Claude Code is not only for developers. Give it a rules file, a file for each marketing job, a few saved playbooks and a reviewer, and it drafts ads, pages, emails and reports the same careful way every day.
Keep three things with a person: spending money, publishing, and anything about a patient. Write those into the rules file, and back the first two with hooks. Want the folder ready-made? Get our free starter kit.
The short versionWhat is Claude Code, and can it do marketing?
Yes, if you set it up for marketing. Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools," available "in your terminal, IDE, desktop app, and browser" (Anthropic's Claude Code overview, read November 13, 2026). Swap "codebase" for "a folder of marketing files" and you have the idea.
Out of the box it knows nothing about your business, your customers or the rules your industry plays by. The work is in the setup: what it reads first, what it is allowed to do, and who checks it. That is what the rest of this guide covers, using the system our agency runs on every day.
The partsWhat goes inside a Claude Code marketing system?
Seven parts. Each one is a plain file or setting you can read and change.

Read the transcript
Here's what's actually inside a Claude Code marketing system. Claude Code is an AI agent that works inside a folder of your files. It reads them, edits them, and runs your tools. It starts with one file called CLAUDE.md. That's your rules, and Claude reads it at the start of every session. Then each job gets its own instructions like a real team, one for ads, one for SEO, one for email, one for reporting. Skills are saved playbooks, and Claude only loads one when the task needs it. Subagents are specialists, like a reviewer that checks the work before it goes out. Hooks are rules that run every single time, not only when the AI remembers. Connectors plug it into Meta, Google Ads, Analytics, Clarity, Ahrefs, and Canva with your sign-in. And scheduled tasks can run a morning job in the cloud, even with your laptop closed. That's a lot to set up right. It's what we do. Book a consultation at peloramarketing.com.
- A rules file (CLAUDE.md). Anthropic's docs call it a file "that Claude Code reads at the start of every session." Ours holds who we serve, how work moves, and what is never allowed. Keep it short: it loads every time.
- One file per job. We call them seats: ads, SEO, email, reporting. A task names its seat, and Claude works to that file and nothing else.
- Skills. Saved playbooks Claude loads only when the task needs one. Anthropic's docs describe skills as a way "to package repeatable workflows your team can share." Ours cover research, ad copy, pages, email, reports and creative briefs.
- A reviewer subagent. A second agent with one job: read the draft as the customer would, and say READY or NOT READY with the exact fix.
- Hooks. In Anthropic's words, hooks "let you run shell commands before or after Claude Code actions." We use them for rules that must hold every time, not only when the AI remembers.
- Connectors. The Model Context Protocol is "an open standard for connecting AI tools to external data sources." It is how Claude reads your ad accounts, analytics and documents, with your own sign-in.
- Scheduled tasks. Anthropic's cloud scheduled tasks "run in the cloud, so they keep running even when your computer is off." Ours start each morning with a check of what the platforms changed overnight.
Day to dayHow does the work move from idea to live?
The same four steps for every task, whether it is a search ad or a monthly report.
- Research first. Read today's platform help page and the best live examples before building. Platforms change their rules without telling anyone, so a page read last month is not good enough.
- Draft into a file. Ads, pages and emails go into a content folder, never straight into an account.
- Review as the customer. The reviewer reads the draft cold. Anything it marks NOT READY gets fixed and read again.
- A person approves. Claude says what is ready and what needs a yes. A person publishes, sends and spends.
In practice that looks like: "You are in the ADS seat. Write three search ads for our main service." Claude reads the seat file, runs the research skill, writes the ads, sends them to the reviewer, and hands back a file and a short list of what a person must click.
GuardrailsWhat should a person always approve?
Three things, every time: anything that spends money, anything that goes public, and anything that touches a customer's personal information. Write those into the rules file, then back what you can with hooks, because a hook runs whether or not the model remembers the rule.
Two more rules save the most trouble. First, every fact has one home: prices, results and claims live in one facts file, and a number with no source is not used. Second, never let it guess. When the facts file is silent, Claude asks.
Get the folder ready-made
The PELORA Marketing Starter Kit is the setup in this guide: a rules file, four seats, six skills, the reviewer, two safety hooks and a morning report. Free, open source, no email needed.
Get the free starter kitPaid mediaCan Claude Code run your Google and Meta ads?
