How Do You Automate LinkedIn Content for a Team?
I built a skill file that holds our voice rules, writing frameworks, and real story sources in one place. The AI reads that file and writes first drafts for each team member. I review and approve every post before it goes out. We went from posting twice a week between three people to publishing 12 posts every single week.
Every Monday at Sucana, the same thing happened. I would stare at a blank LinkedIn editor. Victor my co-founder would stare at his.
Vinod my co-founder would stare at his.
Three people who know exactly what they want to say. None of them writing it down.
The problem was never ideas. We talk about marketing, PPC, and building software all day long.
The problem was starting. Opening a blank page, picking a topic, finding the right tone.
By the time you get through all of that, the day is gone and you posted nothing.
I tried ChatGPT first. Typed "write me a LinkedIn post about building a SaaS" and got back a press release written by a robot.
Victor tried it too. His posts came back full of PPC jargon that even I couldn't follow.
The AI wasn't the problem. The instructions were. We were giving it nothing to work with.
So I stopped prompting and started building a system.
Here are the seven steps I followed, in the exact order I did them.
Step 1: Define the Outcome
Before touching any AI tool, I wrote down one sentence.
"I type one command, and 30 minutes later all three people have their posts in Slack and Basecamp."
That became my north star for every decision during the build.
I also defined what each post should look like. Every post is 150 to 250 words. Every hook is one sentence, max eight words.
Every post is first-person and uses real stories.
No press release tone. No invented stories. Write your outcome before you do anything else.
Step 2: Break It Into Small Steps
My one-sentence outcome sounds simple. But there are a lot of moving parts hiding inside it. The trick is breaking it into steps so small that each one is impossible to mess up.
Here is how I broke mine down:
-
Read the content pool and pick four topics per person
-
Write all four posts using voice rules and frameworks
-
Show me the posts for review
-
Save approved posts as files
-
Send each person their four posts on Slack
-
Create a to-do in Basecamp for each person
Six steps. Each one does one thing.
I wrote each step as a plain English sentence. You can do this on a napkin.
Step 3: Go to Claude Code and Plan Your Build
Now take your outcome and your steps to Claude Code. Not to build yet. To plan.
I opened Claude Code and had a conversation. Not "build me a skill."
More like, "here is what I want to happen every Monday. Here are my six steps. What's the best way to structure this?"
Claude Code asked questions I hadn't thought about. How should the posts be structured? What makes a good hook?
Where does the content pool live?
That conversation took about 30 minutes. By the end, I had a clear build plan before writing any skill file.
Claude Code also told me what it needed from me: a voice file and example posts.
Step 4: Create Your Voice File
The planning step showed me what the skill needed. So before building, I created the inputs.
I opened a blank file and started dumping everything about how I sound. Not how I want to sound. How I actually talk.
Tone calibration: "two tequilas on an Ibiza beach party explaining how cool Sucana is."
That one sentence tells the AI more than a page of generic instructions.
Banned words: "leverage," "synergy," "unlock," "empower." Sentence rules: max 20 words, if a 7-year-old can't read it, it's garbage.
Phrases that are mine: "One last dance." "Let's fucking go." "You cannot out-think a market."
These are things I actually say. The AI picks them up and weaves them in.
My voice file ended up at 596 lines. It took about two hours. The more specific your voice file, the better the posts sound.
Step 5: Collect 3 to 6 Example Posts
The voice file tells the AI how you talk. The examples show it what good looks like.
I grabbed six LinkedIn posts I had written manually that got solid engagement. Posts where people DMed me and commented real replies.
I also grabbed a few from people whose style I admired. Not to copy, but to show the AI the energy and structure I was going for.
Save these somewhere the AI can read them. This step takes 20 minutes. Scroll through your LinkedIn, pick your best ones, save them.
Step 6: Build and Test Each Step
Now you build. I went step by step with Claude Code's skill builder.
"Let's build step 1 first. The AI should read my content ideas file and pick four topics for Virgil."
We built it. Tested it. Moved to step 2.
I embedded my voice rules directly into the skill file. Earlier versions referenced external files, and the AI would skip reading them. Putting everything in one file fixed that.
If you want to go deeper on this approach, I wrote a full guide on building AI workflows for your marketing team. I ran the full skill three times before I trusted it.
First run: Victor's posts were too technical. I added a rule that his posts must be understandable to someone outside PPC.
