How to Make Claude Sound Like You: The Complete Guide for 2026

Claude is different. If you've switched from ChatGPT or use both, you've noticed.

The outputs feel more natural. The reasoning is more nuanced. The writing is less... generic.

Claude still doesn't sound like you, though.

You paste in your request, get back something thoughtful and well-structured, and then spend 10 minutes editing it to match how you actually write. The rhythm is off. The phrases aren't yours. It's good AI writing, but it's not your writing.

The solution isn't to accept generic output. Claude has powerful features for voice customization that most users never touch. Used correctly, they can transform Claude from "helpful assistant" to "assistant who writes like me," a personal writing stylist that knows your patterns.

This guide covers everything: how Claude handles instructions differently than ChatGPT, how to use Projects for persistent voice settings, how to write effective system prompts, and where the limitations are.

---

Why Claude Feels Different (And Why That Matters for Voice)

Before we customize Claude, you need to understand what makes it unique.

Claude's Writing Tendencies

Claude has distinct stylistic defaults:

• More nuanced hedging. Claude tends to acknowledge uncertainty and present multiple perspectives. Great for analysis, sometimes too cautious for direct communication.
• Longer explanations. Claude often provides more context than asked. Helpful when you want depth, verbose when you want brevity.
• Natural sentence rhythm. Claude's default output reads more like human writing than ChatGPT's. Less robotic, but still not your voice.
• Ethical awareness. Claude actively considers implications. This can add thoughtfulness, or slow things down when you just need a quick draft.

These tendencies are features, not bugs. But they become problems when they override your own writing patterns.

If you're direct and Claude hedges, that's a mismatch. If you're concise and Claude elaborates, that's friction. If you prefer confident assertions and Claude adds "however, it's worth noting...", you're editing.

How Claude Differs from ChatGPT

The key insight: Claude follows instructions very precisely. This is your advantage. When you give Claude specific voice rules, it applies them consistently.

The challenge: Claude doesn't have a global "Custom Instructions" field like ChatGPT. You need to use Projects or system prompts to maintain voice settings.

---

Using Claude Projects for Persistent Voice

Claude's Projects feature is your primary tool for voice customization. It's more powerful than ChatGPT's Custom Instructions, but requires setup.

What Are Projects?

Projects are persistent workspaces in Claude. Within a Project, you can:

• Set Project Instructions (the system prompt) that apply to every conversation in that project
• Upload Knowledge files that Claude can reference
• Organize related conversations together

Think of a Project as a "custom Claude" configured for a specific purpose, including matching your voice.

Setting Up a Voice Project

Here's how to create a Project optimized for your writing:

Step 1: Create a New Project

1. Open Claude (claude.ai)
2. Click "Projects" in the sidebar
3. Click "+ New Project"
4. Name it something clear: "My Writing Voice" or "[Your Name] Communications"

Step 2: Add Project Instructions

Click "Project Instructions" to open the system prompt editor. This is where you document your voice.

A basic structure:

## Style Profile: [Your Name]

### Baseline Settings
- Formality: [Describe your default level]
- Sentence length: [Short / Medium / Variable]
- Directness: [How plainly you state things]
- Punctuation: [Your distinctive patterns]

### Patterns to Always Use
- [Pattern 1 with example]
- [Pattern 2 with example]
- [Pattern 3 with example]

### Patterns to Never Use
- Never: [What to avoid and why]
- Never: [What to avoid and why]

### Context-Specific Adjustments
When writing to [audience type]:
- Shift to: [specific changes]
- Include: [required elements]
- Avoid: [specific things]

Step 3: Add Knowledge Files (Optional)

Upload examples of your writing as Knowledge files. Claude can reference these to better understand your style:

• 3-5 representative emails you've written
• Blog posts or documents in your voice
• Any writing that exemplifies your style

Label them clearly: "Sample - Email to Client", "Sample - Internal Update"

Step 4: Test in the Project

Start a conversation within your Project. Ask Claude to draft something typical: an email, a message, a short document.

