Getting Started with Claude Code – for marketers!

—

by

David
in ,

Claude Code is Claude running in your terminal so it can work with files and repeatable workflows. This guide is written for marketers who are curious, perhaps a little apprehensive about technical tools used by software engineers, and want practical wins fast, without “learning to code.”

Terminal is a non-graphical way that you can tell your computer to do things. This guide helps you learn how to use this interface that is essentially keyboard-only. You can still use the mouse to select things, but all of the inputs are pasted or typed, then hit “enter” to run the command.

As we go through this guide, you’ll see text formatted like this which indicates something that you should copy or type into the Terminal, then hit enter to run it. Sometimes these commands will be broken out as a blockquote with one or more commands (each one will be on a separate line):

command1
command --option

Commands are instructions that you’re giving the computer. Hit enter to have the computer follow that instruction.

Let’s get started!

Bold Prediction: sometime in the course of 2026, this guide will become obsolete because a non-Terminal interface supporting work with local files will become available for non-technical users.

Please share your clever uses of Claude Code on social media using #MarketingWithClaude!

Table of Contents

What you’re doing (and what you’re NOT doing)

You’re not “learning to code.” You’re learning:

  1. A tiny bit of command-line actions (like learning a few phrases in a foreign language before traveling to a new country)
  2. How to hand Claude your local files (like data exports from analytics tools, creative images, message content, etc.)
  3. How to ask Claude to analyze, summarize, compare, and draft things for marketing work

Quick Start (~15 minutes)

Step 1) Open Terminal

  • Mac: Applications → Utilities → Terminal
  • Windows: Windows Terminal (or PowerShell)

Terminal is the old-school interface to interact with your computer. It’s often used for things that don’t have or need a graphical interface. In the Terminal, you’ll be using your keyboard almost exclusively.

Tip: when you’re typing a file or folder name in Terminal, you can hit Tab

Step 2) Create a working folder for Claude

In Terminal, type these commands (or copy/paste one at a time) and hit enter after each line:

mkdir -p ~/ClaudeMarketing
cd ~/ClaudeMarketing

You just created a folder and navigated to it! We need to know how to create folders (aka directories) and navigate through them so that we can have Claude run in that folder.

  • mkdir stands for “make directory”
  • cd stands for “change directory”

Note: you can also use your file explorer (like Finder on a Mac) to create folders, but you’ll still need to navigate to it in the terminal

Step 3) Install Claude Code

Follow the instructions in the Claude Code overview.

Step 4) Start Claude Code

Type or copy/paste, then hit enter:

claude

Follow the login prompts given. This command opens Claude in the Terminal, which, once signed in, is where you can enter prompts, just like using Claude on the web at claude.ai.

Step 5) Ask Claude something low-stakes

Try this:

Explain what you can do for a marketing team if I give you data exports from GA4 or Amplitude. Give me 10 examples.

You’re officially started. 🎉

Terminal Basics: just the 6 commands you need

If you only learn six commands, you can do 90% of what you need:

1) pwd — “What folder am I in now?”

pwd

pwd stands for “present working directory”

2) ls — “What’s in this folder?”

ls

ls stands for “list”

3) cd — “Go into this folder”

cd FolderName

cd stands for “change directory”

Tip: as you start to type the folder name, you can hit Tab to quickly fill in the rest of the folder name.

⚠️ Note: if a folder name has spaces, it needs to be “escaped” by adding \ before space, so the computer doesn’t confuse it as another option in the command. Using Tab to fill in the folder name, these escaping slashes will be added automatically.

To go up one level to the parent folder:

cd ..

One dot refers to the current folder, two dots to its parent. Use slashes in the folder structure to go further. The parent’s parent is ../..

4) mkdir — “Make a new folder”

mkdir CampaignQ1

5) cp — “Copy a file”

cp ~/Downloads/ga4_export.csv .

That last dot . means “copy into this current folder,” like mentioned above.

cp stands for “copy”

6) mv — “Move and/or rename a file”

mv ga4_export.csv ga4_260107_traffic.csv

mv ga4_export.csv inputs/ga4/ga4_260107_traffic.csv

mv stands for “move”

Don’t worry about memorizing these commands. Keep this section open the first few times and copy/paste the commands you need.

Folder Structure

Claude gets dramatically more helpful when your files are organized. Think of your folder as a project room full of evidence.

The marketer-friendly project folder template

Create one folder per initiative (campaign, experiment, research study). You can create these folders in Terminal or in your file explorer (Finder or Windows Explorer).

