Make AI Write Like You: The 12-Rule Writing System

The banned-word list you keep adding to

You know the routine. You tell it no em-dashes. You tell it to stop saying "delve." You tell it not to sound like AI.

Then the next draft arrives and it sounds exactly the same, because you never told it how to write. You only told it what not to say.

A ban list is a blocklist. A writing system is a style. One of them scales, and it isn't the list you keep adding to at 11pm.

Vox (@Voxyz_ai) put this bluntly in a post that did 403,000 views in a day, and the line that stuck with me was that every README, landing page, and description ships in the same voice while you ban words one at a time.

The tic that has a name

There's one pattern that gives it away faster than any word. Vox calls it corrective juxtaposition. You've read it a thousand times:

The answer isn't X, it's Y.

X isn't the problem. Y is.

Same mold, any topic. His explanation is the sharpest I've seen: it gives a plain statement the shape of insight without adding a fact.

That's the whole trick. The sentence sounds like it landed something. Read it twice and nothing new arrived.

The structure itself is fine. Rhetoricians have used it for two thousand years. The family is antithesis, the corrective form is correctio, and it gave us "ask not what your country can do for you" and Dr. King's line about the color of skin and the content of character.

As Vox put it, rhetoricians used it for 2,000 years and LLMs wore it out in a few.

Once you know the shape, you can't stop seeing it. Here's one short brief and the six replies that followed.

Orwell already wrote the first half

The fix is eighty years old. George Orwell published six rules in Politics and the English Language in 1946, and they still do most of the work:

1. Never use a metaphor, simile, or other figure of speech which you are used to seeing in print.
2. Never use a long word where a short one will do.
3. If it is possible to cut a word out, always cut it out.
4. Never use the passive where you can use the active.
5. Never use a foreign phrase, a scientific word, or a jargon word if you can think of an everyday English equivalent.
6. Break any of these rules sooner than say anything outright barbarous.

Rule six is the one people skip, and it's the most important. It gives the machine permission to be human when the rules would make it stiff. Without it you get prose that obeys and dies.

The 2026 patch

Orwell never met a language model. Vox added six more rules for the tics that are specific to this decade:

7. Don't build a straw man to knock down. Use not X, it's Y once per piece, max.
8. Two examples are enough. Don't stretch to three.
9. Don't announce what you're about to say. Say it.
10. Don't end two paragraphs in a row with punchlines.
11. Vary the length and shape of neighboring sentences.
12. Break any of these rules sooner than write like a machine.

Rule 9 is the one that will change your drafts today. Watch for "Let's dive into why this matters" and "Here's the thing." That's throat-clearing. Cut it and start with the thing.

Rule 10 explains why AI writing feels exhausting. Every paragraph ends on a mic drop, so nothing lands. Punchlines work because most sentences aren't one.

See it work: the same paragraph, twice

Rules are abstract until you watch them bite. Here's a paragraph about email follow-up, first the way it usually arrives.

Before. Let's dive into why follow-up matters. Here's the thing: most people think follow-up is about persistence. It isn't about persistence. It's about timing. The best operators don't send more emails. They send better-timed ones. Follow-up isn't a numbers game. It's a memory game. And that changes everything.

Count the damage. Two announcements before it says anything (rule 9). Three separate not X, it's Y shells (rule 7). Both paragraphs would end on a mic drop if you let it keep going (rule 10). Every sentence is the same length (rule 11).

Now the same point, through the system.

After. Most follow-up fails on timing, not effort. A second email two days later gets read. The same email two weeks later arrives as a stranger. If you only change one thing, shorten the gap.

Shorter, plainer, and it actually tells you what to do. The information went up while the word count went down. That's the whole system working.

The whole system, ready to paste

Here's the give-away. Paste this into your global instructions once and every session picks it up. No per-chat reminders, no ban list.

WRITING SYSTEM. Apply to everything you write for me.

Orwell's six (1946):
1. Never use a metaphor, simile, or figure of speech you are used to seeing in print.
2. Never use a long word where a short one will do.
3. If it is possible to cut a word out, cut it out.
4. Never use the passive where you can use the active.
5. Never use a foreign phrase, a scientific word, or a jargon word if an everyday
   English equivalent exists.
6. Break any of these rules sooner than say anything outright barbarous.

