Claude skills for writing: Build a voice that sounds like you

TL;DR:

• Generic prompts produce generic writing because "professional" and "friendly" aren't writing instructions.

• The fix is a reusable Voice DNA packaged as a Claude Skill, a Markdown instruction folder that tells Claude exactly how you structure sentences, choose words, format ideas, and avoid AI clichés. Skills activate dynamically across all your Claude conversations, not just within a single Project workspace.

• Before/after examples are especially powerful because they show Claude the transformation instead of merely describing the desired style.

• Break the workflow into small skills: outline, headline, editing, rather than one giant prompt that tries to do everything.

• Keep the human in the loop: AI can execute your patterns, but your judgment and personal details are what make the writing yours.

• Connect Claude directly to Google Docs (33 tools, 10 triggers) and Notion (45 tools, 13 triggers) via Composio so drafts land where you work, no tab-switching or copy-pasting between tools.

Copying and pasting drafts between Claude and Google Docs can quickly become inefficient. By the time generic phrasing is polished and rewritten, the editing process can take longer than writing the draft from scratch. Instead of treating Claude like a vending machine for finished drafts, use it as a co-writing partner.

The key is to give Claude a clear role: let it handle research, organization, and structure while you maintain creative authority over voice, tone, and judgment. Two things make this approach work: a Claude Skill that carries your style framework across every conversation without re-pasting, and a way to eliminate manual copy-pasting between Claude and the tools where you actually write. Claude Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on a specialised task, activating when needed rather than pre-loading into the context window. This guide covers both, step by step.

Build your first Claude skill

1. Select representative writing samples

Don't dump your entire archive into Claude. Choose several pieces that represent different formats: one long-form article, one short email, one social post, and one client-facing document. This gives Claude a cross-section of your register rather than training it on a single mode.

What to look for in your samples:

• Sentence length patterns: Do you write in punchy, short sentences or longer, flowing constructions?

• Structural habits: Do you lead with the conclusion or build toward it?

• Vocabulary preferences: Which words do you reach for repeatedly, and which do you consciously avoid?

• Formatting defaults: Headers and bullets, or dense prose with minimal visual breaks?

Pick samples you're genuinely proud of, not just recent ones. These become the calibration standard Claude works from in every session.

2. Define your personal writing persona

Once you have your samples, extract the rules explicitly. Don't let Claude infer them passively from examples alone. Write them down in a "Voice DNA" document: a set of rules, formatting preferences, and vocabulary guidelines that defines your unique writing style for the model.

Here's a copy-pasteable Voice DNA template to save as a Markdown file (e.g., voice-dna.md) inside a new Skill folder in Claude. If you also use Claude Projects, you can add the same file there as reference material, but the Skill is what makes the Voice DNA available everywhere across Claude, not only inside one Project workspace:

## Voice DNA: [Your Name / Brand Name]

### Tone dimensions
- Primary tone: [e.g., direct, analytical, warm, dry]
- Secondary tone: [e.g., occasionally playful, never sarcastic]
- What I am NOT: [e.g., not corporate, not academic, not cheerleader-y]

### Sentence structure
- Average sentence length: [e.g., 15-20 words]
- Preferred opening: [e.g., lead with the finding, not the context]
- Paragraph length: [e.g., 3-4 sentences max, one idea per paragraph]

### Vocabulary rules
- Words I use frequently: [list 5-10]
- Words I never use: [e.g., "delve," "testament," "synergy," "leverage"]
- Jargon policy: [e.g., define all technical terms at first mention]

### Formatting defaults
- Heading case: [sentence case or title case]
- Lists vs. prose: [e.g., lists for steps and comparisons, prose for narrative]
- Bold usage: [e.g., label terms only, never mid-sentence for emphasis]

### Signature moves
- [e.g., I open analytical pieces with a specific failure case, then work backward]
- [e.g., I use parentheses for brief clarifications rather than dashes]

### Anti-patterns (what AI drafts do that I need corrected)
- Generic openers: "In today's fast-paced world..."
- Passive voice constructions
- Over-qualified claims: "It's worth noting that..."

Claude loads the Skill's instructions dynamically through progressive disclosure when the task requires them, so the Voice DNA is available across all your Claude conversations, not scoped to a single Project workspace.

