A step-by-step guide. No coding needed. Takes about 3 hours.
An AI assistant that writes LinkedIn posts in YOUR voice. Not generic AI slop. Posts that sound like you actually wrote them.
- A Claude Pro account ($20/month at claude.ai). The free version won't work for this because you need Claude Projects.
- Your brain. You need to know what you sound like when you write.
- About 3 hours for the initial setup. After that, writing a post takes 10 minutes.
- Go to claude.ai
- Sign up or log in
- Click your name (bottom left) → Settings → Subscription
- Subscribe to Claude Pro ($20/month)
- You now have access to Projects
- On the left sidebar, click Projects
- Click Create Project
- Name it something like "LinkedIn Posts"
- You now have an empty project. This is where your skill lives.
This is the hard part. And the most important part. You need to tell Claude exactly how you sound.
Open a text file (or Google Doc, or Notion page) and write answers to these questions:
Who are you?
- What do you do? (job title, company, industry)
- How many years of experience?
- What's your audience on LinkedIn? (who reads your posts?)
- What topics do you post about?
How do you sound?
- Are you formal or casual?
- Do you use humor? What kind?
- Do you curse? How often?
- Do you use emojis? Which ones?
- Short sentences or long ones?
- Do you use bullet points? What style? (▷, -, ✅, numbers?)
Grab 3-5 of your best LinkedIn posts. The ones that got the most engagement. Paste them into the file. These are your voice examples.
What do you NEVER say? Think about phrases that would sound weird coming from you. Write them down.
What do you ALWAYS say? Catchphrases, repeated terms, signature expressions.
Now structure all of this into a file called something like MY_SKILL.md. Look at my LinkedIn_SKILL.md in this repo for the format. You don't need all the sections I have. Start with:
- Voice Profile (who you are, how you sound)
- Phrases you use / phrases you never use
- Post examples (paste your real posts)
- Basic writing rules (formatting preferences, length, structure)
This is what makes your posts not sound like AI wrote them. You can use my anti-ai-writing-guide.md directly. It's universal. The banned words and structure rules apply to everyone.
But also add your own rules:
- Words YOU would never use (even if they're not on my banned list)
- Formatting rules specific to your style
- Things that bug you about AI writing
Save this as a separate file or add it to your skill file.
- Open your "LinkedIn Posts" project in Claude
- Click the Knowledge section (or the paperclip/upload area in project settings)
- Upload your files:
- Your skill file (voice profile, writing rules, examples)
- The anti-AI guide (you can use mine directly)
- Any other reference files (post examples, content calendar, etc.)
In your project settings, there's a field called Custom Instructions (or Project Instructions).
This tells Claude HOW to behave. Not what your voice sounds like (that's in the uploaded files), but what to do when you talk to it.
Here's a starter template you can copy and adjust:
You are my LinkedIn post writing assistant. Follow the skill file exactly.
When I say "write a post about [topic]", write a full LinkedIn post. Don't ask questions. Just write it.
Output the post text ready to copy-paste. No code blocks. No markdown.
After every post, tell me:
- Character count
- Post type
- Whether it passes the anti-AI check
Rules:
- Use my voice from the skill file. Not generic AI voice.
- Never use words from the banned list.
- One thought per line. Posts must breathe.
- Include real numbers and specific details when possible.
- Take a clear stance. Don't hedge everything.
Customize this based on what you need. My full project instructions are in LinkedIn PROJECT_INSTRUCTIONS to paste.md in this repo.
Start a conversation in your project and try:
- "Write a post about [something you recently experienced at work]"
- Read the output. Does it sound like you?
- If not, tell Claude what's wrong: "That's too formal" or "I would never say 'crucial'" or "Make it shorter"
- Each correction helps you figure out what to add to your skill file
Do this 5-10 times. Each time, update your skill file with new rules based on what Claude gets wrong.
If you post regularly, build a workflow:
- Pick your posting days. (I do Tuesday, Wednesday, Thursday)
- Create a content backlog. A simple list of post ideas in Notion, Google Sheets, or even a text file.
- Plan weekly. Every Sunday, pick 3 topics from your backlog.
- Write daily. Open Claude, say "write Tuesday's post about [topic]", review, post.
My full weekly system is in linkedin-weekly-system.md.
Your skill gets better the more you use it. After every post:
- Did Claude use a word you'd never say? Add it to the banned list.
- Did the structure feel off? Add a new rule.
- Did a post perform really well? Add it as an example.
- Did Claude miss your tone? Sharpen the voice profile.
Think of your skill file as a living document. It's never "done."
"I uploaded my files but Claude ignores them." Make sure you're starting conversations INSIDE the project, not in regular Claude chat. The project files only work inside the project.
"The posts still sound like AI." Your voice profile probably isn't specific enough. Add more real examples. Add more banned words. Be more explicit about your tone. "Casual" isn't enough. "Like a voice memo I'd send to a friend about something that pissed me off at work" is better.
"Claude keeps using the same structure." Add a rule like: "Never start with a generic intro. Never end with a summary. Start mid-story. End with the punchline." AI has strong default patterns. You need explicit rules to break them.
"My posts are too long / too short." Add a character target to your rules. "Aim for 1,000-1,300 characters for story posts."
| What | How Long |
|---|---|
| Initial setup (Steps 1-6) | ~3 hours |
| Testing and refining (Step 7) | ~30 minutes |
| Weekly planning | ~15 minutes on Sunday |
| Writing a single post | ~10 minutes |
| Ongoing skill improvements | ~5 minutes per week |
After the initial 3 hours, you're looking at roughly 45 minutes per week for 3 polished LinkedIn posts. Not bad.
Open an issue on this repo or connect with me on LinkedIn.