Onboarded.

How to set up Claude for your marketing team in four steps: identity, projects, skills, and your stack.

Most marketing teams open Claude, type a request into a blank box, and judge the whole thing by what comes back. Then they decide AI is fine but overhyped.

That's not using Claude. That's using the demo.

A team that sets Claude up properly isn't getting a faster writer. It's getting something that already knows the brand, the audience, and the job before anyone types a word. That's the line between AI-assisted and AI-operated, and it comes down to four things you can set up this week.

[PROOF — drop your receipt here before sending. Example: "This is the exact setup my own marketing operation runs on. Every post, email, and client deliverable starts from these four layers, which is why one person can run the output of a small team." Swap in your real numbers or a client before/after.]

Here's the part most people miss: none of this is technical. There's no code. If you can write a clear brief to a new hire, you can do all four.

Two things before we start

1. Save this and block 45 minutes this week to set it up. Reading it does nothing. The setup is the whole point.
2. Send it to one marketer or founder who's still typing into a blank box and blaming the model.

If it saves someone an hour, forward it. That's how it spreads.

Skip this if you've never opened Claude

You need a paid plan and the desktop or web app. That's it. If you've ever pasted a prompt and gotten back a usable first draft, you're ready, jump to step one. If you've never signed in, do that first, ask it one real question from your week, then come back. The rest of this assumes you've seen what a single good answer looks like and you want that to be the floor, not the ceiling.

1. Define its identity (global instructions)

Global instructions are a standing brief Claude reads before every task, in every chat. Set it once, and it applies everywhere.

This is the most valuable 15 minutes you'll spend. Most people skip it, which is why their output reads like everyone else's. The blank box has no idea who you are. Global instructions fix that permanently.

Go to Settings, then Profile (or "Personal preferences" depending on your app). Paste a brief that tells Claude who it's working for and how you expect it to write.

Start from this and make it yours:

You are the marketing operator for [COMPANY], a [what you do] serving [who you serve].
Our audience: [describe them — role, what they care about, what they already know, what they're skeptical of].
Our positioning: [the one thing you want to be known for].
How we write:
- Plain, specific, confident. Short sentences are fine.
- No buzzwords (synergy, leverage, game-changer, robust, seamless).
- No hedging or throat-clearing. Open with the point.
- Specifics over adjectives. Real numbers, real examples.
- Sound like a sharp practitioner, never a brand account.
When I give you a task:
- Ask a clarifying question if the brief is unclear. Don't fill gaps with filler.
- Give me the work, not a description of the work.
- If something's a bad idea, say so and tell me why.

The "how we write" block is what stops the generic-AI smell. Be specific about what you hate. The more concrete your bans, the less it regresses to the mean.

Pro tip: write the banned-words list from your own pet peeves, not a template. The words you're sick of seeing are the exact ones your audience is sick of reading.

[Screenshot: the global instructions field in Settings, filled in with the brief above.]

2. Give it a workspace (projects)

A project is a dedicated workspace with its own instructions and its own files. Everything you do inside it inherits that context automatically.

Global instructions cover who you are. Projects cover what you're working on. You'll want a few: one for content, one for a specific client or product, one for campaigns. Each one carries its own brief and its own reference material, so you stop re-explaining the situation every single chat.

Create a new project. Name it for the job ("LinkedIn content," "[Client] account," "Q3 launch"). Then do the two things that make a project worth having:

1. Write project instructions for that specific job. Same idea as global instructions, but narrower: the goal of this workstream, the format you want, the audience for this work specifically.
2. Add files. Drop in your real material, past posts that worked, your brand guide, a transcript of a sales call, the product one-pager. Claude reads these as context for everything inside the project.

That second step is what people underuse. A project with three of your best-performing posts in it will match your voice far better than any instruction you could write about your voice.

Project instructions — LinkedIn content
Goal: turn one idea per week into 3 LinkedIn posts that sound like me and book calls.
Audience: founders and CMOs who know AI matters but feel behind.
Format: hook in the first line, one idea per post, a concrete example or prompt, a soft CTA.
I've added my 5 best posts to this project. Match their rhythm, their hooks, and their length. When I give you a topic, draft 3 angles before writing anything full.

Pro tip: feed a project your best work, not your average work. It learns the ceiling you give it.

[Screenshot: a project sidebar showing custom instructions and 3-4 uploaded reference files.]

