The Claude Skills That Finally Made AI Write Like Me (And How to Build Yours)
Learn how to build Claude Skills for writing—including Voice DNA, Audience Profiles, and Business Context—so AI output sounds exactly like you. Step-by-step tutorial with prompts.
It’s no secret that one of the core activities of what I’m doing now is writing. And it’s also no secret that Claude Skills have been a game changer for me. Writing is actually where I saw the first major improvements.
So today I’m sharing exactly how I built the Claude Skills I use to write these newsletters with Claude, plus the only method I’ve found that really makes AI write in your voice.
I’ve tested every humanizing prompt out there (spoiler: they don’t work). I’ve tried custom instructions, prompts, feeding it examples of my writing, asking it to “sound more like me”.
None of it worked the way this does. These skills cut my writing time in half.
The part where you go back and forth with Claude trying to get the tone right, the structure right, the voice right. That used to take me hours. Now I go in with my messy draft, and I’m closer than ever to a final version on the first pass.
Instead of telling you how good it is, let me show you.
Here’s a snippet of how today’s article came to life. I started from a 12-page messy draft and a detailed prompt on what I wanted. Claude used all of my writing skills (the ones I’m about to teach you how to build) to turn that into what you’re reading right now:
My raw draft, my voice skills, my audience profile, my business context, all loaded. And the output sounds like me.
Now let me show you exactly how to build yours, step by step. The result will blow your mind.
What Claude writing skills actually do (and why they work)
Before we get into the how, I want you to understand what this changed for me, and what it'll change for you, because this is more than a productivity hack:
It makes AI write like you, not like AI.
You know that flat, overly polished, vaguely corporate tone that comes out of every ChatGPT or Claude draft? That happens because the model has zero context about who you are, how you think, and how you talk.
These skills fix that. They give AI a deep, structured understanding of your voice so the output sounds like something you’d publish.
It cuts your editing time dramatically.
Before I built these skills, I’d spend as long editing Claude’s output as I would writing the draft myself. The back and forth was exhausting, “make it more casual”, “no, that’s too casual”, “add more personality”, “not like that.”
Now? The first output is already 80% there. I go from draft to almost-final in a fraction of the time.
It makes your writing more consistent.
When you have a documented voice, audience profile, and business context, you stop drifting. On a good day, you might write a detailed prompt with all the right context. On a tired Thursday night, you won't. And that's when the output suffers.
The skills hold the standard even when your energy doesn't, because all that context is already there.
It works for any kind of writing.
Even though I built these for long-form newsletters, the process applies to anything: LinkedIn posts, email sequences, sales pages, course materials, Substack Notes.
Because we’re not all newsletter writers, but we all write. And the same problem (AI doesn’t know how you sound) applies everywhere.
And once these are built, you can take it further with a brand design skill that makes everything Claude creates look like yours too. And create branded lead magnets in minutes, straight from articles I'd already written.
Before we start
I built all of this in Claude and I’ll walk you through it that way because that’s my writing workflow.
But the prompts I’m sharing can run in ChatGPT, Gemini, or whatever you prefer. The only difference is where you store the result:
In Claude, you save them as Skills
In ChatGPT, you add them to a Custom GPT or project instructions
In Gemini, you can drop them into a NotebookLM project and reference them from a Gem or a conversation
Use what AI model fits how you work.
If you’re using Claude
You’ll need:
Cowork or Claude Code if your articles run long like mine, they handle larger context without choking on it (like the Chat does)
And if you’re wondering what skills are and how to create a skill in Claude, I already covered that in this article, so I won’t go over it again here.
For each skill, there are two starting points depending on where you are right now.
Path 1: Starting from scratch. You don’t have an existing audience, published writing, or data to work with yet. You'll use guided prompts to help AI build these profiles based on what you know about your goals, your topic, and the audience you want to reach.
Path 2: You already have data. You have subscribers, published posts, testimonials, DMs, survey responses. You’ll feed all of that into AI alongside the prompts, and the output will be dramatically richer because it’s built on real evidence instead of projections.
Both paths produce something useful. Path 2 just gets you further, faster.
I’ll walk you through both for each skill. Start where you are, and improve over time.
If you want the fastest path, I built Amplifiers (the AI Blew My Mind MCP for Claude) that has all three skills (Audience Profile, Voice DNA, and Business Profile) already loaded inside it.
You set it up in two minutes, and then you can just ask Claude to run any of these prompts. It walks you through the whole process without you having to copy-paste any of my prompts below. (It’s way faster and easier!)
So you have two options:
Build it yourself by following the step-by-step instructions below (great if you want to understand how it all works and customize it deeply)
Use the MCP and have it all ready to go immediately
What follows is detailed and step by step for building each skill. I'm giving you everything you need to build this today. This whole process shouldn't take you more than a couple of hours. And those couple of hours will save you hundreds later, while giving you output that sounds like you wrote it.
