Hello Reader

Claude was a long-time cult favorite, beloved by developers and largely invisible to everyone else.

Then, late last year, it swept across the AI conversation and is now one of the most popular AI platforms in the world, regularly trading places with ChatGPT at the top of the charts.

People and teams love it.

And almost everyone I talk to is confused by it.

That confusion has a specific cause: Claude was originally designed for coders, and much of the language and systems are still visible in the platform.

As a marketer, that leaves us scratching our heads. What's a Bash command? Why does Cowork need to work on local folders? What's a Skill?

A few weeks ago I ran a live lightning lesson and 400+ marketers and Claude-curious folks showed up gushing about the platform. "Life changing!" "My fav coworker!" "Phenomenal partner." "At work I'm a superstar." "Spend 90% of my work day in Claude."

But they're also confused.

"I’m excited, but daunted by the complexity." "Still haven't unlocked the true potential." "I am more confused than ever!"

This edition is the answer to all of that.

Well, at least some of it — you can't explain Claude in a 1,000-word newsletter, but I can clear up some of the confusion, and point you in the right direction for the rest.

#1 Does the model really matter?

People ask me constantly which Claude model they should use.

My answer is almost always: Sonnet.

I run Sonnet for about 90% of my work because it handles the vast majority of tasks well and doesn't obliterate my tokens the way Opus does.

Haiku I use for lightweight, high-volume tasks where I need speed over depth.

It is worth knowing that Anthropic just launched Claude Opus 4.7 and it’s now at the top of the model leaderboard for performance.

But even though it’s topping the charts, is it really that useful for your work?

Well it depends.

If you're doing competitive analysis on a 60-page document, building a multi-step strategic brief or working through a long client deliverable that requires Claude to hold a lot of context and make nuanced judgment calls, Opus is the right tool.

It's also noticeably better at creative and professional work quality — interfaces, slides and documents it produces tend to be more polished.

The other improvement worth flagging: Opus 4.7 is better at file system-based memory. For teams using Claude on longer multi-session projects, it retains relevant notes across sessions and requires less re-prompting to get back to speed on a task.

That said, the cost per task is meaningfully higher.

If you have an unlimited token budget, go for it.

If you're watching usage, Sonnet 4.6 is still the right default for most marketing work.

​
​#2. What’s the difference between Projects and Skills?

This was the most-asked question in the session, so let’s clear it up.

A Project is a persistent context container.

Think of it as a standing brief for a specific body of work. You load it with brand guidelines, past campaigns, reference documents, a system prompt that gives Claude context about who it's working for and what it should know.

Projects work best for ongoing work where the context is consistent: a client account, a product line, a brand initiative. Every conversation inside that Project starts with that context already loaded.

A Skill is a reusable instruction set for a specific task.

"Write copy in my brand voice." "Build a weekly briefing from these sources." "Review this for brand compliance."

Skills load automatically when Claude detects you need them, regardless of which Project you're working in.

The simplest way to think about it: Project = the who and what context. Skill = the how-to-do-it instruction.

You can and should use both together.

A client Project with a brand voice Skill loaded on top means Claude has the account context and the voice instructions every time, without you having to re-explain either.

If you want to build this setup for your team from scratch, the Claude for Marketing Leaders workshop on Maven walks through exactly this.

#3. When should I use Chat vs. Cowork?

Think of Chat for conversations: brainstorming, drafting, quick research, iterating on a piece of copy or anything where you're working through an idea and the output lives in the conversation.

I use Chat 90% of the time.

Cowork is for execution. Projects that require multi-step workflows, file orchestration, working across folders and documents or building repeatable systems that connect to external tools.

One important constraint: Cowork is only available on the Claude Desktop App, and it works with local files, not shared drives.

​
​#4. Should I use Claude vs. ChatGPT/Copilot/Perplexity?

Everyone has their favorite large language model and many of the tools overlap, but they're each strongest in different places.

Here’s a breakdown:

Claude is strongest for workflow orchestration, long-form reasoning and context-heavy work. If you're building a system that needs to hold a lot of information, follow a structured process and produce high-quality documents, Claude is where I'd put that work. You can also create Artifacts — interactive dashboards and apps.

ChatGPT covers similar ground to Claude for most tasks. The meaningful differences: it has native image generation, and it doesn't have a Cowork-equivalent for local file and workflow orchestration.

Copilot is the right choice if your organization runs on Microsoft. It integrates directly with Outlook, Teams, Word and Excel, etc. You can also access Claude and ChatGPT models inside Copilot, which gives you flexibility without leaving the Microsoft environment.

Perplexity is a research and search tool, not a workflow tool. It's excellent for fast web research with citations. I use it when I need to pull current information quickly and want sources I can verify. For anything beyond research and discovery, I switch to Claude.

#5. How do I manage drift and hallucinations?

A hallucination is when Claude generates something that isn't accurate — a statistic it made up, a quote that doesn't exist, a fact it got wrong. It happens when Claude is working from weak or missing source material and fills the gap with plausible-sounding content.

To fix hallucinations give Claude strong source material and require it to cite or verify its claims. Use prompts that ask Claude to flag when it's uncertain rather than fill in the gap. Always verify your outputs.

Context drift is when Claude starts forgetting the instructions you gave it at the beginning of a conversation. You notice it when outputs stop matching the brief, tone shifts or Claude starts making assumptions that contradict earlier instructions.

If this is happening, you’ve hit your context window (or memory limit).

To fix drift keep your conversation length under 150,000 words and use Projects to store your context and knowledge docs. It’s more efficient than loading them into every chat.

#6. What’s a realistic token budget for a marketing team?

Anthropic doesn't publish a token calculator for departments. Anyone who gives you a precise number without knowing your workflows is guessing.

Most token waste comes from poor workflow design, rather than heavy usage. The two biggest culprits: long-running chats that re-read the entire conversation history with every message, and Projects loaded with documents that don't need to be there. Both create context sprawl, where Claude is processing far more than it needs to for each task.

Model selection is the main lever on cost.

Haiku for high-volume, low-complexity tasks: subject line generation, content tagging, quick lookups.

Sonnet for most marketing work: drafting, analysis, research synthesis, campaign planning.

Opus for complex strategy work that genuinely requires it. A team that defaults everything to Opus will run up a bill quickly.

Here’s Claude’s documentation on managing usage for development teams, but some of the same concepts apply for non coding users.

#7. What's the best way to roll out Claude to my team?

It's one thing to use Claude yourself, getting your entire team set up is a different challenge. Here’s how to do it:

Step 1: Define the operating model. Clarify where Claude will be used in your marketing workflow. Identify the highest-impact use cases by role — content, research, campaign planning, reporting, SEO, customer insights.

Step 2: Build a small champion group. Select three to five high-agency team members across functions. Train them first on workflows, prompting and quality control. Have them document wins, templates and repeatable processes. This group becomes your internal benchmark.

Step 3: Roll out workflow by workflow. Introduce Claude through real work, not generic training sessions. Start with one repeatable workflow per team. Pair every AI workflow with human review standards and a clear definition of what "good output" looks like.

Step 4: Operationalize and scale. Create a shared library of prompts, Projects and Skills. Share them with the team so you're not rebuilding context every time someone new starts using the system.

Note: sharing functionality for Projects and Skills requires Claude Teams or Enterprise, and your IT or admin team needs to activate it.

Track adoption, time savings, output quality and business impact. Expand successful workflows into standard operating procedures.

If you want to stop figuring Claude out by trial and error, Build a Claude Marketing System in 2 Hours is the shortcut. Projects, Skills, team setup — all of it, with real workflow examples. On-demand, $79.