3. Teach it your repeatable jobs (skills)

A skill is a saved set of instructions for a specific, repeatable task. Instead of re-explaining the job every time, you trigger the skill and it runs the way you taught it.

This is where AI-assisted becomes AI-operated. Anything you do more than twice a month is a candidate: your weekly LinkedIn post, your nurture email, a competitor teardown, a first-draft proposal. You write the recipe once. After that, you ask for the output and it follows the recipe.

Think of it as documenting how you'd train a new hire to do one task, then handing that document to Claude.

Write the skill as a plain instruction file: what the task is, when to use it, the steps, and an example of a great finished version. Here's the shape of one:

Skill: Weekly LinkedIn post
When to use: when I say "write this week's post" or give you a topic for LinkedIn.
Steps:
1. Ask me for the raw material if I haven't given it: the win, lesson, or idea this week.
2. Propose 3 hook angles. Wait for me to pick one.
3. Write the post: strong first line, one idea, a concrete example or copy-paste prompt, a soft CTA.
4. Keep it in my voice (see global instructions). No buzzwords, no hedging.
What good looks like:
[paste one of your best posts here as the gold standard]
Length: 150-250 words. Sentence-case. No emojis in the hook.

The example at the bottom does most of the work. A skill with a real gold-standard example beats one with ten rules and no example.

Pro tip: build your first skill from a task you did well last week. You already have the finished version to use as the standard, so the skill writes itself.

[Screenshot: a skill file open, with the "what good looks like" example highlighted.]

4. Connect your stack

Connectors let Claude read and act inside the tools you already use, your email, calendar, docs, and more, so it works with your real data instead of guessing.

The first three steps make Claude smart about your brand. This one makes it useful inside your actual operation. Once your tools are connected, the work stops being copy-paste. Claude can pull the real numbers, read the real thread, and draft inside the real document.

In Settings, open Connectors and connect the tools your marketing actually runs on. Start with the two or three you live in. For most teams that's your email, your calendar, and wherever your docs live.

Then the prompts change shape. Instead of pasting context in, you point at it:

Read my last 5 emails with [client name] and draft a follow-up that moves the proposal forward. Match my voice from global instructions. Keep it short.

Look at my calendar for this week, find every external meeting, and write me a two-line prep note for each one: who they are, what we last discussed, what I want out of the meeting.

That's the AI-operated layer. The setup reads your world and acts in it, instead of waiting for you to describe your world one paste at a time.

Pro tip: connect one tool, not all of them, on day one. Get one workflow running end to end before you add the next. A single connected workflow you actually use beats five you set up and forget.

[Screenshot: the Connectors settings screen with two or three tools connected.]

Your first 45 minutes

You don't need a weekend. You need one focused session.

Minutes 0-15: Write your global instructions. Who you serve, how you write, what you ban. Save it.

Minutes 15-30: Create one project for your most frequent work. Write its instructions and drop in three pieces of your best past work.

Minutes 30-40: Build one skill for a task you repeat. Use last week's best output as the gold-standard example.

Minutes 40-45: Connect one tool. Run one real prompt against it.

When you're done you have a version of Claude that knows your brand, has a workspace for your main job, can run one task on command, and can reach into one real tool. That's not a chatbot anymore. That's the first hire in an operation.

Where this falls short

It only knows what you tell it. A vague identity brief produces vague output, and no amount of clever prompting fixes a thin setup. The work is in being specific upfront.

It still needs a review pass before anything client-facing goes out. This setup gets you a strong first draft fast, not a final draft you send blind.

Skills take a round or two to get right. Your first version won't nail it. Treat the first few runs as training, correct it in the moment, and fold those corrections back into the skill.

Connectors depend on your plan and, in some companies, on your admin's settings. If a tool won't connect, that's usually a permissions question, not a dead end.

The close

I run this setup every day. The four layers above are how one operation produces the volume of a team, and it's the same thing I build with clients moving from AI-assisted to AI-operated. I'm not theorizing about where this goes. I'm reporting what's already working.

My reason for writing this: marketing is changing fast as it moves from assisted to operated, and I'd rather you set this up now than watch competitors do it first.

If you want this built properly for your team, the brand layer, the projects, the skills, the connected stack, wired to how you actually work, that's exactly what I do. Book a free call and we'll map it.

And if this was useful, send it to one person who's still fighting the blank box.

— Louis

P.S. Don't do all four today. Do the first 15 minutes. Write your global instructions, save them, and watch your next answer come back in your voice instead of a stranger's.