You’re reading Excellent AI Prompts.
AI prompt chains, agent skills, and agentic workflows for professionals learning to use AI to earn more, think sharper, and live better.
Copy them. Use them. Build with them.
You probably asked AI to match your voice a dozen times today.
I get it. This sounds like a reasonable request for large language models and the high level of reasoning they are capable of today. It’s not. This problem would be easier to remedy if you have copious examples of your writing style. Even then, sometimes the models get it wrong. If you don’t have enough examples for AI to recognize your linguistic habits, “write in my voice” is just another vague instruction. The model has to guess what your voice means, which parts matter, and how those qualities should change for the work you want it to do. So it does what it does. It defaults to a strange mix of copywriting and persuasive writing.
I started thinking seriously about this after watching an experiment unfold on X around ASD-STE100, the controlled language developed for European aerospace documentation.
Andrew Carr recommended telling AI to report in ASD-STE100 Simplified Technical English. Matt Pocock tested the idea in his global Claude instructions, then reported the following day that the passive instruction had not changed the model’s output as much as he hoped.
A few days later, @levelsio shared a much more positive result, and Dan McAteer described reported the same technique as having a significant impact on their experience with Claude.
When I learned of this, I wanted to try it out. I immediately created a Claude Project to only talk to me in the ASD-STE100 style. Once I received the output, I realized I’ve been doing a version of this all along. Not with ASD, but with various fiction frameworks and non-fiction writing styles via Claude Projects and AI agents. I use investigative journalism style quite a bit. I also use SBVR to ensure that I understand any topic I may be learning or writing about. A lot of these styles you’d only know if you’re in those particular industries. But the biggest idea is that better AI co-writing comes less from asking the model to sound better, and more from giving the model a writing system appropriate for the job. It understands systems exceedingly well. Your specific style of writing? Less so.
For instance, an operating procedure needs different language from a personal essay. A governance policy needs different logic from a client update. An executive summary should not unfold like a feature story, and a newsletter should not read like an aircraft maintenance manual. It seems basic, I know, but it’s not something we immediately think about when we’re co-writing with AI.
The premise of this issue is to stop asking AI to make it sound better when you’ve not even given it the appropriate writing system.
YOU are in control of how AI produces the work you request, not the other way around.
In This Issue
Four writing systems worth adopting now: Explore ASD-STE100, SBVR, the inverted pyramid, and narrative prose.
One set of facts, four different outputs: See how structure transforms writing without changing the truth.
Build a writing stack: Combine registers so each part of a project does its job.
The Multi-Register Editorial Engine: Copy a reusable prompt/instructions for ChatGPT, Claude, Gemini, Codex, Hermes, and other AI tools via Projects, AI agents, etc.
Commands you can try today: Get starting points for SOPs, policies, executive updates, and newsletters.
The much bigger map: Discover more writing systems to add to your editorial toolkit.
Why your taste is the operating system: Understand the judgment AI cannot supply for you.
Related “Throwback” Issue
When we were all new to Claude Projects