Second run: the Basecamp delivery failed. The skill had placeholder variables instead of real API credentials. I hardcoded the real values.
Third run: everything worked. Twelve posts written.
Three Slack messages sent. Three Basecamp to-dos created.
Step 7: Roll It Out to Your Team
Once the skill works, rolling it out is the easy part. If you want to document your process for the whole team, I covered creating AI SOPs for your agency separately. I run it every Monday. It takes 30 minutes to review all 12 posts.
Each person gets a Slack DM with their four posts. Copy, paste, post. That is literally it.
Before the skill, we posted maybe twice a week between the three of us. Now we publish 12 posts per week. Every week.
The posts are better too. Not because the AI is a better writer. Because it starts from real stories with a clear structure.
We edit from 80% done instead of building from zero.
Victor told me his posts got more engagement after we started. Not because the AI wrote something magical. Because he actually posted consistently for the first time.
Consistency beats brilliance.
What the Delivery Layer Taught Me
Writing the posts is only half of it. Getting them in front of the right person at the right time is the other half.
We tried saving posts to a shared folder. Nobody opened the folder.
We tried emailing them. Nobody read the emails.
Slack DMs worked. A direct message with four posts, ready to copy and paste.
They see it, they grab it, they post it.
The Basecamp to-do adds accountability. It shows up in the weekly task list with a due date. If you haven't posted by Friday, the to-do is still staring at you.
Systems only work when the output lands where people already look. The same principle applies when you pitch AI services to clients: meet them where they are, not where you wish they were.
One Thing I Would Do Differently
I would write the voice file first. Before anything else.
I did it in step 4 because the planning conversation told me I needed it.
But if I started over, I would record myself talking for 30 minutes and transcribe it. Clean that up before even opening Claude Code.
The voice file is the single biggest factor in output quality. My first version was two paragraphs. The posts were generic.
The 596-line version produces posts I barely need to edit. Start there.
Frequently Asked Questions
How long does it take to build a LinkedIn content skill from scratch?
The initial build took about six hours over a few days. Two hours on the voice file, one hour collecting examples, 30 minutes planning, and the rest on building and testing.
After that, the weekly run takes 30 minutes for all 12 posts.
Does the AI write the posts or do I?
The AI writes the first draft based on real stories, your voice rules, and specific frameworks. You review, edit, and approve.
Think of it as starting from 80% done instead of a blank page. The human always makes the final call.
Can this work with ChatGPT instead of Claude Code?
The preparation steps are the same: voice file, examples, frameworks, real content sources. ChatGPT can handle the writing part.
The delivery automation needs a tool that can execute commands. That is where Claude Code or a workflow tool like Make comes in. I compared AI marketing automation tools that actually work if you want a full breakdown.
What if the posts don't sound like me?
Your voice file needs more detail. My first version was two paragraphs and the output was generic.
The current version is 596 lines. It covers everything from sentence length to banned words to how I describe problems. More specificity means better output.
How do I handle different team members with different expertise?
Define each person's domain in the skill. I handle marketing. Victor covers PPC.
Vinod writes about software development.
The voice rules apply to everyone, but the topics and framing shift based on who the post is for.
What writing frameworks work best for LinkedIn posts?
We use three. How-to posts framed as "How I did this" in three steps. Personal stories as raw narratives.
And industry takes with an opinion on trends.
Every post has one idea, one hook, and ends with a question.
How do I make sure the AI doesn't make up stories?
Hard rule in the skill file: never invent. The AI pulls from two sources, a content ideas file and a notes folder.
If no source exists for a topic, it skips that topic. When referencing someone else's work, it frames it as "I read this." Never pretending something happened to you.
What was the biggest mistake during the build?
Leaving placeholder variables in the skill file instead of real values. The Basecamp integration had generic placeholders where the account ID and API token should be.
The AI got confused and tried to open a browser instead of using the API. Once I hardcoded real credentials, it worked instantly.
How many posts per week should each person write?
We settled on four: two how-to posts, one story, one industry take. That covers Monday through Thursday.
Consistency matters more than volume. Pick a number your team can sustain week after week.
Can I use this approach for content besides LinkedIn?
The same steps work for any repeated content task. Email newsletters, Twitter threads, blog outlines, client updates.
The ingredients are identical: a voice file, a structure, examples, and real content. We started with LinkedIn because it was our worst blank-page problem.