It can help run them, and it should not run them alone. Through connectors it can read campaigns, search terms and results, and draft new ads and keyword lists. We connect ad accounts with read access wherever the platform allows it, and a spend guard hook goes by each connector tool's name: anything not named as a simple read, like get_ or list_, waits for a person's yes, which covers a tool named for a budget, a bid or a campaign's status. Names can mislead, so after you connect an account, have Claude list its tools and add any that change things to the guard's list (the kit's README shows how). It watches connector calls only, not scripts, so never save a password or API key that can make changes in the folder.
Where it shines is the work people skip: reading every search term each week, checking each ad against the platform's current policy page, and writing the change log nobody keeps. Where it needs a person is judgment on money: how much to spend, and when to stop.
Health businessesIs Claude Code safe for healthcare marketing?
For marketing work, it can be, with one hard line. Anthropic's Claude Code legal page says a business associate agreement with Anthropic "extends to Claude Code in two configurations": its HIPAA configuration or Zero Data Retention. It also says cloud sessions are not covered under the HIPAA configuration (read November 13, 2026). So keep patient information out unless your organization has that agreement and runs a covered setup, and every tool or connector it uses is covered too.
Ads, pages, email to people who opted in, and reports built from platform numbers do not need patient information. Health ads also carry their own platform rules on claims and on what an ad may imply about the viewer. Our kit adds a health claims list: a draft that uses a listed phrase is blocked, and Claude has to rewrite it. Have a person review health copy too. The list does not make anything compliant, and nothing does on its own.
Next stepShould you build it yourself or have it built?
Build it yourself if you have someone who enjoys setting up tools and has a few afternoons. Start with the free kit, fill in your facts, and add one rule to the rules file every time something goes wrong. That last habit is how the system gets better.
Have it built if you want it running across ads, search, email and reporting without learning it first. We set up the full system inside your business and teach someone on your team to run it. See how we build your marketing in-house with AI.

Read the transcript
Tired of marketing agencies promising you the moon? You're not the only one. And here's the twist: we're a marketing agency. Plenty of businesses still want us to run everything, and we do. But some owners want their marketing in-house, with AI. So we build it for you. A complete Claude Code system for your marketing, set up on your computer and saved in your own GitHub. It connects to the platforms you already use, through MCP connectors and APIs. Like Meta Ads. Google Ads. Google Analytics. Search Console. Ahrefs. Microsoft Clarity. Local Falcon. Apify. Canva. And your inbox. Inside, it's organized like a real team. Ads. SEO. Content. Influencers. Reporting. Each with its own instructions, skills and checks. Every morning it can pull your numbers, write the report, and flag what needs fixing. It drafts your ads, your posts and your creator briefs. It tracks your marketing finances: budgets, cost per lead, and ad spend. And a person on your team approves anything before it goes live. For health businesses, patient data stays out unless a signed BAA covers every tool. It runs in your accounts. You own it. We train your team, and hand it over. Built by an agency that runs on this same system every day. Ready to bring your marketing in-house? Book a consultation below.
Rather have it built?
Book a free 15-minute strategy call. We look at how your marketing runs today and tell you what we would set up first.
Book a Strategy CallQuestionsFrequently asked questions
Do I need to know how to code to use Claude Code for marketing?
No. You type requests in plain English. Some surfaces start from a terminal, and Anthropic's docs link a terminal guide "if you haven't used a terminal before." The desktop app needs no command line at all. The web version works from a GitHub repository (an online copy of a folder), so you would put the kit there first.
What do I need to get started?
Anthropic's overview says "most surfaces require a Claude subscription or Anthropic Console account." Then a folder with your rules and facts. Our free starter kit gives you that folder ready to fill in.
How long does it take to set up?
An afternoon gets the basics working: your facts, your voice, and a first ad or page through the reviewer. A full system across ads, search, email and reporting takes longer, because the rules get written as real work shows what they need to say.
Can Claude Code publish or send on its own?
It can be connected to tools that publish and send, which is why we do not let it. In our setup it drafts and a person publishes, sends and spends. A hook goes by each connector tool's name: anything not named as a simple read, like get_ or list_, waits for a person's yes.
Is PELORA affiliated with Anthropic?
No. PELORA Marketing is an independent agency that uses Claude Code. We are not affiliated with or endorsed by Anthropic. Claude and Claude Code are trademarks of Anthropic.
What we checked: Anthropic's Claude Code overview and its legal and compliance page (both read November 13, 2026), against the setup PELORA runs today. Not affiliated with or endorsed by Anthropic.
Last updated November 16, 2026. By Preston Durnford. Newport Beach, California.