ClaudeMarketing/
  260107_growth-audit/
    brief/
    inputs/
      ga4/
      amplitude/
      mailchimp/
      snowflake/
    working/
    outputs/

Naming rules that save your sanity

  • Use dates as a prefix. I like YYMM, as it sorts nicely. Add the date for even more granularity, e.g., 260107
  • Include folders for the source of the contents within the folder: ga4, amplitude, mailchimp, snowflake
  • Include what is in the file: visits_by_source, email_performance, nps_comments

Examples:

  • 260107_ga4_visits_by_source.csv
  • 260107_amplitude_signup_funnel.csv
  • 260107_mailchimp_email_performance.csv
  • 260107_snowflake_nps_comments.csv

Context: how Claude stays oriented

Most “AI frustration” comes from missing context. Here’s the simple model to ensure you have all context provided for the task:

  • Role: what role should Claude play?
  • Objective: what are you trying to achieve?
  • Context: which files (exports, notes, results) matter?
  • Constraints: guardrails that should be considered
  • Output format: bullets, tables, slide summary, etc.
  • Definition of done: what means that Claude is done with the task?

The easiest context clue: create a CLAUDE.md file

Inside each project folder, create a file called CLAUDE.md and write down common context items that should be consistent in every task. Claude Code reads this file every time. The format is Markdown (which comes with the .md extension), which is a convention of formatting that is simple to read without changing font formatting.

Example CLAUDE.md for marketers

# Project: 260107 Growth Audit

## Context
Our tools:
- GA4
- Amplitude
- Pendo
- Marketo
- Databricks

### Key Moment Definitions
- Activation = completed signup + created first project
- Conversion = request demo OR purchase

## Output format
- Executive summary (5 bullets)
- Findings (by funnel stage)
- Recommendations (impact vs effort)
- What to test next (A/B ideas)

Why this matters: It reduces repetition and keeps Claude aligned across the whole project. You don’t have to input this context with every prompt, like you would otherwise.

How to give Claude your files

In Claude Code, you can reference files and folders directly using @ so Claude can read them without you copy/pasting everything. Claude will also smartly understand requests to review files without a direct reference. One useful indicator is to wrap a filename or folder name in with these characters: `. For example: `260107_growth_audit`.

Analyze a single file

Prompt: “Summarize channel performance and key changes month over month. @01_inputs/ga4/ga4_2026-01_channels.csv”

Compare two exports

Prompt: “Compare these two Amplitude funnels and explain what changed. @amplitude_funnel_dec.csv @amplitude_funnel_jan.csv”

Get oriented in a folder

Prompt: “What’s in @inputs/ and what should I analyze first?”

Model Context Protocol (MCP)

If files are “bringing Claude the receipts,” then MCP is like plugging Claude into your tools directly.

MCP is a standard for connecting Claude to external tools and data sources via “MCP servers.” You can think of it as a connector layer.

Three data access levels

  1. Level 1 (we started here, without MCP): Export → Claude (CSV files, docs, notes)
  2. Level 2 (Read-only MCP integrations): Claude can pull info from connected tools
  3. Level 3 (Action MCP integrations): Claude can do things inside tools (with approval)

Tip: Start with exports (Level 1). Only consider integrations once your team agrees on safety rules and access permissions.

Debugging without tears

If Claude “can’t find a file”

Usually it’s one of these:

  • You’re not in the right folder (use pwd)
  • The filename/path is slightly different (use ls)
  • The file is in Downloads and not in your project folder yet

Try:

pwd
ls

Then copy the file into the right place:

cp ~/Downloads/yourfile.csv 01_inputs/ga4/

Note: you can also just use your file explorer to move files, if that’s easier!

Tip: ~ means the home directory, which on a Mac is /Users/<yourusername>.

If Claude asks for permission

This is normal and good. You stay in control. If it asks to read a file you intended to share, approve it. If not, deny its access.

If the results look wrong

Don’t “argue.” Tighten the request.

  • Ask Claude to list assumptions
  • Ask what information it used from the file
  • Reduce scope (“only analyze these 3 columns”)

One useful way to think about Claude and its capabilities is as a highly-educated, inexperienced intern. Give it the right amount of direction, and it will do a great job.

Tip: Don’t know what to do next? Claude can help you out…ask Claude what to do or for help!

Marketing example prompts

Here are some practical examples about how to get started using Claude Code for marketing.

Example 1: Experiment result evaluation (GA4 or Amplitude)

Export: experiment/variant performance or funnel conversion by variant

Put files in: inputs/ga4/ or inputs/amplitude/

Prompt:

# ROLE
You are a senior data analyst with experience evaluating marketing test results using Google Analytics and Amplitude.

# CONTEXT
- Data exports are in the folders `inputs/ga4/` and `inputs/amplitude`
- This experiment consisted of two groups: a control and variant
- The target KPI is overall funnel conversion, defined as [give a specific definition based on your data]
- Guardrail KPIs include cart abandonment rate ([give a specific definition based on your data]) and signup rate ([give a specific definition based on your data])

# OBJECTIVE
Evaluate the test results, declare a winner, and output executive-level summary results for this experiment.

# HARD CONSTRAINTS (NON-NEGOTIABLE)
- **Use only data in the provided export files.** Do not introduce external data not included in the provided files. Do not use placeholder data. If additional data are needed to accomplish the objective, define a specific data ask.
- Do not guess about the data. If you’re unsure, **read files, inspect, and ask for clarification**.

