# How do I make Claude write like me?

There are three routes, from free to paid: write a short style note in Claude's instructions, give Claude 3 to 10 real samples of your writing in a Project or a skill, or connect Ownhand, which rewrites each draft from a Hand built from your samples and learns from what you send. Samples beat a description, and the free routes are enough if you mostly write one kind of thing.

By Jordan Gibbs, founder of Ownhand at Thalient Labs. Updated 2 October 2026.

Claude has four free places to put your style: your profile instructions, a Project, a skill, and memory. Claude Code adds two more. Each one helps, but each runs out at a predictable point. Below are the steps for each, as Anthropic's help center describes them today, and then where Ownhand takes over.

## The options at a glance

| Option | Where | What it holds | Cost |
| --- | --- | --- | --- |
| Instructions for Claude | Settings | A few lines about how you write, in every chat | Free |
| Project instructions and knowledge | A Project | A style note plus real samples, for chats in that Project | Free (five Projects on the free plan) |
| Skill | Customize > Skills | Rules and example texts in a skill.md file | Free |
| Memory | Settings > Memory | Notes Claude keeps about you | Free |
| CLAUDE.md or an output style | Claude Code | Standing instructions for every session | Comes with Claude Code |
| Ownhand | A custom connector or MCP server | A Hand from 5 to 10 samples, 12 occasions, rules learned from what you send | $1 free, then about half a cent per rewrite |

## 1. Write a style note in your instructions

This is the quickest fix. It costs nothing. Click your initials in the lower left corner, choose Settings, and describe how you write under Instructions for Claude. It applies to your whole account.

Most style notes fail because they're full of adjectives. "Casual but professional" gives you the model's idea of casual. Write rules Claude can check against a draft:

- Length: "Slack replies are one or two sentences. Emails stay under 120 words."
- Openers and sign-offs: "No greeting in Slack. Sign emails with my first name only."
- Punctuation and casing: "Lowercase is fine in DMs. No exclamation marks."
- Words you never use: pick them from [the list of phrases that read as AI](https://ownhand.dev/answers/ai-writing-tells).

A note is enough when you mostly want Claude to stop padding. It fixes length and openers well. It does poorly on habits you'd never think to write down, like how you hedge, where you put the ask, or how blunt you are with a peer compared with a client.

## 2. Give Claude samples in a Project

Examples show what a description misses. Anthropic's prompting guide calls examples one of the most reliable ways to steer Claude's tone and structure. It suggests 3 to 5 of them. A Project is the free place to keep them, and free accounts can have up to five.

1. Create a Project named after the kind of writing, such as "Client email" or "Slack".
2. Click Set project instructions. Add your style note and one more line: "Write like the samples in this Project. Match their length and how they open."
3. Add 3 to 10 things you wrote and sent yourself to the Project's knowledge. Paste them as text, each labeled with where it went. Leave out anything an AI drafted.
4. Start chats inside the Project whenever you want a draft in your voice.

When that is enough: you write one kind of thing, and you're fine starting those chats in one place. The samples only apply inside that Project. The same samples serve every kind of message. Fixing a draft before sending doesn't update anything.

There's a ceiling too. A 2025 study of more than 400 real authors found that models given a few samples copied style fairly well in news and email, but struggled with blogs and forum posts.

## 3. Package your style as a skill

A skill is a folder with a skill.md file. Claude loads it when a task calls for it. Anthropic says skills work on Free, Pro, Max, Team and Enterprise plans. They recommend putting example inputs and outputs in the file. That makes it a good home for a "write as me" guide that works outside any one Project.

1. Create a folder with a skill.md file. Start it with a name and a description that says when to use it, such as "Use when I ask for a message or an email in my voice."
2. Below that, add your rules and 3 to 5 samples, each with a line saying where it was sent.
3. Zip the folder, then upload it under Customize > Skills with Add.

This is enough when you want one portable guide. It has the Project's blind spots. You update it by hand, and it can't tell a Slack DM from a cold email unless you write both cases out.

