all of it at once: a model regresses to the statistical mean, so its prose "tends toward the most statistically likely result."

The same guide warns against outsourcing the judgment: detector tools "have non-trivial error rates," and the patterns matter more than any score. You are the detector. This lesson gives you the rubric and the tooling.

⌘K

search lessons and terms…