Critique before you rewrite

When Claude drifts from your style, don't just ask it to rewrite. Use the "critique before rewrite" method: ask Claude to first analyze its own draft against your style guide, identify specific violations, then generate a corrected version. This feedback loop produces better corrections than a simple "try again" instruction.

A prompt that works:

"Review the draft you just wrote against my Voice DNA style guide in the project instructions. List the top 3 places where the tone or structure doesn't match my guidelines. Then rewrite those sections only, keeping the rest intact."

This targets the phrases that sound generic while preserving the parts of the draft that are working.

Claude skills to refine your style

Claude skills vs. standard prompts

A standard prompt is a temporary sticky note. Claude reads it, generates a response, and forgets everything when the conversation ends. Claude Projects and Claude Skills work differently and serve complementary purposes.

A Claude Project provides static background knowledge, Voice DNA documents, style guides, sample articles, research files, that loads into the context window at the start of every conversation within that workspace. This knowledge persists across all conversations in the Project without re-pasting.

A Claude Skill is a folder of instructions, scripts, and resources that Claude loads dynamically to improve performance on a specialised task. Skills work everywhere across Claude, including outside Projects, and activate through progressive disclosure when a task requires them rather than loading at the start of every conversation.

The structure layers like this: profile preferences apply globally to every Claude conversation you have; project instructions and uploaded documents add static background knowledge that loads into the context window only inside a specific Project workspace; and Skills are dynamic folders of instructions that activate when needed and work everywhere across Claude, including outside Projects.

According to Anthropic's Claude Skills documentation, project instructions apply to every conversation within a workspace, while profile preferences set baseline behavior globally. You don't need to repeat "be concise" in every project if it's already in your profile. Here's how the two approaches compare:

Table 1: Claude Projects and Skills vs. standard prompts

Why AI drafts sound generic

The fear that AI writing always sounds robotic is legitimate. Words like "delve," "testament," and "it's worth noting" are signals that the model is filling space without direction. The fix is specificity in your style instructions, not length.

When Claude is grounded in concrete style rules rather than vague instructions, you'll notice fewer generic phrases to edit out, because Claude is producing on-brief output rather than filler you have to rewrite. The quality gap between "write professionally" and a detailed Voice DNA document is the difference between spending hours on edits versus spending five minutes.

Build a consistent brand tone

Draft your voice and tone framework

Beyond personal style rules, Claude needs to understand your audience. Save this Audience Profile template as a second Markdown file (e.g., audience-profile.md) in the same Skill folder as your Voice DNA. Both files load together when the Skill activates. If you use a Claude Project for a specific client, you can upload the same files there as static reference material, but the Skill is what carries them across all contexts:

## Audience Profile

### Primary reader
- Role: [e.g., independent consultant, small business owner, technical PM]
- What they know: [e.g., familiar with APIs but not low-level code]
- What they want: [e.g., practical steps they can act on immediately]
- What they don't want: [e.g., long preambles, academic citations without context]

### Reading context
- Where they read: [e.g., on a laptop, skimming first then reading in depth]
- How long they'll spend: [e.g., 4-6 minutes for a long article]
- Decision they're making: [e.g., whether to adopt this tool for their workflow]

Add before and after examples

Before/after examples show Claude the transformation you want rather than a description of it. Add 2-3 pairs directly to your style guide:

Before (what Claude produces without guidance): "In today's fast-paced digital landscape, professionals are increasingly seeking ways to optimize their content creation workflows."

After (your actual voice): "Most writers spend more time moving text between tabs than actually writing. Here's a faster setup."

That gap is what you're training Claude to close. Showing the transformation directly is more effective than telling Claude to "be direct."

Build a banned words list

Add a concrete banned words list and stylistic rules to your Voice DNA document:

• Never use passive voice ("it was found that" becomes "we found")

• No semicolons in body copy

• No em dashes (use parentheses or a new sentence instead)

• Banned vocabulary: "delve," "testament," "synergy," "empower," "elevate," "seamless," "robust," "game-changing"

• No rhetorical questions in body paragraphs

• No sentences starting with "It is important to note that"

The more specific your guardrails, the less editing you do after the draft is generated.

Streamlining tasks with custom AI workflows

Speed up writing with custom prompts

Creating a Claude Skill is straightforward: create a folder, add your instructions and any supporting files (scripts, resources, reference documents), and Claude loads them dynamically through progressive disclosure when the task requires them. Each modular instruction below can be its own Skill folder or combined files within one. The key is that Skills activate when needed and work everywhere across Claude without re-pasting.

Instead of one massive system prompt, build modular sub-skills. Each sub-skill is a short, focused instruction for a specific writing task:

• Outline Generator: "Given a topic and target audience, create a hierarchical outline using my heading structure preferences. Lead with the main finding, then evidence, then implications."

• Headline Polisher: "Rewrite this headline five ways. Each version must be under 10 words, use sentence case, and lead with the reader's outcome rather than the method."

• Editing Pass: "Review this draft for passive voice, banned vocabulary, and sentences over 30 words. Flag each issue with a line number and a suggested fix."

Modular skills are easier to update, easier to share, and easier for Claude to follow because each instruction set is narrow and unambiguous.

Keep every task on brand

The practical challenge in long-form writing isn't running out of context. It's maintaining stylistic consistency across sessions when you're working on a project over days or weeks. A running "State Document" in your Claude Project prevents style drift: log decisions from previous sessions (chapter summaries, topic arc decisions, formatting choices) and paste the relevant excerpt as context at the start of each new session. This prevents Claude from resetting your established choices every time a conversation starts cold.

Repurpose one draft across multiple channels

The same core draft can serve multiple channels with minimal effort. A 2,000-word analytical article becomes a Twitter thread opener, a LinkedIn post leading with the main finding, and a five-bullet email newsletter excerpt, all generated in sequence using your Voice DNA as the baseline.

Table 2: Manual workflow vs. Composio-connected workflow

How to link Claude directly to your text editors

Now the workflow shifts from "Claude generates text, you move it around" to "Claude generates text directly where it needs to go." Composio is the action layer that gives Claude a single, governed path into your writing tools: Google Docs, Notion, your CMS, and 1,000+ other integrations from one account. Instead of managing separate connections for each tool, you configure access once and Claude can read, write, and update content across your entire stack from a single instruction.

Sync Claude skills to Google Docs

Composio's Google Docs integration gives Claude 33 tools and 10 triggers covering the full document lifecycle: Create a document, Insert Text into Document, Replace All Text in Document, Create Paragraph Bullets, Insert Table in Google Doc, and Export Google Doc as PDF, among others. Claude can read a document's current content, write new sections directly into it, apply formatting, and save changes without you opening a browser tab.

Once connected, Claude operates inside the document rather than alongside it. You can run an Outline Generator skill, a full section draft, and an Editing Pass all within the same document without switching tabs or reformatting pasted text. Access is configured once during setup, and credentials persist across sessions so you are not re-authenticating every time.

Import Notion notes into Claude

Stop manually copying research notes into Claude. Composio's Notion integration gives Claude 45 tools and 13 triggers: Create Notion page, Append text blocks, Query database, Update Page, and Fetch Notion Block Children, among others. Claude can query your Notion workspace directly, pulling specific page content or database entries it needs for a given draft.

This works through Anthropic's Model Context Protocol (MCP), a shared connector that lets AI models pull in data from other tools without custom one-off integration code built for each one.

Stop these 3 habits from ruining your AI voice

Defining a weak brand identity

The problem: "Write in a professional yet friendly tone" is not a style guide. It's a suggestion Claude will interpret differently every time because it has no concrete anchor.

The fix: Replace vague tone descriptions with specific rules. "Never open with a definition" is actionable. "Use contractions in body copy but not in headings" is actionable. "Lead every article with a problem the reader has already experienced" is actionable. Specific rules produce consistent drafts.

Overcomplicating your instructions

The problem: A 5,000-word system prompt sounds thorough, but dense instructions cause Claude to weight some rules over others in unpredictable ways. Contradictions inside long prompts compound the problem.

The fix: Modular, focused instruction sets. Each skill covers one task with clear input requirements and output expectations. When a skill produces bad output, you know exactly which instruction block to revise.

Skipping the human pass

The problem: Treating Claude's output as final copy rather than a strong first draft. AI handles pattern execution well, but judgment and lived context still require human input.