## 4. Know what memory does

Memory is on by default for Free, Pro and Max plans. Claude saves topics as you chat. The help center lists communication preferences and working style among them. You can read, edit or delete each topic under Settings > Memory. Use it for standing rules, like "I sign emails with my first name only." It holds notes about you. On its own, it won't reproduce your voice.

## 5. In Claude Code: CLAUDE.md and output styles

Claude Code reads `~/.claude/CLAUDE.md` in every project. It's the place for personal rules about commit messages and PR descriptions. An output style goes further. A Markdown file in `~/.claude/output-styles` sets Claude's tone and format for a whole session, and `/output-style` switches to it. Anthropic's docs say both are instructions Claude follows, with no guarantee. For pull requests in particular, see [PR descriptions in your own style](https://ownhand.dev/answers/pr-descriptions-in-my-style).

Other coding tools do the same. Cursor keeps personal rules under Rules in its Customize menu and project rules as `.mdc` files in `.cursor/rules`. GitHub Copilot reads repository instructions from `.github/copilot-instructions.md`, and its docs list personal instructions for Copilot Chat on GitHub.com. Both are style notes, so they have the limits of a style note.

## 6. Connect Ownhand for drafts written from your samples

Everything above gives Claude your style once and leaves the rest to the model. Ownhand moves the writing into a tool. Claude drafts as usual, then calls `write` with the draft and where it is going. The tool rewrites it from your Hand, checks it for AI tells, and returns the text for you to approve.

### Claude Code

1. Run the command under Install at the bottom of this page.
2. Run `/mcp`, pick `ownhand`, and sign in in the browser.
3. Say "Set up my Ownhand Hand" and paste 5 to 10 things you wrote.

### Claude on the web, desktop or mobile

1. Go to Customize > Connectors, click **+ Add**, then **Add custom connector**.
2. Name it Ownhand, paste `https://ownhand.dev/mcp`, click Continue, then Add.
3. Sign in, pick or create your Hand, and click Allow.
4. In a chat, turn it on from the + button under Connectors.

Free Claude plans allow one custom connector, so this works there too. On Team and Enterprise plans, an owner adds the connector under Organization settings first. Then each member connects it.

Then ask the way you normally would: "reply to Dana in my voice" or "write the PR description as me". To make it a habit, paste the agent blurb from the [docs](https://ownhand.dev/docs#blurb) into your instructions or `CLAUDE.md`. It tells Claude when to call the tools and to report what you sent.

## What changes in the output

A style note gives every message the same treatment. A Hand is built from your own sentences. Each draft is fitted to one of 12 occasions, from a Slack DM to a cold email. Replies match the casing, length and punctuation of the thread you're in. Here's one AI draft, rewritten for two different writers:

These examples are real output from the live API. Each AI draft was rewritten by a demo Hand: a fictional writer (an engineer, a founder, an ops lead, a designer or a consultant) built from a handful of samples.

**AI draft:** "Hi Priya! I hope you're doing well. I wanted to reach out to see if you might have a moment to review my PR #482 before the 3pm deploy."

**Demo Hand, engineer:** "priya, can you review pr #482 before the 3pm deploy?"

**AI draft:** "Hi Priya! I hope you're doing well. I wanted to reach out to see if you might have a moment to review my PR #482 before the 3pm deploy."

**Demo Hand, ops:** "Hi Priya, can you review PR #482 before the 3pm deploy?"

Both writers got the same facts and the same ask. One wrote lowercase with no greeting. The other kept the greeting and the capitals. A shared style note would have given them the same message. A status update shows the same thing on a longer draft:

**AI draft:** "I'm pleased to report that the migration is approximately 80% complete. We remain on track to finish by Friday, barring any unforeseen issues."

**Demo Hand, ops:** "Migration: 80% done. On track to finish Friday."

The other change comes after you send. When you edit a draft, Claude reports the exact text you sent, and your Hand learns from the difference. A reason you give becomes a rule right away. A pattern in your edits becomes a rule once it shows up twice. More on that in [how Ownhand learns from what you send](https://ownhand.dev/answers/how-ownhand-learns-from-edits).

## Which one to pick

- You write one kind of thing and rarely edit Claude's drafts: a style note, plus a Project with samples.
- You want the same guide in every chat: a skill.
- You work mostly in Claude Code: `CLAUDE.md` for rules and an output style for tone.
- You send many kinds of messages through Claude and keep fixing the drafts by hand: Ownhand.

