My 8-Step Substack Writing Workflow (+ How to Turn It Into a Claude Cowork Agent)

The full workflow I use, and how a Claude Cowork agent takes it from 8 steps to one prompt

May 24, 2026

Howdy, Wealth Gang🤠

timo and dj working on a claude writing agent

Today I want to walk you through my full AI workflow to create Substack articles.

AI can save you a ton of time and energy (if you use it the right way) and I’m not super deep on the technical side of it, I just care if the output is good. :)

So today I share the practical workflow I actually use, every single time I publish, and in part 2, my friend Dheeraj Sharma from Gen AI Unplugged breaks down how to level this up into a Claude Cowork agent, so one prompt handles everything.

In this article, you’ll learn:

✓ My full 8-step AI workflow for writing Substack articles from idea to published
✓ The exact tools I use at each step and why I chose them
✓ How to level this up into a Claude Cowork agent that handles everything in one prompt

Let’s goo! 🤠

My 8-Step AI Article Workflow

Step 1: Start With the Idea (Zero AI)

The first thing to know about my workflow is that I don’t use AI for ideation at all.

I already have enough ideas. Always more than enough, honestly.

So when I go into my writing setup, I already have the idea in mind.

The point is: AI shouldn’t decide what you write. You do.

That’s actually one of the things that makes this workflow different from what most people do. A lot of creators outsource their thinking to AI from the very start.

Step 2: Build the Outline in Claude

The LLM I use to write articles is Claude, and in there I have a project, with my Article Architect.

It includes the system prompt I use for my articles, plus two files in the knowledge base: an article example and a target persona reference for the specific person I’m writing for.

What’s a target persona reference?

It’s basically a document that describes exactly who you’re writing for. Their struggles, their goals, how they think. So every article is written with that one person in mind.

Once I’m inside the Article Architect, the first thing I do is build out the full outline of the article.

I don’t start writing yet. I figure out the structure first.

This is where I decide what sections the article will have, what order they go in, and what the point of each section is.

Only once the outline looks right do I move on to actually writing.

Step 3: Write Section by Section

Once the outline is locked, I start writing.

And the way I do it is section by section, not all at once.

Also I don’t use AI for research. I don’t look for studies or outside sources.

I always write from personal experience or personal opinion. That’s my whole philosophy.

So what I do is take whatever context I have saved (a story, a lesson, an observation) and paste it into the Article Architect. Then me and Claude build the section together from there.

Once every section is written and approved, I’ve got the full draft.

That’s when the real finishing work begins.

Step 4: Run the Internal Linker Skill

Once I have the full article draft, the first thing I do is run the Substack Internal Linker skill.

It analyzes the article I just wrote and suggests internal links to articles I’ve already published on Substack.

The way it works is that it’s connected to a Notion database of all my published articles. It looks through all the titles, compares them to what I’m writing right now, and finds the spots where I could naturally link back to something I wrote before.

Why does this matter?

Internal links keep readers on your Substack longer. They discover older articles they missed. And it builds a web of content that makes your whole newsletter feel more valuable over time.

The skill handles all of that automatically. I don’t have to manually scroll through 50 articles trying to remember what I wrote six months ago.

Step 5: Run the Product CTA Linker Skill

After the Internal Linker, I run the Product CTA Linker skill.

Same idea, it analyzes the article draft and finds natural spots where I can add links to my digital products.

But here’s the part I’m actually really proud of.

It automatically creates UTM links.

What’s a UTM link?

It’s basically a tracking tag added to the end of a URL. So when someone clicks a product link inside an article and buys something, I can see track exactly which article sent that sale.

So instead of just knowing “someone bought the Article Architect today,” I know “someone bought it from this specific article.”

That’s the kind of data that helps me figure out what content actually converts.

Step 6: Create the Thumbnail

After the two skills, I create the thumbnail.

And for this I actually use ChatGPT, not Claude.

I have a GPT set up specifically for my Substack thumbnails. I always give it this reference image, and it generates a thumbnail that matches my brand.

Now since the ChatGPT Image 2.0 upgrade dropped, this step got way easier.

Before that I had to reprompt it five, six, seven times to get something close to what I wanted, now I reprompt it maybe once or twice and it nails it. :D

Step 7: Generate the Title

After the thumbnail, I create the title.

For this I use something I call the Headline Hero, basically a prompt where I paste in the full content of the article and it spits out a list of potential titles that work well for Substack.

Step 8: Publish + Turn It Into Notes

Once the title is done, I publish the article.

And right after publishing, I use another GPT I’ve set up, the Article-to-Post GPT.

I paste the article into it and it creates multiple Substack Notes from the content.