Compare the output to your actual writing. What matches? What doesn't? Refine your Project Instructions based on the gaps.

---

Writing Effective System Prompts for Claude

System prompts are instructions you give Claude before your actual request. In Projects, this is the Project Instructions. In conversations, you can include it at the start.

Claude is precise about following system prompts. The more specific your instructions, the better the match.

The Anatomy of a Great Voice Prompt

A comprehensive voice prompt has five sections:

1. Identity Statement Tell Claude who it's writing as:

You are writing as [Name], a [role] who communicates with [audiences].
Your writing is characterized by [2-3 key traits].

2. Baseline Voice Rules Define your defaults:

VOICE DEFAULTS:
- Formality: Professional but not stiff. Use contractions. No corporate jargon.
- Length: Prefer brevity. Most sentences under 15 words.
- Structure: Lead with the main point. Context comes second.
- Tone: Confident assertions over hedged suggestions.

3. Explicit Patterns Document what you do:

PATTERNS TO USE:
- Start emails with the action or decision, not "I hope this finds you well"
- Use em-dashes for asides—like this—rather than parentheses
- Use "we" for team accomplishments, "I" for personal commitments
- End with clear next steps, not vague closings

4. Anti-Patterns Document what you avoid:

PATTERNS TO AVOID:
- Never use: "Per my last email", "reaching out", "touching base"
- Never use: "Moreover", "Furthermore", "In conclusion" (these sound like AI)
- Never: Start with excessive pleasantries
- Never: Bury the lead in context

5. Context Shifts Define how you adjust:

CONTEXT ADJUSTMENTS:
For executives: More formal, numbers-first, 3 points maximum
For direct reports: More personal, explain reasoning, invite questions
For clients: Benefits-focused, professional warmth, explicit next steps

Example: A Complete Voice Prompt

Here's what a full prompt looks like:

## Style Profile: Alex Chen (Product Manager)

You are writing as Alex Chen, a Senior PM at a B2B SaaS company. Alex communicates primarily via email and Slack with executives, engineers, and clients.

### Voice Baseline
- Formality: Professional but human. Uses contractions. No buzzwords.
- Sentence length: Short by default (8-12 words). Longer for complex context.
- Directness: Very high. States the point first, explains second.
- Punctuation: Em-dashes for emphasis. Minimal exclamation points.

### Consistent Patterns
- Opens emails with the decision/action/news, never with greetings or preamble
- Uses "we" for company/team, "I" for personal ownership
- Presents trade-offs explicitly rather than hiding downsides
- Closes with specific action and timeline: "Let me know by Friday if this works."

### Never Patterns
- Never: "I hope this email finds you well" → just start
- Never: "Per my previous email" → say "Following up on..."
- Never: "Synergize", "leverage", "circle back" → use plain words
- Never: "I think we should maybe consider..." → be direct

### Context Shifts

**To executives:**
- More formal tone
- Lead with metrics and outcomes
- Keep to 3 bullet points maximum
- End with clear ask and deadline

**To engineers:**
- Technical depth welcome
- Acknowledge trade-offs explicitly
- Ask for pushback
- Use precise language

**To clients:**
- Warmer but still professional
- Focus on their outcomes, not our process
- Be explicit about next steps
- No internal jargon

---

Claude-Specific Techniques

Beyond basic system prompts, Claude responds to specific instruction patterns.

Technique 1: The "Never/Always" Framework

Claude follows explicit constraints very precisely. Use this to your advantage:

ALWAYS:
- Start with the main point
- Use active voice
- Keep paragraphs under 4 sentences

NEVER:
- Use "delve", "robust", "leverage"
- Start with "I wanted to reach out"
- Use more than one exclamation point per message

The more specific your ALWAYS/NEVER lists, the more consistent Claude's output.