For the style note question side by side, see [Ownhand or custom instructions](https://ownhand.dev/answers/ownhand-vs-custom-instructions). If you also use ChatGPT, see [make ChatGPT write in your voice](https://ownhand.dev/answers/make-chatgpt-write-in-my-voice).

## Follow-up questions

### How many samples should I give Claude?

Anthropic's prompting guide suggests 3 to 5 examples to steer tone and structure. Ownhand asks for 5 to 10. Use things you wrote and sent yourself, from more than one place. Keep the short ones. A two-line Slack reply shows Claude how short you go.

### Can Claude learn my style from my past chats?

Only a little. Memory keeps notes about you, including communication preferences, and paid plans can search past chats. The trouble is what Claude sees there. It's how you write to Claude, which is rarely how you write to a client or teammate. Real sent messages make better samples.

### Does Ownhand work on the free Claude plan?

Yes. Anthropic's help center says free users can add one custom connector, and Ownhand needs only one. Ownhand bills separately. New accounts start with free credit, and after that you pay per rewrite, with no subscription.

### Is what I give Ownhand stored?

Your samples, drafts, and the text you return are stored in your account so the tool can work. The thread passed along with a reply is used for that one request and not saved. Ownhand doesn't train its own models on your writing. You can delete a Hand any time.

## Sources

Checked on 2 October 2026. If something here is out of date, email [hi@thalientlabs.ai](mailto:hi@thalientlabs.ai) and we will fix it.

- [Claude Help Center: Understanding Claude's personalization features](https://support.claude.com/en/articles/10185728-understanding-claude-s-personalization-features): Instructions for Claude under Settings, project instructions, skills
- [Claude Help Center: How can I create and manage projects?](https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects): Set project instructions, project knowledge, five projects on free accounts
- [Claude Help Center: How to create custom skills](https://support.claude.com/en/articles/12512198-how-to-create-custom-skills): plans, skill.md with name and description, example inputs and outputs, Customize > Skills
- [Claude Help Center: Use Claude's chat search and memory](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context): memory on by default for Free, Pro and Max, Settings > Memory, topics you can edit
- [Claude Help Center: Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp): Customize > Connectors, one custom connector on the free plan, owner setup on Team and Enterprise
- [Claude Docs: Prompting best practices](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices): examples as one of the most reliable ways to steer tone, 3 to 5 examples
- [Claude Code Docs: How Claude remembers your project](https://code.claude.com/docs/en/memory): ~/.claude/CLAUDE.md for personal preferences in every project
- [Claude Code Docs: Output styles](https://code.claude.com/docs/en/output-styles): custom styles in ~/.claude/output-styles, /output-style, no guarantee
- [Wang et al., Catch Me If You Can? Not Yet (EMNLP Findings, 2025)](https://arxiv.org/abs/2509.14543): style imitation from a few samples across 400+ authors: better in email and news, weak in blogs and forums
- [Cursor Docs: Rules](https://cursor.com/docs/context/rules): User Rules, project rules as .mdc files in .cursor/rules, AGENTS.md
- [GitHub Docs: Support for different types of custom instructions](https://docs.github.com/en/copilot/reference/custom-instructions-support): personal instructions in Copilot Chat on GitHub.com, repository instructions
- [Ownhand source code](https://github.com/jordan-gibbs/ownhand): occasions, the checker, reply matching and the learning loop

## Install

In Claude Code it is one line. You sign in through your browser, so there is no API key to copy.

```
claude mcp add --transport http --scope user ownhand https://ownhand.dev/mcp
```

For Claude, ChatGPT, Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed, Goose and Cline, see [Connect](https://ownhand.dev/connect). New accounts get $1 of free credit. After that you pay as you go, and each rewrite costs about half a cent.

More answers: [all questions](https://ownhand.dev/answers). Full reference: [Docs](https://ownhand.dev/docs).

---

Page: https://ownhand.dev/answers/make-claude-write-like-me. Index for agents: https://ownhand.dev/llms.txt