Two skills. One that makes AI answer first and stop, one that makes it sound like a person. Both ready to copy.
There are two completely different versions of the AI writing problem, and almost everyone tries to fix them with one prompt.
The first is when AI writes something for other people. An ad, an email, a landing page. The failure there is that it sounds like AI, and your reader can tell.
The second is when AI is talking to you. The failure there is not tone, it is volume. You asked one question and got six paragraphs, four of which restate the question before answering it.
Those need two different skills, so here are both. The short one is built on three specific ideas, and it is worth knowing what each one contributes.
One
Simplified Technical English
A controlled writing standard built for aircraft maintenance manuals, where a misread sentence is a real problem.
What it contributes: 53 writing rules and a dictionary of roughly 900 approved words, each with exactly one meaning and one part of speech. You use "make sure," never verify, check, confirm, or ensure. Active voice, simple tenses, short sentences, one instruction per sentence. The full standard is here. You do not need all 53 rules. You need the discipline of one meaning per word.
Two
Attention Span
An open-source set of output styles for Claude Code that changes how the AI talks to you, not how it works.
What it contributes: the answer-first rule. Conclusion in line one, no wind-up, say the least that fully answers and then stop. Its own benchmark measured roughly 43% shorter output while code quality held at a 97% pass rate, which is the useful finding here. Shorter did not mean worse. The repo is here and it is worth installing on its own if you use Claude Code.
Three
Zinsser's Four Principles
From On Writing Well, the nonfiction writing book that has been in print since 1976.
What it contributes: clarity, brevity, simplicity, humanity. The one doing the most work is brevity, and specifically his standard for it: every sentence is guilty until proven essential. The humanity principle is the reason the skill below does not turn your AI into a robot that answers in fragments.
This is for when AI is talking to you. Paste it into your custom instructions, your project instructions, or save it as a skill.
This is for when AI writes something other people will read. The ban list below is the real payload. It is the specific vocabulary that makes a reader think "this was AI" before they can explain why.
These two do opposite jobs, so do not run them together in the same place.
One honest note on the shorter-is-cheaper point. Fewer output tokens is a real saving if you are paying per token through an API. On a flat monthly subscription it mostly buys you time and attention rather than money, which is the better reason to do it anyway.