Technique 2: Formality Levels

Define a scale Claude can reference:

FORMALITY SCALE (1-5):

Level 1 (Casual): "Hey, quick heads up—deadline moved to Friday."
Level 2 (Relaxed): "Quick update: the deadline has been moved to Friday."
Level 3 (Neutral): "I wanted to let you know the deadline is now Friday."
Level 4 (Professional): "Please note the deadline has been revised to Friday."
Level 5 (Formal): "Please be advised that the submission deadline has been adjusted to Friday."

Default to Level 3 for unknown contexts.
For executives, default to Level 4.
For close colleagues, Level 2.

Then in conversations: "Draft this at Formality Level 4" gives Claude a precise target.

Technique 3: Reference Examples

Include a short example in your instructions:

EXAMPLE OF MY WRITING:

"We need two more engineers for Q2. Without them, we miss the mobile launch. Budget is $180K annually.

Let me know if you need more detail, but I'd recommend we move on this before the headcount freeze."

KEY PATTERNS IN THIS EXAMPLE:
- Leads with the need (engineers)
- States consequence (miss launch)
- Provides number (budget)
- Closes with action and reason

Claude uses examples as calibration. Show it your actual writing, then explain what makes it yours.

Technique 4: The Instruction Hierarchy

Claude prioritizes instructions in order:

1. System prompt (Project Instructions)
2. Earlier conversation context
3. Current message instructions

If you need to override your Project defaults for a specific message, be explicit:

"For this email only, use a more formal tone than usual—this is going to the board."

Claude will adjust while maintaining other aspects of your voice.

Technique 5: Multilingual Style Profiles

If you write in multiple languages, Claude handles this well, but needs explicit guidance:

LANGUAGE-SPECIFIC VOICE:

ENGLISH:
- Default formality: Level 3
- Direct, active voice
- Short sentences preferred

JAPANESE:
- Default formality: 敬語 (formal) for external, タメ口 for internal
- Provide context before conclusions (different from English)
- Longer sentences acceptable

FRENCH:
- Default: "Vous" unless told otherwise
- More formal opening than English
- Accept longer constructions

Tell Claude which language and context: "Write this in Japanese, formal, to a client." If you communicate in multiple languages, our free multilingual writing checklist can help you audit consistency across languages before building your profiles.

---

Limitations of System Prompts (The Honest Truth)

System prompts and Projects help. They're better than nothing. Significantly better. But they have real limits.

What System Prompts Can't Capture

Strategic variation. You don't use short sentences uniformly. You use them for emphasis and longer ones for context. A rule saying "prefer short sentences" applies too broadly.

Contextual instinct. You know when to be more formal within a single email, building up to a difficult request. System prompts apply rules uniformly.

Vocabulary depth. You have signature phrases, transitions, and words that appear in specific contexts. A short system prompt can't document all of them.

Audience history. Your relationship with specific people affects how you write to them. System prompts don't know your history with Sarah from Marketing.

What You'll Still Need to Edit

Even with a well-tuned Project:

• Opening lines. Claude's openings often feel generic. You'll adjust to match your typical start.
• Closing lines. Your sign-offs are specific to you. Claude approximates.
• Subject lines. Email subjects are highly personal. Claude guesses.
• Humor and personality. If you use dry humor or specific references, Claude usually misses the tone.

When System Prompts Are Enough

System prompts work well for:

• Rough drafts you plan to edit
• High-volume, lower-stakes communication
• Maintaining general consistency
• A starting point that's close enough to polish

They work less well for:

• Communication where your voice is critical (client-facing, executive, sensitive)
• Writing that needs to be indistinguishable from your organic output
• Multilingual communication with nuanced formality shifts

---

Testing and Refining Your Claude Voice

Don't expect perfection on the first attempt. Voice matching is iterative.

The Testing Protocol

Test 1: Audience Variation

Give Claude the same content to communicate to three different audiences:

1. Your CEO
2. Your team
3. A client

Do the outputs shift appropriately? If they all sound similar, your context rules need work.

Test 2: The Friend Test

Share Claude's output with someone who knows your writing. Ask: "Does this sound like me?"

If they say no, ask what feels off. Use that feedback to refine.

Test 3: The Edit Audit

Draft something with Claude, then track every edit you make. Categorize them:

• Tone adjustments
• Word choice changes
• Structural changes
• Opening/closing fixes

Patterns in your edits reveal gaps in your instructions.

Test 4: The Sensitive Content Test

Ask Claude to draft something difficult: declining a request, giving critical feedback, delivering bad news.

Your voice shows most in sensitive situations. If Claude misses the tone here, add specific instructions for these contexts.

Refinement Cycle

1. Test with a real writing task
2. Note what doesn't match
3. Add or clarify a specific rule
4. Test again
5. Repeat until outputs need minimal editing

Most people need 3-5 refinement cycles to get their Project Instructions dialed in.

---

A Complete Template for Claude Voice Setup

Here's a fill-in-the-blanks template for your Claude Project Instructions:

## Style Profile: [Your Name]

### About
I am a [role] at [organization/context]. I communicate primarily via:
- Email: To [audiences]
- [Channel]: To [audiences]
- [Channel]: To [audiences]

Success metric: [What makes your communication effective?]

Languages: [List if multilingual]

---

### Voice Baseline

- Formality default: [Level 1-5 with description]
- Sentence length: [Typical pattern]
- Active/passive: [Preference]
- Punctuation signature: [What's distinctive]
- Directness: [How plainly you state things]

---

### Always Patterns

1. [Pattern]: [Brief example]
2. [Pattern]: [Brief example]
3. [Pattern]: [Brief example]
4. [Pattern]: [Brief example]

---

### Never Patterns

- Never: [Avoid this] → [Why/what to use instead]
- Never: [Avoid this] → [Why/what to use instead]
- Never: [Avoid this] → [Why/what to use instead]
- Never: [Avoid this] → [Why/what to use instead]

---

### Context Shifts

**For [Audience A]:**
- Formality: [Level]
- Lead with: [What]
- Include: [Required elements]
- Avoid: [What]

**For [Audience B]:**
- Formality: [Level]
- Lead with: [What]
- Include: [Required elements]
- Avoid: [What]

**For [Audience C]:**
- Formality: [Level]
- Lead with: [What]
- Include: [Required elements]
- Avoid: [What]

---

### Output Format Rules

- Emails: [Preferred structure]
- [Format type]: [Preferred structure]
- [Format type]: [Preferred structure]

---

[IF MULTILINGUAL]

### Language-Specific Rules

**[Language 1]:**
- Default formality: [Level]
- Key differences from English: [What shifts]
- Context variations: [How you adjust]

**[Language 2]:**
- Default formality: [Level]
- Key differences from English: [What shifts]
- Context variations: [How you adjust]

---

### Example of My Writing

"[Paste a representative sample - 100-200 words]"

Key patterns visible:
- [What makes this example "you"]
- [Another pattern]
- [Another pattern]

---

The Ceiling You'll Hit

The system above works. A week from now you'll have a Claude Project that sounds noticeably more like you.

Then you'll hit the wall.

The Never/Always list captures rules you know how to articulate. But the patterns that make your writing recognizably yours are mostly ones you've never thought to name. The specific pace you use when you're building toward a conclusion. How your sentence length shifts when you're being direct versus when you're hedging. The function words you overuse without noticing. The punctuation habits you inherited from nobody.

You can't write down what you don't know you're doing.

There's also a hard structural problem with the DIY approach: you're asking Claude to describe your writing patterns, and then pasting Claude's description back into Claude. At no point does anyone actually measure anything. It's qualitative all the way down.

---

What Stylometry Actually Does

Forensic linguists use computational stylometry to identify anonymous authors. It's how the Federalist Papers authorship was resolved. It's how courts attribute disputed documents. The technique doesn't ask a human to describe their style — it measures it algorithmically.

