I posted an infographic this week called Claudemaxxing. The premise: two people pay for the same Claude plan and one gets 10x the output. The comments made it clear a lot of people recognize themselves in the wrong column.

The infographic listed 8 things. This email breaks down all 8, how they actually work, and what they look like inside a real marketing operation — mine.

1. Projects, Not Chats

This is the one that explains most of the gap.

When you open a blank chat, Claude starts with zero context. It doesn’t know your brand, your offer, your customer, what good copy sounds like from you, or what words you hate. So it gives you the average of everything it’s ever read. That average is what people mean when they say Claude sounds generic.

A Project fixes this by loading context once — your voice, your offers, your ICP, your writing rules — and carrying it into every conversation inside that Project. You never re-explain yourself. Claude reads a brief, not your history.

Here’s what I load into my marketing Project: two pages of voice notes with real examples of copy I’d actually publish, an ICP document with the specific objections my buyers raise, service descriptions in plain language, a list of words I never want to see (Claude is instructed to flag “leverage,” “unlock,” and “elevate” before they reach me), and 3–4 past campaign examples Claude can reference instead of making things up.

That setup took about two hours. It has saved me that two hours every single week since.

2. Model to the Stakes

Not every task deserves the same model. Using Opus for a first-draft subject line is like taking an Uber Black to the corner store. Using Sonnet for a client-facing proposal when the contract is worth $30K is the opposite mistake.

My actual rule: Sonnet handles volume work — social posts, email drafts, briefs, repurposing. Opus runs when the output will be seen by a client, used in a pitch, or represents my thinking publicly. The cost difference between them adds up fast if you’re not paying attention, and so does the quality ceiling if you use the wrong one.

The practical version: decide which outputs are high-stakes before you start the week, and open Opus only for those.

3. Make It Ask First

The single fastest way to cut revision cycles.

Most people give Claude a task and judge the first output. When it misses, they rewrite the prompt. Then rewrite again. Three drafts later they have something close.

The faster path: instruct Claude to ask what it needs before it writes anything. I have a line in several of my Project instructions that says exactly this: “Before you start, ask me the three questions that would most change your output.” Claude then surfaces the gaps — audience, tone, constraint, whatever it’s missing — and fills them before the first draft exists.

Three questions beat five redrafts. Every time.

4. Repeat Prompts → Skills

If you’ve typed the same instruction twice, you’ve done unnecessary work.

A Skill is a saved prompt template — a reusable command that carries your format, your rules, and your context without you re-explaining them. I have Skills for my weekly LinkedIn posts, my nurture emails, client briefs, workshop curriculum sections, and a few internal review processes.

The practical test: if you find yourself writing “write this in my voice, use sentence case, no bullet points, no AI words” before every request — that’s a Skill. Write it once, name it something you’ll remember, save it in your Project instructions or as a document Claude can reference. You stop typing format instructions and start giving Claude only what changes.

Over a 12-function marketing workflow, this is where the hours stack up fastest.

5. Claude Design for Visuals

Claude Design (the built-in design tool, previously called Artifacts) is genuinely underused.

I’ve built brand guides, pitch decks, lead magnets, landing page layouts, and infographic structures inside it. The workflow: describe what you need, iterate in conversation, export. No Canva subscription required for early-stage work. No back-and-forth with a designer for a first draft.

The “built once, reused forever” part matters most for templates. I have a Claude Design template for my workshop slides that I update, not rebuild, for every new cohort. A deck that used to take half a day to restructure now takes 20 minutes.

It won’t replace a designer for finished production work. But for 80% of the visual work in a marketing operation, it’s faster and cheaper than anything else.

6. Wire the Stack with MCP

MCP (Model Context Protocol) is how Claude connects to your tools directly — Google Drive, Gmail, calendar, Notion, your CRM.

Without it, you paste context into Claude. With it, Claude reads the context itself.

Concrete example: I used to copy-paste my calendar, email threads, and project notes every time I wanted Claude to help me prep for a client call. Now Claude pulls that itself when I ask. The prep happens in a fraction of the time and Claude has everything, not just what I remembered to paste.

The connectors that made the biggest difference for me: Google Drive (so Claude can read my briefs and docs without me uploading them), Gmail (for drafting responses in context), and Calendar (for scheduling and meeting prep). If you’re running any kind of account-based work, your CRM is the fourth one worth wiring up.

Setup is a one-time hour. The return is every conversation you’d have otherwise spent pasting walls of context.

7. Hand Cowork Outcomes

Cowork is Claude’s agentic mode — you give it an outcome, it plans and executes the steps.

The difference between this and a regular chat: you don’t direct every move. You describe the deliverable. Claude figures out what it needs, runs the steps, and surfaces the output.

“Prep my pipeline review for Thursday” is a Cowork task. In a regular chat, that’s six separate asks — pull the deals, sort by close date, flag stalled ones, check last contact, draft talking points, format it for the meeting. In Cowork, it’s one.

This matters most for multi-step work that has a clear output but messy middle steps. Reviews, reports, research, prep docs. If you’d normally spend 45 minutes gathering and formatting before you can actually think — that’s the Cowork use case.

8. Build with Claude Code

Claude Code is where you go when the output is software, not text.

I use it for internal tools my marketing operation runs on: a lead tracking dashboard, automated tagging logic for my email list, a custom brief generator that pulls from my docs. None of these required me to hire a developer. I described what I needed, Claude Code built it, and I’ve been using it for months.

The distinction the infographic makes is important: not just drafts. Code means production. If there’s a repetitive process in your marketing stack that you’ve been doing manually because “we’d need a developer for that” — Claude Code is worth an hour of your time to test.

The pattern across all eight

None of this is prompt hacking. It’s architecture. The people I see running real leverage on Claude have made upfront investments: they’ve loaded context, set up templates, connected tools, and defined what “done” means before they ask for anything. Then the asks are fast and the outputs are usable.

The people who aren’t getting value are starting from zero every time. They’re typing the same setup into every chat, using the wrong model for the task, and doing manually what could run itself.

Same plan. Different operation.

I work with marketing teams to build this kind of setup end to end — the Project architecture, the Skills library, the tool connections, the workflows. If your team is using Claude but not getting the output you expected, that’s usually a setup problem, not a Claude problem. I can spot the gaps in about 30 minutes.

If that’s worth a conversation, book a free call here:

— Louis