# OUTPUT FORMAT
- Generate a powerpoint slide summarizing the outcome of the experiment. Include the primary and guardrail KPIs for each variant, an indicator of the winning variant, and commentary about learnings from the experiment and proposed next steps and next experiments to run.

# DEFINITION OF DONE
Only consider the work complete when:
- All data has been analyzed.
- All output documents are fully created and saved (all requirements accounted for).
- You provide a final summary including: what you analyzed, how you approached the analysis, and where to find the output.

Example 2: Copy pre-testing (subject lines, landing pages, ads)

Provide: copy variants in a doc or text file

Prompt examples:

  • Initial prompt
# ROLE
You are a senior marketer with experience in consumer behavior running a focus group with members of the target audience.

# CONTEXT
- Output copy will appear in email subject lines, on landing pages, and in online ads.
- Brand voice guidelines are provided in the file @brandvoice.txt
- The product value proposition is documented in @valueprop.docx
- Target audience personas are provided in @personas.md

# OBJECTIVE
Generate options for marketing copy to use to convince prospects to try and buy this product. Pre-test the copy options with each persona, make edits based on their feedback, and prioritize the options.

## PERSONA SUBAGENTS
Run each persona as a subagent that can respond to your questions and debate with other participants.

# HARD CONSTRAINTS (NON-NEGOTIABLE)
- Copy outputs must conform to the system constraints. Apply best practices for each surface.

# DEFINITION OF DONE
Only consider the work complete when:
- All copy options have been documented and saved to a file
- All best practice research questions have been answered
- Each persona has given feedback about each copy option
- The feedback and recommendations have been saved to a document
  • Follow-up prompt: Generate test variants

    Prompt: “Based on the discussions with the target personas, create 12 subject line variants: 4 curiosity, 4 benefit, 4 urgency. Under 50 characters. Here’s the initial email draft: @email_draft.txt”
  • Follow-up prompt: Align with brand voice

    Prompt: “Rewrite these headlines to match our brand voice rules. Keep meaning, improve punch.”

Example 3: Survey responses – consumer research

Data Export: CSV with open-ended responses, or a text file with one response per line

Prompt examples:

  • Theme clustering

    Prompt: “Cluster responses into themes, estimate frequency, and give representative quotes. Then suggest messaging improvements. @survey_open_ended.csv”
  • Positioning angles

    Prompt: “Propose 3 positioning angles and 3 message pillars based on what people value/complain about. Include proof points. @survey_open_ended.csv”
  • Objections

    Prompt: “Extract top objections and anxieties. Draft landing page copy that addresses each. @survey_open_ended.csv”

Example 4: Campaign setup help — generate assets you paste into tools

Marketo nurture campaign kit (ready to paste)

Prompt: “Create a 4-email nurture sequence for [persona] who did [action] but didn’t [convert]. For each email: subject line, preview text, body copy, CTA, and goal metric. Keep voice [friendly/professional]. Also include send timing, segmentation rules in plain English, and what to A/B test in each email. Optimize the output for Mailchimp.”

Pendo in-app guide kit

Prompt: “Write in-app guide copy for a 3-step walkthrough introducing [feature]. Each step under 25 words, include a benefit, avoid jargon. Also write 2 tooltip variations per step.”

Cross-tool consistency

Prompt: “Ensure the Marketo nurture and the Pendo guide share the same message pillars and terminology. Create a shared mini style guide + phrase bank. @marketo_sequence.md @pendo_guide_copy.md”

The prompt formula that works

This structure provides Claude with all the information to maximize success. Treat it as guidelines because every situation is different. You can even ask Claude to improve your prompt, then use the output with minor edits to get an even better result!

  1. Role: what role Claude should play for this working session
  2. Context: what Claude needs to know about the situation, including any files you’re giving
  3. Objective: what you’re trying to accomplish
  4. Constraints: what should Claude be careful not to do?
  5. Output format: what should Claude output? Be specific…Claude will follow instructions
  6. Definition of done: what constitutes that Claude is done with this task in its entirety?

Copy/paste prompt template (we used it in example 1):

# ROLE


# CONTEXT
- 
- 

# OBJECTIVE


# HARD CONSTRAINTS (NON-NEGOTIABLE)
- 
- 

# DEFINITION OF DONE
Only consider the work complete when:
- 
-

Hope you found this guide helpful!

Please share your clever uses of Claude Code on social media using #MarketingWithClaude!

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *