Claude Code for marketing: a guide for paid media people who don't code
A plain guide for paid media leads who don't code: what Claude Code is, what it costs as of November 2026, the permission settings to fix first, and a search terms task you can finish in an afternoon.

Claude Code for marketing means using Anthropic's agent to do the file work around your ad accounts: cleaning exports, joining reports, drafting Google Ads scripts and building repeatable checks. You describe the job in plain English, and Claude Code reads the files, writes and runs the code, and asks before changing anything. You never have to write the code yourself.
You probably know the Monday version of this. There are six CSV exports in your downloads folder, a search terms report with a few thousand rows, a LinkedIn demographics file that refuses to line up with the CRM, and a weekly readout due before lunch. None of it is hard, and all of it is slow.
This guide is for the head of paid media or PPC lead who has used Claude or ChatGPT in a browser and wonders whether Claude Code is worth the setup. It covers what the tool is, what it costs as of November 2026, the safety settings to fix on day one, and a first task you can finish in an afternoon.
Claude Code is Claude working inside a folder on your computer
Claude Code is an AI agent that can read, create and edit files and run commands in a folder you point it at, which the ordinary chat window is not built to do. Anthropic's overview describes it as a tool that "reads your codebase, edits files, runs commands, and integrates with your development tools", and it runs in the terminal, in code editors, in the Claude desktop app and in a browser at claude.ai/code.
For a paid team, the word "codebase" is misleading. Your codebase is a folder of exports, a notes file and a few saved instructions. Claude Code treats that folder the way a developer's agent treats source code: it opens every file, writes a small script when the job needs one, runs it and shows you the result. Scalewell puts that to work with paid teams on their own folders, so weekly exports become finished analysis and a person still reviews anything before it reaches an account.
So when do you need it instead of the ordinary chat? The table below is the rule we'd use.
| The job | Use the Claude app chat | Use Claude Code |
|---|---|---|
| A one-off question or a draft of ad copy | Yes | Overkill |
| One small file you can paste or attach | Yes | Either |
| Several exports that need cleaning and joining | Awkward | Yes |
| The same job every week, done the same way | Rewrite the prompt each time | Yes, saved as a skill |
| A job that should run while you're away | No | Yes, as a scheduled task |
| A Google Ads script you'll test and maintain | Drafting only | Yes, with the file kept |
If most of your AI use is the first two rows, stay in the chat and read our guide to Claude for paid media teams instead. Claude Code earns its place once the work involves files and repetition.
You need a paid plan and a short setup
Claude Code needs a paid Claude plan or an Anthropic Console account, because Anthropic's setup page says "the free claude.ai plan does not include Claude Code access." As of November 2026, the Claude pricing page lists these plans with Claude Code included.
| Plan | Price as of November 2026 | Who it suits |
|---|---|---|
| Pro | $20 a month, or $17 a month billed annually | One person testing a few weekly jobs |
| Max | $100 a month | Daily, heavy use by one person |
| Ultra | $200 a month | Heavier use than Max |
| Team | $25 a month per standard seat, $20 billed annually | A paid team sharing settings and skills |
| Enterprise | $20 per seat a month billed annually, 20-seat minimum, plus usage at API rates | Larger companies with security review |
Start on Pro. A few weeks of real use will show whether you hit its usage limits, and you can move up to Max or Ultra if you do.
There are two ways in, and neither needs coding:
- The desktop app. Download Claude for macOS or Windows, sign in and click the Code tab. The overview notes that the app includes Claude Code, so there is nothing else to install. This is the route we'd pick for anyone who has never opened a terminal.
- The terminal. On a Mac, open Terminal and paste
curl -fsSL https://claude.ai/install.sh | bash. On Windows, open PowerShell and pasteirm https://claude.ai/install.ps1 | iex. Open a new window, typeclaude --versionto confirm it installed, then go to your working folder and typeclaude. It asks you to log in through the browser the first time.
The app and the terminal use the same engine, so the CLAUDE.md file, settings and connections you set up in one work in the other.
Five paid media jobs Claude Code does without you writing code
Claude Code is most useful on jobs with a clear input file, a clear rule and an output you can check. These five fit that description.
1. Cleaning and joining exports. Ask it to merge a month of Google Ads and LinkedIn campaign exports into one table with the same column names, convert currencies and flag rows where spend is missing. It writes a short script, runs it and saves the result, so next month it reruns the same script instead of guessing again.
2. Search terms triage. Hand it the search terms report and your rules for what counts as irrelevant, and it returns candidate negatives with the reason for each. The rules themselves are in our negative keywords guide. This is the first task below.
3. LinkedIn demographics against your account list. LinkedIn's Campaign Manager lets you export a demographics report that shows "information about members who interacted with your ad sets and ads", including company names. Claude Code can match those companies against your target account list from the CRM and tell you what share of engagement came from accounts you actually want, which is the number a B2B team needs.
4. Google Ads scripts you can keep. Claude Code can draft a pacing alert or a search term sweep and keep the file in the folder, with a note on what changed in each version. The brief and the four-stage test still apply, and our guide to writing Google Ads scripts with AI covers both.