The fix: After Claude's Editing Pass skill runs, read the draft out loud. Rewrite any paragraph that sounds like it's performing expertise rather than sharing it. Add one specific, personal detail that only you could know. That single addition is what makes a draft sound human.

Troubleshooting your Claude skill configuration

Managing multiple Claude writing styles

Create a separate Skill folder for each client's Voice DNA and Audience Profile so the style is available everywhere you work with that client in Claude. If you also use Claude Projects for that client, upload the same files there as static reference material. The Skill travels with you; the Project stays in one workspace.

Use a clear naming convention. For example: "Acme Corp - Blog Posts" or "Personal Brand - LinkedIn." Each project holds that client's Voice DNA, audience profile, and relevant context documents. Switching clients means switching projects, not re-pasting style guides.

Maintaining your authentic writing tone

A practical editing sequence:

1. Run Claude's Editing Pass skill on the draft first to catch mechanical violations.

2. Read the output out loud for natural flow and rhythm.

3. Rewrite any paragraph that sounds like it's performing expertise rather than sharing it.

4. Add one specific, personal detail or example that only you could know.

Transferring skills to other Claude users

Claude added native skill sharing in December 2025. To share a Skill, the owner provisions access directly from the Skill's settings. Recipients can import it into their own Claude account where it works immediately across all their conversations, including inside Projects. For teams, an internal directory lists available Skills so members can find and add them without a manual handoff. Three things native sharing does not currently cover: there is no use-without-read permission (anyone you share with can see the Skill's instructions), no in-product log of who has used a Skill, and no built-in version history. If your Voice DNA or Audience Profile documents contain sensitive client details, share the Skill with that in mind or strip the sensitive content before provisioning access.

Build your first connected writing workflow today

You now have the framework: a Voice DNA document that defines your style, modular Claude Skills for specific writing tasks, and a connection pattern that eliminates manual copy-pasting between Claude and your workspace. These Skills activate through progressive disclosure across all Claude conversations, including inside Projects, so the workflow you build isn't locked to one workspace.

Start with Composio's free tier (100,000 tool calls per month, no credit card required) to connect your first Claude skill to Google Docs or Notion. Every workflow you build on this foundation compounds. The style guide you write today calibrates Claude for every article you publish this year.

FAQs

How much does it cost to connect Claude to Google Docs using Composio?

You can start for free. Composio's free tier includes 100,000 tool calls per month with no credit card required, and paid plans start at $29 per month.

Do I need coding skills to build custom Claude writing workflows?

No. You can set up your Voice DNA and style guides directly in the Claude Projects UI using plain text.

Can Claude write a 50,000-word book in my voice all at once?

No. Claude generates content in discrete turns rather than all at once, so long-form projects work best chapter by chapter, using a State Document in your Claude Project to maintain style and narrative consistency across sessions.

What is the Model Context Protocol (MCP)?

MCP is a shared connector created by Anthropic that lets AI models securely read data and use tools from other apps. Instead of building a separate custom bridge for every tool, MCP gives any compatible AI client a single consistent way to plug in without extra setup.

Is my writing data secure when I use Composio?

Yes. Composio is SOC 2 and ISO 27001 certified.

Can I use Composio to connect Claude to tools beyond Google Docs and Notion?

Yes. Composio gives Claude access to 1,000+ pre-built integrations. The same account and auth setup extends to any tool in the catalog without additional credential management.

Key terms glossary

Model Context Protocol (MCP): A shared connector created by Anthropic that lets AI models securely read data and use tools from other apps. It gives any compatible AI client a single consistent way to plug in, replacing custom per-app integration code.

Voice DNA: A document containing tone rules, formatting preferences, banned vocabulary, and before/after examples that defines your unique writing style for Claude across every project session.

Claude Project: A dedicated workspace in Claude that loads uploaded documents and custom instructions into the context window at the start of every conversation within it, replacing the need to re-paste style guides or background materials each session. Projects provide static background knowledge scoped to that workspace.

Claude Skill: A folder of instructions, scripts, and resources that Claude loads dynamically to improve performance on a specialised task. Skills work everywhere across Claude, including outside Projects, and activate through progressive disclosure when needed rather than loading statically at the start of every conversation.

Context window: The total amount of text Claude can process in a single session. Large static file uploads consume this space and reduce the room available for the actual writing task.