Why ChatGPT and not Claude for this?

I compared both and for Notes, I preferred ChatGPT. It just fits better for that style.

So that’s the full split in my stack: Claude for the long-form article writing, ChatGPT for the visual and short-form stuff around it.

The Notes are then used to promote the article after it goes live.

Real Talk: My Workflow Isn’t Perfect

Okay, so that’s my full eight-step workflow and I’m genuinely happy with it.

But… there’s a lot of platform hopping involved.

It works, but it’s not 🏃running🏃 at full efficiency.

Every time I switch platforms and copy-paste something, that’s a small friction point and small friction points add up over a week of publishing.

The other thing is that each tool works in isolation. They don’t talk to each other, so I’m the one doing all the connecting manually.

Is there a better way?

Yeah, and that’s exactly what Part 2 is about. :D

My friend Dheeraj Sharma from Gen AI Unplugged is going to show you how to take this exact workflow and collapse it into one Claude Cowork agent, so instead of hopping between tools, one prompt handles everything.

It’s a level up and more accessible than it sounds.

How to Level This Up With a Claude Cowork Agent

This section was put together with the help of Dheeraj Sharma from Gen AI Unplugged. He builds AI systems and workflows and broke this down in the most practical way possible.

So when DJ looked at my workflow, the first thing he noticed was this:

Every step I’m doing manually (the platform hopping, the copy-pasting, the switching between tools) an agent could do all of that for you.

That’s what a Claude Cowork agent makes possible.

And the way DJ explains it, every agent is built on the same basic three layers.

Don’t let the word “layers” scare you. This is actually very simple once you see it. :)

Layer 1: The Integration Layer

This is how your agent connects to the outside world.

In my case, that means connecting to Notion, where I store all my context (opinions, experiences) for the individual articles.

Think of this layer as giving your agent access to your tools. Without it, the agent is working in isolation with no outside data.

For my workflow specifically, the integration layer would connect Claude Cowork to Notion so it can automatically pull my context without me copy-pasting anything.

This layer answers one question: what external tools does your agent need access to?

Layer 2: The Context Layer

This is where you teach the agent who you are.

Your writing voice, your brand, your target persona, your article examples. Everything that makes your content sound like you and not like generic AI output.

In my current setup, most of this lives inside my system prompt. DJ’s point is that in a proper agent setup, you’d have this organised into dedicated files: a brand voice file, a content strategy file, your target persona reference, your article examples.

The context layer is the hard part. But once it’s built, everything else just works.

Layer 3: The Instructions Layer

This is where your full workflow lives.

Think of it as the operating manual for your agent. Every step I walked you through in Part 1 gets written down here in order. And instead of me doing each step manually, the agent reads the instructions and executes them one by one.

It also knows which skills to use at each step. So when it gets to Step 4, it automatically runs the Internal Linker. When it gets to Step 5, it runs the Product CTA Linker. All without me switching tabs.

This is basically your workflow, just written down so an agent can follow it instead of you.

What the Upgraded Workflow Looks Like

Once all three layers are set up, the whole thing changes.

Instead of eight manual steps across multiple platforms, you give the agent one prompt with your idea and your context.

It builds the outline, writes the sections, runs the Internal Linker, runs the Product CTA Linker, generates title options and even creates a draft ready to paste into Substack.

The One Thing That Makes or Breaks It

The context layer is the hardest part to build. It takes real work upfront to document your voice, your workflow, your examples properly.

But here’s what he said that stuck with me:

“Once that is there, everything is more or less just copy-pasting.”

And yeah, if it would be easy, it would be boring. We want some challenges in our lives. :D

The point is: don’t expect to set this up in an afternoon, but do expect that once it’s set up, your whole content operation gets a serious upgrade.

You don’t need to have everything in place from the start. Try it, optimize as you go, and grow with it.

Takeaway

Here’s the full picture.

Part 1 is the workflow I use right now: eight steps, real tools, built from personal experience. It works and it’s something you can start using today without any technical knowledge.

Part 2 is where this is heading: one agent that handles the whole thing so you can focus entirely on your ideas.

A big thank you to Dheeraj Sharma for breaking down Part 2. Go check out his newsletter Gen AI Unplugged, if you want to go deeper on building AI systems and workflows, that’s your place.

If you liked this workflow breakdown, you will love the Substack Side-Hustle Sprint — a 5-day masterclass I created to go from Substack newbie to having the foundation of a $2K/month Substack business that replaces your 9-5.

Subscribe to Write Your Way To Wealth and get the Substack Side-Hustle Sprint for free.

See ya soon

Timo Mason🤠