5. Saved procedures and scheduled runs. Once a job works, save the instructions as a skill so anyone on the team can run it the same way. Our guide to Claude skills for marketing covers which ones to build first. You can then schedule it, which we cover in step 8 below.
The best public example comes from Anthropic itself. Its July 2025 write-up says the Growth Marketing team "built an agentic workflow that processes CSV files with hundreds of ads, identifies underperformers, and generates new variations within strict character limits." A January 2026 follow-up says that for growth marketer Austin Lau, "what used to take 30 minutes per ad now takes 30 seconds", and quotes his first reaction: "I have zero idea what this product is for." Treat those as Anthropic's claims about its own team. We found no independent study of time saved by paid teams using Claude Code, so be wary of anyone quoting a precise figure.
Set the safety rules before your first session
Claude Code can change files and run commands, so the permission mode you work in matters more than any prompt. Anthropic's permissions page lists the modes, and four of them are worth knowing.
| Mode | What it does | When to use it |
|---|---|---|
Manual (default) | Reads without asking, and asks the first time it wants to edit a file or run most commands | Your first weeks |
| Plan | Reads files and proposes a plan without editing anything | The start of every new job |
| Accept edits | Approves file edits in the working folder without asking | Jobs you've already run several times |
| Auto | A background classifier reviews actions instead of you | Once you trust the setup |
Check which mode you start in. Anthropic's permission modes page says that from version 2.1.283, auto mode is the starting mode for interactive terminal sessions. In the terminal, press Shift+Tab to cycle through the modes. In the desktop app, use the mode selector under the prompt box. The permissions page also describes a mode that skips permission prompts entirely, and Anthropic's advice is to use it only "in isolated environments like containers or VMs." A laptop holding client exports is no such place.
Four more rules keep you out of trouble:
- Never paste credentials into a prompt. That includes Google Ads developer tokens, LinkedIn access tokens and CRM keys. If a connection needs them, set it up through the connection's own login, and keep any keys file out of reach with a deny rule.
- Block the files Claude should never read. A
Readdeny rule in your settings file, such asRead(./.env)orRead(./secrets/**), stops Claude's file tools from opening them, according to the permissions page. - Treat downloaded pages and files as untrusted. Anthropic's security page recommends that you "avoid piping untrusted content directly to Claude", because text inside a scraped page can try to give it instructions.
- Keep ad account access read-only at first. Connecting Claude Code to Google Ads or LinkedIn through MCP is possible, and our guides to the Google Ads MCP and the LinkedIn Ads MCP cover the routes. Start with exports, then read-only connections, and leave changes to the account in your hands.
Here's a settings file you can put at .claude/settings.json inside your working folder. It starts each session in Manual mode, blocks reading secret files, and makes Claude ask before running curl or wget commands. The permissions page warns that a rule like Bash(curl *) matches the command as Claude usually writes it and is not a security boundary, so treat it as a speed bump that makes you look twice.
{
"permissions": {
"defaultMode": "default",
"deny": [
"Read(./.env)",
"Read(./secrets/**)"
],
"ask": [
"Bash(curl *)",
"Bash(wget *)"
]
}
}
Your first task: turn a search terms export into a negatives file
This first task takes a Google Ads search terms report and produces a file of candidate negative keywords that you review and import through Google Ads Editor. It's a good first job because the input is one file, the rules are yours, and nothing reaches the account until you accept it.

- Make a working folder. Create a folder called
paid-mediawith a subfolderexports. Open it in the desktop app's Code tab, or typecdfollowed by the folder's path in the terminal and thenclaude. You'll have a folder Claude can see and nothing outside it. - Export the report. In Google Ads, download the search terms report for the last 30 days as a CSV with campaign, ad group, search term, impressions, clicks, cost and conversions. Save it in
exports. You'll have one dated file, such assearch-terms-2026-11-09.csv. - Write the CLAUDE.md file. Claude Code reads a file called
CLAUDE.mdin the folder at the start of every session, as the overview explains. Put your account rules in it once so you never retype them. The template below is a starting point.
# Paid media working folder
## Account
- B2B SaaS, selling to finance teams at companies with 200+ employees.
- We do not sell to consumers, students or job seekers.
- Brand terms: [your brand], [product names]. Never suggest these as negatives.
- Competitor terms are reviewed by hand. List them, don't exclude them.
## How to work
- Exports live in /exports. Never edit the original export. Write new files to /output.
- State the rule behind every suggestion.
- If a column is missing or a number looks wrong, stop and tell me.
- Never invent figures. Every number must come from a file in this folder.
- Ask for a plan first. Switch to Plan mode and paste a prompt like this one:
Read the newest file in /exports. Using the rules in CLAUDE.md, find search terms
that should be negative keywords. Only consider terms with at least 10 clicks or
$50 of cost and zero conversions. Group them by reason: wrong audience, wrong
product, informational, competitor (for my review only). Do not edit anything
yet. Show me the plan, the columns you will use and any rows you cannot classify.
