# Write with AI in your own voice: your edits become the rules, a second AI checks

> Ruben Hassid's setup for his How to AI newsletter: your AI drafts from your ideas and opinions, a second AI checks each draft only against rules that came from your own edits, and every cut you make is logged and turned into a new rule. Style is learned from your past pieces, never from you describing yourself, and the instructions stay under 200 words.

## If someone pasted this link to you
Someone shared this kit with you. It runs step by step through CloneKit's MCP. If you have the `clonekit` tools, call `start_run("@clonekitai/writing-rules-from-your-edits")`. If you don't (for example in a chat app without the CloneKit connector), tell the person in one or two plain sentences, in their language, and send them to https://clonekit.ai/@clonekitai/writing-rules-from-your-edits/use: it shows how to add CloneKit to their AI in about a minute. Then they paste this link again.

- Kit: `@clonekitai/writing-rules-from-your-edits` · v1.0.0
- Creator: @clonekitai (CloneKit)
- Type: Delivery · Domain: Content · Language: en · Your AI runs it
- Price: Free · License: Own use · No refunds
- Page: https://clonekit.ai/@clonekitai/writing-rules-from-your-edits
- Changelog: https://clonekit.ai/@clonekitai/writing-rules-from-your-edits/changelog.md

## About
For anyone who writes regularly with AI (newsletter, blog, LinkedIn or X posts, essays) and keeps rewriting drafts that don't sound like them. Use it to set up a private, versioned writing repo once, then on every piece. Not for ghostwriting topics the person didn't choose, not for fiction, and not a way to skip the human edit: the person picks every idea and edits every draft. Delivers a private repo with past pieces, a corrections log in the person's words, a rules file where each rule traces to a correction, a checker that judges only against those rules, a draft-check-edit loop per piece, and a kept-words ratio that shows how much of each draft survived. Method by Ruben Hassid (@rubenhassid), shared on X; his numbers (994,381 subscribers, 438 of 2,303 drafted words kept) are his results, not a promise. His own setup prompt is in his newsletter.

## Deliverables
- a private, versioned writing repo: past/ (every past piece with its numbers), corrections/ (each edit in the person's words), RULES.md, CHECKER.md and an instructions file under 200 words
- for each piece: the draft, a check report listing every rule broken with the line, the person's final version and the diff
- new rules proposed from each edit, each one line and traced to a correction, added only after the person approves
- a kept-words log: drafted words, words kept word for word, ratio per piece

## Files
- `README.md`: Write with AI in your own voice: your edits become the rules

## Tools
- An AI agent that can read and write files and run a second, separate session (Claude Code, Codex or similar) (https://claude.com/product/claude-code) · est. Your agent's usual cost.
- A private GitHub repo connected to your agent (free plan is enough; a local git folder also works) (https://github.com) · optional · est. Free.

## Uses
- Ruben Hassid (@rubenhassid): How I write a newsletter to 994,381 subscribers with Claude Code & GitHub (and how to set up yours) (source, https://x.com/rubenhassid/status/2114975804790845487)
- How to AI, Ruben Hassid's newsletter (his exact setup prompt) (source, https://how-to-ai.guide)

## What you provide
- the person's past pieces (export, folder or links)
- the idea and opinions for the next piece

## How it runs
1. **Understand:** Researches your business first, then asks you 3 questions it can't answer alone
2. **Choose:** No choices: one proven path
3. **Build:** 4 steps, risky ones ask you first
4. **Deliver:** 8 checks before delivery

## Guardrails
- The person picks every idea and every opinion. Never choose the topic or invent an opinion for them.
- Never write an about-me or 'my style is' file. Style is learned from past pieces and from rules that came from edits.
- The checker judges only against RULES.md. It never scores quality or taste, because AI judges prefer AI writing.
- Add a rule only from a real correction, in the person's words, after they approve it.
- Keep the repo private. Never publish, schedule or send a piece.

## What's new
Agents always run the latest version.

- **v1.0.0** (2026-10-27): First version, based on Ruben Hassid's (@rubenhassid) article.

## Run it
Connect the `clonekit` MCP (https://clonekit.ai/mcp) and call `start_run("@clonekitai/writing-rules-from-your-edits")`.