Anthropic's anti-mannered-prose prompt improves writing across Claude and GPT, with hands-on tutorial

omarsar0 · x · 2026-09-08

DAIR.AI's Elvis shares an official Anthropic prompt that he says dramatically improves writing density in Claude Fable 5.1 — and, surprisingly, works on GPT-5.6 Sol too. He now uses it as a default rule wherever he edits or writes with AI.

The prompt defines "mannered prose": metaphor and flourish substituting for direct statement (e.g., "a dial worth turning" instead of "a parameter worth varying"), explains why it irritates readers — the writer performs at the reader's expense — and asks the model to test and revise each sentence. Unlike "be concise," which states a goal without criteria, it names what to remove and what replaces it.

He also shipped an interactive tutorial on DAIR.AI Academy where you practice on a 120-word customer update; Anthropic also offers the shorter variant "Please remove all mannered prose."

Related event: Anthropic's Writing Prompt Goes Viral, Improving Multiple Models' Prose(2 posts)→

Original post →

More from Apps

Apps channel →