You'll get a written plan and a list of questions before a single file changes.
- Approve, then run. Answer its questions, approve the plan, and let it write two files to
output: a review sheet with every candidate, its spend and the reason, and a separate negatives CSV with the columns Campaign, Keyword and Type. Google Ads Editor's CSV column guide says to enter "Campaign negative" in the Type column for a campaign-level negative. - Review the sheet yourself. Read every row, delete anything you disagree with, and pay most attention to competitor and brand-adjacent terms. Expect to change some of them, and treat each change as a rule to add to CLAUDE.md.
- Import as proposed changes. In Google Ads Editor, choose Account, then Import, then From file. The import guide says imported rows stay as proposed changes until you accept or reject them, and the preview shows only the first 100 rows, so scroll the full list before you keep anything. Then post the changes as you normally would.
- Save and schedule it. Ask Claude Code to save the instructions as a skill at
.claude/skills/search-terms-review/SKILL.md, which the skills page says anyone can then run by typing/search-terms-review. To run it weekly, the desktop app's scheduled tasks run on your machine with access to local files, "but only fire while the app is open and your computer is awake." Cloud scheduled tasks run with the computer off but start from a fresh copy of a repository, so they can't see your downloads folder. For a job built on exports, the desktop task is the right one. Have it prepare the review sheet, and keep the import manual.
Imagine, as a purely hypothetical example, two teams running this. Team A asks for "negatives from this report" in a chat window each week and gets a slightly different answer each time. Team B writes the rules into CLAUDE.md once and corrects them after each review. By week four, Team B's file needs a few edits, while Team A is still arguing with the chat about whether "free" is a negative.
Where paid teams get stuck with Claude Code
Most problems with Claude Code for marketers come from trusting the output before the rules are written down. Here are the ones to watch for.
- Invented numbers in summaries. A total calculated by a script can be checked, while a number typed into a prose summary is much harder to trace. Ask it to show the script or the formula behind any total, and keep the "never invent figures" line in CLAUDE.md.
- Stale exports. If yesterday's file and last month's file sit side by side, it may read the wrong one. Date every filename and tell it to use the newest.
- Auto mode on day one. The classifier is reasonable, but you learn nothing about what the tool does if you never see the prompts. Work in Manual mode until you could predict each prompt.
- Skipped scheduled runs. A desktop task does not fire while your laptop is asleep. The scheduled tasks page says the app runs one catch-up run when it wakes, so a Monday job can arrive on Tuesday afternoon. Put the date range in the prompt so a late run still checks the right week.
- Building an app when you needed a file. Claude Code will happily build a dashboard with a login screen if you ask. A CSV and a short readout are usually enough, and they are easier to check.
Claude Code is the wrong tool if your account is small enough to review in fifteen minutes, if nobody owns checking the output, or if your company has not cleared client data for use with AI tools. Anthropic's security page notes that Free, Pro, Max and Ultra plans fall under its consumer terms and Team and Enterprise under its commercial terms, so check with whoever approves vendors before you put client exports in the folder.
You'll know the setup worked when the Monday job takes minutes, your review edits get fewer each week, and the negatives you add stop coming back as wasted spend in the next report.
Where to start this week
Install the desktop app, make the paid-media folder, and run the search terms task once in Plan mode on last month's export. If you'd like help setting up the folder, the rules file and the first skills around how your team already works, Scalewell does that for paid media teams.
Frequently asked questions
Is Claude Code free for marketers?
Claude Code is not part of the free Claude plan. As of November 2026 it's included in the Pro plan at $20 a month, and in the Max, Ultra, Team and Enterprise plans. You can also use it through an Anthropic Console account, which Anthropic's authentication page recommends for organizations that prefer API-based billing.
Do I need to use the terminal to use Claude Code?
You don't. The Claude desktop app for macOS and Windows has a Code tab that runs Claude Code without any terminal. You pick a folder, type what you want in plain English and approve its actions with buttons. The terminal offers the same engine, and some guides assume it, so a short look at Anthropic's free terminal guide helps.
Can Claude Code make changes in Google Ads or LinkedIn directly?
Claude Code can connect to ad platforms through MCP servers, and some of those servers can write changes as well as read data. We'd start with exports, move to read-only connections, and keep every change behind a human review in Google Ads Editor or Campaign Manager until the setup has run cleanly for weeks.
How long does it take a non-developer to learn Claude Code?
There's no reliable published figure. Anthropic offers free self-paced courses, including Claude Code 101, through Claude Academy, and its own growth marketer built a Figma plugin in about 45 minutes to an hour by Anthropic's account. Expect a first useful task in an afternoon and a dependable weekly routine after a few weeks of corrections.
Is it safe to put client ad data in Claude Code?
It can be, with the right plan and settings. Team and Enterprise plans fall under Anthropic's commercial terms, and consumer plans let you change privacy settings. Keep credentials out of the folder, block secret files with deny rules, and get approval from whoever signs off on vendors before using client data.