Hello Reader
Over the past few months, I’ve watched a lot of marketing teams adopt Claude.
Most start in roughly the same place.
Someone needs a campaign brief, so they ask Claude to write one.
Someone else needs a LinkedIn post, so they open a new chat and start prompting.
A third person uploads campaign data and asks for a summary.
Everyone is technically using AI, but everyone is using it differently.
Different prompts.
Different context.
Different standards.
Different outputs.
And usually, the two or three people who’ve figured out how to get great results from Claude become the unofficial AI help desk for everyone else.
That’s a hard way to scale.
I’ve spent the last few months building a framework that moves teams from random acts of AI to a coordinated system.
That became the 3F Model:
I’ve turned the whole thing into a three-part field guide for marketing leaders.
PS: This newsletter grows from your shares. Sharing is free and helps me create more AI content for marketers.
01. Foundation: Give Claude Context
Most people open Claude and start typing in the chat window without providing any context.
There’s nothing wrong with that, but after 25 interactions you know there has to be a more efficient approach.
Without your ICP, messaging hierarchy, brand standards, campaign plans and examples of good work, it fills in a lot of blanks (and usually not in a good way).
Five minutes of setup can make a significant difference.
Think of it like onboarding a new team member. You wouldn't expect them to produce great work without a proper briefing.
Start with your settings
Run through all the settings and make sure you have them set up the way you want. The most important settings are turning on “connector search” (this allows Claude to connect to all of your external platforms) and “code execution and file creation” (this allows Claude to create and change files for you).
Here's what that looks like in the settings section:
Prompting is still important
One of the quickest ways to improve your outputs is to improve your prompting.
There’s plenty of chatter online about how prompting doesn’t matter anymore, but in my experience, these five principles help me give Claude the context and direction it needs to do the work well:
Create up your Projects
A Project is a workspace with memory.
Load it with your brand guidelines, campaign brief, ICP and messaging hierarchy and every conversation inside the Project inherits that context.
Create a Project for each project, campaign or client you work on.
For example, create a Brand Checker Project and upload your brand and style guide. Then you can share a blog post or ad copy in that Project and Claude will identify any brand violations.
If you’re on the Teams plan, you can also share Projects with your team.
Choose the right model for the task
Anthropic just released Opus 5.5 and Sonnet 5.5, with Sonnet positioned for faster everyday work and Opus for complex tasks that require more careful reasoning.
For marketing work, I’d start with Sonnet 5.5. It's good at drafting emails, summarizing meeting notes or turning approved content into social posts.
I’d choose Opus 5.5 for work with more uncertainty, such as weighing campaign strategies, making sense of conflicting customer research or investigating why performance has changed.
Try both on a task you repeat regularly, using the same brief and source material. Compare how much editing each output needs, whether it follows your instructions and how well it supports its conclusions.
02. Function: Turn Your Processes into Skills
Setting up Claude with context is a good first step but it’s not enough to scale.
Say your best demand gen person has developed a great process for building nurture campaigns. You can capture that process in a Skill so other people on the team can use the same methodology.
Here's what Skills look like in Claude:
A good Skill defines four things:
- Role
- Rules
- Format
- Example
Instead of teaching everyone on the team how to construct the perfect prompt, you're giving them a tested way to do the job.
Use Cowork
When you’re ready for Claude to take action on your behalf, toggle to Cowork.
Just click the Cowork button in the chat window:
Then give Claude a clear task, point it to the files and tools it needs and let it work through the steps.
You can have it pull together research, analyze campaign data or build a first draft, then review the result before you pass it on.
People often ask me what the difference is between Chat and Cowork. Here's the quickest breakdown:
- Chat: Where you work through questions and ideas with Claude.
- Cowork: Where you give Claude a task to carry out across your files and connected tools.
Want to know more about Cowork? Claude just released an update. You can find all the info here.
03. Flow: Let the System Run
By this point, you have context and repeatable processes, but you still have to initiate the work.
Flow is where you start handing that job to Claude.
With Cowork toggled on, tell Claude when and how you’d like to initiate the workflow.
For example:
Every Monday at 8am: pull current performance data → run your Performance Analyst Skill → build a dashboard → post it to Slack.
Or:
Every Friday at 4pm: research current competitor activity → combine it with sales messaging data → produce a competitive intelligence brief → save it to Airtable.
You can adjust the timing of any workflow in the Scheduled Tasks section:
Part 3 of the series includes six complete workflow recipes covering campaign intelligence, performance reporting, content production, brand review, competitive intelligence and SEO/AEO auditing.
Each recipe gives you the Skill, connector, files, trigger, output and Cowork task prompt required to build it.
Build The System
Foundation → Function → Flow.
- Give Claude the context it needs to understand your business.
- Turn the processes your team already uses into repeatable Skills.
- Schedule the work so it runs repeatedly.
Once those three layers are working together, Claude has the infrastructure to get work done for you.
I put the entire system into three practical guides, including the setup instructions, prompts, Skills and workflow recipes.
Start with Part 1 if you're still working primarily in individual chats.
If your Projects are already well configured, Part 2 will help you turn the processes you're repeating into Skills.
And if you already have a working Skills library, Part 3 shows you how to connect those pieces into workflows that can run with far less intervention.
One series. Three guides. The complete system.
Get all three Marketing Director’s Guides to Claude →
Scale this Across Your Team
You can build a beautiful Claude setup for yourself and still have five marketers operating five different systems.
For a team, I’d set it up like this:
1. Create shared Projects around the work
For Team and Enterprise plans, Projects can be shared workspaces.
Give the appropriate people view or edit access so they’re working from the same campaign briefs, ICP documents, messaging and brand context.
This is how you share a Project:
2. Name an owner for each Project
Someone needs responsibility for keeping the source material current, maintaining the instructions and removing outdated documents. Otherwise your AI system gradually starts operating from old information.
3. Centralize your Skills
Decide which processes deserve to become shared Skills and who can create or modify them. I’d start with one or two people building and testing Skills before making them available to the wider team.
4. Give the team a starting point
Don't hand people an empty chat window and tell them to “use Claude.” Tell them which Project to work in and which Skill to use for common jobs.
For example:
- Campaign reporting → Marketing Performance Project + Performance Analyst Skill
- Copy review → Brand Project + Brand Steward Skill
- Email campaign → Campaign Project + Email Campaign Planner Skill
5. Keep human review in the workflow
Especially as you move toward automation. Assign owners, define what needs approval and periodically review the outputs your Skills are producing.
Get all three Marketing Director’s Guides to Claude →