The 2026 patch:
7. Do not build a straw man to knock down. Use "not X, it's Y" once per piece, max.
8. Two examples are enough. Do not stretch to three.
9. Do not announce what you are about to say. Say it.
10. Do not end two paragraphs in a row with punchlines.
11. Vary the length and shape of neighboring sentences.
12. Break any of these rules sooner than write like a machine.

How I write (fill this in yourself):
- Words I never use:
- How I open:
- How I close:
- My rhythm:
- What I would never say:

Before you hand me anything, check it against all twelve and fix what fails.

The last block is the part that makes it yours. Rules 1 through 12 stop it sounding like a machine. Only your five lines stop it sounding like everyone else who pasted the same twelve rules.

Where to put it so it actually sticks

A writing system only works if every session reads it without being asked.

Global beats per-chat every time. If you have to remember to paste it, you'll forget on the day you're rushing, and the rushed drafts are the ones that ship.

The same save-it-once logic applies to the rest of your recurring work, not just your voice. If you want a starting set, here's a 20-prompt pack for a one-person business.

The exact steps, in the Claude desktop app

Four clicks. No terminal, no file to create.

1. Open the Claude desktop app on your Mac or PC.
2. Open Settings from the menu under your name.
3. Find the box called "Instructions for Claude." That's the account-wide one. Anthropic's own description: "Any instructions you add here will be applied to all of your conversations with Claude."
4. Paste the whole block from above and save.

Then open a brand new chat, because a conversation you already started won't pick it up.

Test it in ten seconds. In that new chat, ask: "What writing rules are you working under right now?" If it reads the twelve back to you, it's live. If it doesn't, you're still in the old chat.

One warning worth reading. These instructions apply to everything, including code comments and quick factual questions. Keep the block to writing rules only. Don't put project details or client facts in there, because they'll leak into unrelated conversations.

Filling in your five lines

The five blanks are the whole ballgame, and you don't have to invent them. You already wrote them.

Pull five to ten things you actually sent: newsletter issues, client emails, DMs you're proud of. Then hand them over with this:

Here are 5 to 10 things I actually wrote and sent: [paste them].

Study them and tell me, in plain language:
1. Words and phrases I clearly never use.
2. How I open. What is my first move?
3. How I close. Do I land it, ask, or trail off?
4. My rhythm. Short punchy sentences, long rolling ones, or a mix?
5. What I would never say, based on what is missing from these samples.

Write it as five short lines I can paste into a writing system.
Do not flatter me. If my writing has a bad habit, name it.

That takes about ten minutes and it's the difference between generic and yours.

My receipts, including the embarrassing one

I'm not writing this from theory. I got beaten by this exact tic on my own site.

I built a checker that scans every article for the not X, it's Y shell before anything publishes. The first full run flagged 81 instances across pages I had already called finished.

Then it got worse. About an hour after I cleaned all 81 out, I rewrote an introduction and planted a brand new one. My own gate caught me.

That's the argument for a system over willpower. I knew the rule, I had just spent an hour enforcing it, and I still did it. A rule that lives in your head is a rule you break when you're tired.

It happened again the week I wrote this. A reformatting pass on twenty articles reintroduced two fresh ones, and the gate caught both before they shipped.

What this won't fix

Honesty matters more than the pitch here, so here's where the system stops.

• It won't give you something to say. Voice is delivery. Judgment is the product.
• It won't survive a lazy prompt. "Write me a post about leadership" produces generic writing in your voice, which is worse than generic writing.
• It won't catch everything. Rules are a floor. Reading your own draft out loud still finds things no rule does.

Read it out loud before it ships. If you run out of breath, rule 11 needed you and you weren't listening.

The part worth remembering

Banning words is whack-a-mole. Twelve rules, pasted once, plus five lines only you can write, and the voice problem mostly goes away.

And once you've written those five lines down, they're reusable. That same block belongs in your email drafting, your newsletter, your landing pages, and anything else you hand to a machine.

That's the shift worth making. Stop correcting output. Start configuring the writer.

Want the finished version instead of building it?

I've built this into the skills I use every day, so the voice rules travel with every draft.

You get all of them, plus a direct line to me, inside the Solo Creators AI Studio. See what's inside

And if you'd rather not build any of it yourself, that's what I do one to one. Bring me the writing you already have and I'll turn it into a system your whole business runs on.