Skill 1: How to build an audience profile
What this skill does: It tells AI exactly who you're writing for. Their role, their daily frustrations, what moves them from free to paid, how they talk about their own problems, even what makes them unsubscribe. (Mine has five distinct personas, each with their own conversion triggers, content preferences, and the exact words they use to describe why they read.)
What it solves: Without it, you’re asking AI to write for “my audience” with zero specifics. With it, AI knows your reader’s daily frustrations, their decision-making patterns, what makes them subscribe, and what makes them leave.
Here’s a snippet from mine so you can see what this looks like when it’s done:
Path 1: Build your profile from scratch
If you don’t have subscribers yet but you know who you want to reach, these two prompts will help you build a detailed persona.
(Credit where it’s due: the foundation for this process comes from Andy Crestodina, an expert in SEO and digital marketing. I adapted and expanded them for newsletter creators.)
1. Get clear on the basics
Before you run anything, sit with yourself and think through a few things:
Is your newsletter B2B (business professionals) or B2C (individual consumers)?
What role does your ideal reader have?
What industry, company size, or context are they in?
What specific value does your newsletter deliver, and what challenges does your audience face that your content can address?
2. Run the first prompt
Once you have that, replace everything in brackets and run this:
You are a [B2B / B2C] audience strategist and market researcher, skilled at creating reader personas for newsletters.
Create a detailed audience persona for a [role / job title / identity] in [industry / company size / geography] who would subscribe to a newsletter about [topic or domain].
This person works in or is part of a [business type / life context], and is responsible for or personally invested in [goals, decisions, skills, or areas of interest].
They are currently trying to accomplish [specific outcome or progress], but face challenges with [specific problems, constraints, uncertainty, or recurring frustrations].
They are actively seeking a newsletter (and related content or products sold through it) to help them make better decisions, improve performance, or gain clarity in this area.
Answer the following:
1. What are their hopes and goals in subscribing to this newsletter? Focus on how they want to think, decide, or perform differently as a result of reading it.
2. What are their fears and concerns about finding the right source of insight or guidance? Include skepticism from past experiences, fear of low signal-to-noise, wasted time, shallow insights, or misaligned advice.
3. What emotional triggers would prompt them to subscribe, keep reading, or upgrade to paid? Consider moments of confusion, ambition, frustration, curiosity, or fear of falling behind.
4. What criteria do they use to judge whether a newsletter is “worth it”? Include relevance, originality, depth, credibility of the writer, consistency, and signal-to-noise ratio.
5. How does this person typically discover newsletters like this, and what makes them stay? Think about social proof, recommendations, past experiences, and reading habits.
6. What would make them unsubscribe or disengage, even if the topic is relevant?
Describe as well this person’s internal dialogue when they open the newsletter in their inbox.
If you have multiple personas you’re targeting, include them all.
3. Go deeper
Now stay in the conversation. Here’s a follow-up prompt to build a more robust profile. You can paste it in as it is or drop each question one by one:
Using the audience persona above, expand it into a more comprehensive profile of the ideal newsletter reader and buyer.
Add detail in the following areas:
- Day in the life - Describe a typical day or week for this person, focusing on moments where the newsletter topic becomes relevant, frustrating, or top-of-mind.
- Why they value a newsletter like this - From their perspective, what job is this newsletter doing for them? Be specific about how it helps them think, decide, perform, or feel more confident.
- Prior experiences and biases - What other newsletters, content, tools, or products have they tried before? What disappointed them, and what expectations or skepticism does that create now?
- What slows or stops their decision to subscribe or pay - What objections, doubts, or practical constraints get in the way? Think time, money, trust, perceived redundancy, or “I can get this free elsewhere.”
- Decision dynamics - If B2B: who else influences or approves this subscription? If B2C: what internal trade-offs or competing priorities affect the decision?
- Before and after - Describe their situation before subscribing and after reading consistently. Focus on concrete changes in clarity, confidence, decisions, or outcomes.
- What makes this newsletter distinct in their mind - How do they explain to someone else why this one is worth reading or paying for?
4. Turn it into a skill
Once you’re happy with the result:
Based on everything we surfaced about my audience profile, create an audience-profiles skill that we can apply across writing content, marketing, offer creation, pricing, and product decisions. Use the creator-skill skill to package this as a Markdown file that we can reuse whenever we work on anything related to my Substack audience.
Path 2: Build from real data
This is where things get really powerful.
1. Run the same prompts
Follow exactly the first three steps from Path 1. Use the same prompts. They work as the foundation whether you have data or not.
