How to Train Claude to Write Like You
Turn representative writing samples into explicit voice rules, then test and refine them until Claude matches your style reliably. This guide walks through the full process—from picking examples to storing approved rules for reuse.
Updated 5 min read
To get Claude to write in your voice, collect two or three pieces of your real writing, extract the patterns that make them sound like you, and write those patterns as explicit rules. Claude cannot read your personality, but it can follow a precise description of your sentence length, word choices and structure. Tools like fenbs keep your approved style rules outside any one chat, so every Claude session you open starts with the same voice brief.
Most people skip straight to telling Claude "write like me" and are disappointed when it produces generic text. The missing step is the extraction work: turning a feeling about your writing into instructions specific enough for an AI to follow.
Pick the right writing samples
Choose two or three pieces you are proud of—emails, blog posts, reports, whatever represents the writing you want Claude to match. Make them roughly the same kind of writing as what you will be asking Claude to produce. Your presentation slides and your personal journal are likely to show different voices.
Longer samples work better than short ones because patterns have room to show themselves. A single-paragraph email is hard to analyze; a 400-word blog post gives Claude something to work from. Avoid pieces written under unusual constraints or from a role you do not normally play.
Extract the patterns that make it yours
Read your samples with these questions in mind:
- Sentence length: Do you write short punchy sentences, longer flowing ones, or a mix?
- Paragraph length: How many sentences per paragraph on average?
- Vocabulary: Do you favor plain words or precise technical ones? Do you use contractions?
- Structure: Do you open with the main point or build toward it? Do you use lists or run ideas through in prose?
- Tone: Formal, conversational, direct, warm? What words give it away?
Write down what you find. You are looking for patterns that repeat across samples, not unusual choices. Two samples that both start paragraphs with the conclusion tell you something real about your style.
Write explicit rules from your patterns
Turn each pattern into a rule Claude can follow. Vague instructions produce inconsistent results. "Sound casual" is hard to apply. "Use contractions; keep sentences under 20 words; start with the main point of each paragraph" is specific enough to test.
A working style brief usually has six to ten rules. Too few and Claude fills the gaps with defaults. Too many and the rules start to conflict with each other. Aim for the set that captures the differences between your writing and something generic.
Paste your samples alongside the rules and say: "Here are examples of the writing style to match. The rules below describe it explicitly." Both together give Claude the pattern and the description at the same time.
Test and revise before you trust it
Give Claude a real writing task using your new style brief. Read the result and compare it to your samples. Mark where it drifts—too formal, too long, wrong vocabulary—and trace each problem to the rule (or missing rule) that caused it.
Revise one rule at a time. If you change everything at once you cannot tell which adjustment made the difference. After two or three revision cycles your rule set usually stabilizes: Claude matches your voice consistently and corrections become rare.
One thing to check: Claude may match your style well in short outputs but drift on longer ones. If that happens, the issue is usually that the style instructions got pushed too far back in the context window. Add a reminder at the end of your style brief: "Apply all the rules above to every paragraph, including later ones." That keeps the instruction active through the full output.
This is worth comparing to the approach for other AI tools. The process of testing and correcting works the same way in ChatGPT; the steps are covered in how to train ChatGPT on your writing style. The general approach to shaping Claude's output in any direction is in how to train Claude AI.
Store your approved style brief for reuse
Once your style brief works, keep it somewhere you can find it quickly. Pasting it at the top of each new session is effective but easy to forget. Claude Projects (in the Claude app) lets you write a persistent instruction set for a project—useful if all your Claude work lives in one place.
If you want the same rules available in other AI apps—in Cursor for code comments, in a shared team session—an external tool that any connected AI can read is more portable. fenbs stores your style rules and sample prompts in Skills and prompts, and any connected assistant reads them at the start of a session over MCP.
Keep your style brief where every AI session can read it
Store your voice rules and best sample prompts in fenbs Skills and prompts. Add the approved decisions—"contractions always," "main point in the first sentence"—to Decisions and rules so they appear first in every AI context. When you refine a rule, update it in one place and every connected assistant picks it up.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account and add a project for your writing work.
- Paste your finished style brief into Skills and prompts. Name it clearly, for example "Voice rules - blog posts."
- Add each core rule to Decisions and rules—state the rule, the reason and when it applies.
- Connect Claude to fenbs from Integrations. Ask it to read the project context and the style prompt before drafting anything.
- When you update a rule after testing, edit the decision and the skill in fenbs. The next Claude session picks up the change automatically.