Mean sentence length. Standard deviation. Lexical diversity (how much you repeat yourself). Function word frequencies. Punctuation density. Burstiness — whether your paragraph length varies or stays flat. Metadiscourse markers — how often you hedge versus assert. These aren't adjectives. They're numbers.

My Writing Twin runs this analysis on your actual writing corpus. Not 3–5 samples fed to an AI. A real corpus, measured with a real stylometry engine, producing quantitative fingerprints that no amount of self-description could replicate.

The result shows up as a radar chart: your 6-axis writing fingerprint plotted against AI-generated prose baselines. You can see exactly where your writing diverges from what Claude produces by default — and how far.

---

We Build It for You

This is the distinction that most people miss when they first look at My Writing Twin.

This whole guide — the Never/Always list, the formality scale, the context shifts — is about building your own voice instructions and pasting them into Claude. My Writing Twin does not do that. My Writing Twin is not a guide. It is a service.

You provide the raw material (your writing samples). My Writing Twin delivers the finished product: a production-ready system prompt, already optimized for the platform you're deploying to.

If you choose Claude Projects, you get a Runtime Block formatted and calibrated for Claude's instruction format — ready to paste into Project Instructions. Done.

If you choose ChatGPT, you get a Custom GPT instruction set optimized for GPT's behavior and the 8K character limit — with a one-click copy and a step-by-step setup guide.

If you choose Gemini, you get a Gem-ready instruction set, again with platform-specific formatting and a guided setup flow.

You are not configuring anything. You are deploying a finished product that was built from your actual writing.

The output has two components. First, a Writing Style Profile: a ~3,000-word human-readable analysis of your voice, with annotated real excerpts and a breakdown of how your style shifts across contexts. This is for you — to understand your own patterns in a way you probably never have.

Second, the Runtime Block: the production-ready instruction set. Not a description of your style. An execution-ready voice profile that fires on every generation, across every conversation, without you thinking about it.

The difference between a self-written system prompt and a stylometry-extracted Runtime Block is the difference between explaining how you walk and handing someone your gait analysis.

---

The Time Argument

The Every.to method — the gold standard of the manual approach — takes 2–4 weeks of iterative conversation to produce a usable style guide.

My Writing Twin takes under 5 minutes after you upload your samples.

Same output quality? No. Better. Because 2–4 weeks of asking an AI to describe your writing still doesn't produce quantitative stylometry. It produces a well-organized document of qualitative observations.

---

If You Write in More Than One Language

The manual approach is effectively English-only. Asking Claude to analyze your writing voice in Japanese and your writing voice in French and produce a unified cross-language profile is an exercise in hallucination.

My Writing Twin handles multilingual analysis natively. If you write professionally in English, Japanese, French, and Spanish, the profile captures each voice separately — because your formality, rhythm, and vocabulary don't just translate across languages, they shift. The profile captures those shifts. The Runtime Block deploys them.

This is not a feature most people know they need until they try to set it up manually and realize it's impossible.

---

What to Do

Set up the Project system from this guide. It works, and you'll see the difference immediately.

When you hit the ceiling — when Claude is still smoothing out your edges even with a solid prompt — that's the moment to move to My Writing Twin.

You can start with a free Voice DNA analysis: upload your samples, get the stylometry breakdown, see the radar chart of your writing fingerprint vs. AI baselines. That alone tells you things about your own writing you didn't know.

The Runtime Block — your production-ready deployment package for Claude Projects, ChatGPT, or Gemini — is part of the paid plans. That's where the DIY-to-built gap closes.

Run the same prompt test you ran with your manual setup. The edit rate drops again. That's what the difference between describing your patterns and actually measuring them looks like in practice.

→ Start your free Voice DNA analysis

---

For ChatGPT users, our complete guide to custom GPT instructions covers the same principles for that platform. And to understand the deeper methodology, see how to build custom instructions that analyze your writing patterns. For role-specific guides, see Claude for executives, Claude for developers, and Claude for creative writers. Or explore our full Claude platform page for an overview of how Style Profiles transform Claude's output.