2. Collect everything you have about your subscribers
Before you go deeper, gather what you’ve got. I focused mostly on paid subscribers and fed in:
Testimonials
Messages people sent when they converted to paid
Emails from readers
Data from my onboarding survey in the LAB (over 400 responses with roles, needs, interests)
Attach all of this to the same conversation where you ran the prompts. The more data you give it, the richer the output.
3. Use tools that surface what you can’t find manually
I also used StackContacts, a Substack CRM built by Finn Tropy that gives you audience intelligence you can't easily get on your own: DMs, comments, engagement patterns, and subscriber data, all surfaced right inside Claude.
With it, I could build profiles based on real conversations, real feedback, and real engagement patterns.
4. Compress if needed
What I created from all this data + the prompts was more than 30 pages of insights about my audience.
Way too long for a single Claude Skill, so I had to compress it. I used Cowork for this because it handles larger context.
5. Turn it into a skill
Same as Path 1. Once you’re happy with the result, use the final prompt to package it as a reusable skill.
*The result
My audience profile is built from actual subscriber stories. The pain points come from real DMs. The conversion triggers come from the exact messages people sent when they upgraded to paid.
Collecting all of this takes time. But it gives you information you can’t get any other way. And even if you’re starting with just the prompts and no subscriber data yet, you’ll still get something valuable that improves as you grow.
One more thing worth knowing: skills are active everywhere. In chat, in projects, in Cowork, in Claude Code. So once you build your Voice DNA, Audience Profile, and Business Profile, they automatically apply across all your conversations and projects in the Claude desktop app. You don't need to load them manually.
Skill 2: Voice DNA — The method that makes AI write like you
This is the one that changed everything for me.
What this skill does: It tells AI how you write. Your tone, your rhythm, your sentence patterns, your vocabulary choices, words you never use, how you open and close, how you transition, how you teach, how you handle opinions. Everything.
What it solves: Your audience profile tells AI who you’re writing for. Your Voice DNA tells AI how you sound. Together, they’re the reason Claude’s output sounds like something I’d publish.
To give you a sense of what this produces, I ran the prompt on my most popular articles, and the result was a 4,500-word voice analysis. Clear, thorough, and specific enough that when Claude uses it, the difference is immediate. Here’s a snippet:
And you can build multiple Voice DNA skills for different parts of your writing for extra precision (intros, endings, how-to sections etc.).
Example: I have a separate skill for article intros built with a similar process but focused entirely on hooks and openings.
Why? Because for me, the intro is one of the most important parts of any article. After I built that skill, the improvement in how fast Claude polishes my intro drafts was night and day.
Path 1: You’re just getting started
If you haven’t published yet, you can still build a strong Voice DNA.
1. Let AI interview you
Have AI throw random questions at you about your topic, your worldview, your opinions, your approach.
The key: record yourself talking through your answers using voice input (Wispr Flow, LLM’s voice feature, or even just your phone’s voice-to-text). Let it transcribe how you naturally speak.
This gives AI raw material of your actual voice, not what you think you sound like when you’re “writing”.
Ruben Hassid created a great flow for exactly this, where you interview yourself about how you see the world and speak back to AI. Worth checking out if you want a structured way to do it.
2. Analyze and create your skill
From there, use the prompt I’m sharing in Path 2 below to analyze what came out and create your skill.
This path is more straightforward and produces the richest results.
1. Gather your writing samples
You’ll need 5 to 10 posts. The more, the better. Attach your best-performing posts or your personal favorites, the ones that feel most like you.
You have several options for how to get them in:
Option 1: Copy-paste posts directly into the conversation.
Option 2: If your posts are saved in Google Docs or similar, attach them directly in a new Claude conversation.
Option 3: If you don’t have them stored anywhere, go to Settings → Export all your data. Among the export you’ll find your articles (even drafts) as HTML files, which any LLM can read perfectly.
Option 4: If your go-to LLM is Gemini, add your articles to a NotebookLM project, then use the “add file” feature in Gemini Chat to pull from that project directly.
If you want to build a voice skill for different types of writing (emails, LinkedIn, short-form), use the same process but with different samples. Your article voice is different from your email voice, and the skill should reflect that.
2. Analyze your writing and extract the Voice DNA
Now, time for the prompt that does the heavy lifting.
This prompt analyzes everything. I iterated on it with Claude across multiple sessions over several months. It started basic, and through dozens of rounds of testing and improving, it became something that really, truly respects your voice.
This is what the prompt looks like:
Probably one of the most thought-through and worked-on prompts I’ve ever written, because it was built over months of practice. It has 3,500 words to ensure it covers everything and properly analyzes your writing to extract your voice.
Because it’s too long to include in the article, I uploaded it in a separate doc. Click here to grab it:
