Claude Code for Marketers: The Complete 2026 Guide

Claude Code is a program on your computer that reads and edits the files in a folder you choose and runs commands from plain-English instructions. Marketers open it for the same reason developers do: when the job is a folder of things rather than one document. This guide is the mechanics, from a developer who runs a content site with it: which jobs fit, when chat or Cowork is the better tool, how to brief it, five workflows with prompts, and what goes wrong.

Why a marketer would open a terminal agent at all

Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." Swap "codebase" for "content folder" and the description still holds. The five jobs where it beats a chat window:

• Bulk content operations across files. Thirty posts need a new CTA block, a renamed product, or a frontmatter field. Chat handles one at a time; Claude Code edits all thirty and shows you the diff.
• Exports and CSV work. A Search Console export, a HubSpot contact dump, an ad-platform report. It writes and runs a small script, produces the chart or the pivot, and leaves the script in the folder so next month is one command.
• Landing pages without an engineer. A campaign page as plain HTML and CSS, in a folder, previewed in your browser, ready for whoever deploys.
• Research through web MCP servers. With Firecrawl or Exa connected, it can scrape competitor pages, search the web, and write a research file with a URL beside every claim.
• Repeatable checks. A brand-voice pass, a link check, a "does every stat have a source" audit, run the same way on every draft with one command.

Anthropic's overview lists "Bulk operations across files" as a thing the CLI is for, with the example of piping a list of files into claude -p. If your week contains none of these jobs, stay in claude.ai.

Claude Code vs claude.ai vs Cowork for marketing

Three surfaces, one subscription, one shared usage pool. The wrong choice wastes an afternoon.

The difference between the last two is what is left behind. Cowork returns documents. Claude Code returns the work, the script that did it, and a record of every change, which matters when the job repeats. The full surface map, including Claude Design, Claude in Chrome, and connectors, is in Claude for Marketing Teams.

Setup in plain language

Two routes: the desktop app (download, sign in, click the Code tab) or the terminal (one install line, then type claude inside a folder). Both need a Pro, Max, or Team subscription; the free plan does not include Claude Code. The current install commands are in Installing Claude Code and this guide will not repeat them.

Two things matter more than the commands. You start Claude Code inside a folder, and that folder is the boundary of what it can see and change. Make one for your content, put your posts, exports, and style guide in it, and start there, never in your home directory. And in Manual mode it asks before every file edit and most commands; read the proposal before you press Enter. The permission tiers, the auto mode paid plans default to later, and the deny rules that hold regardless of what Claude decides are explained in the founders pillar, Claude Code for Non-Developers; read that section first.

A CLAUDE.md for a content repo

Every session starts with a blank memory. CLAUDE.md is the file Claude reads first; Anthropic describes it as "instructions you write to give Claude persistent context" and recommends targeting under 200 lines, with instructions "concrete enough to verify." For a content folder that means the layout, the frontmatter rules, and the editorial rules a new writer would need on day one:

# Acme blog content
 
Posts live in `content/blog/<slug>.md` with YAML frontmatter:
title, description (120-155 chars), date, author, tags.
Exports from Search Console and HubSpot go in `data/`. Never edit them.
Style guide: @brand/voice.md
 
## Rules
- Every statistic needs a source URL in the same sentence. No source, no stat.
- Never use: "game-changer", "unlock", "seamless", "in today's fast-paced world".
- US English. Product is "Acme Ledger", never "the Ledger".
- Do not create new posts unless asked; edit existing ones in place.
- Show the plan before touching more than three files.

The @brand/voice.md line imports your style guide so it loads with the briefing. Run /init and Claude drafts a starting file from whatever is in the folder. For rules that apply only to part of the repo, .claude/rules/ files with a paths: field load only when Claude touches matching files, which keeps a landing-page rule out of a blog-post session. The developer-depth version is CLAUDE.md best practices. It is context, not enforcement: anything that must never happen belongs in a permission rule.

Workflow 1: a brand-voice skill

A skill is a folder with a SKILL.md file describing a repeatable procedure. Claude Code loads it when your request matches its description, or you invoke it as /skill-name. The same file runs on claude.ai (as a ZIP upload) and in Cowork, so a skill is the portable unit for anything you want done the same way every time.

The first one to build is a voice profile, because every other workflow depends on it. brand-voice-profiler reads a set of your best pieces and produces a written brand voice document: tone attributes, sentence habits, vocabulary to use and avoid, before-and-after examples. Put the output at brand/voice.md and import it from CLAUDE.md.

Use brand-voice-profiler on the six posts in content/best/. Write the
profile to brand/voice.md. Include ten banned phrases with the reason
for each, and three before/after pairs taken from these posts.

From then on, /brand-check runs a draft against that file and reports deviations by severity. Building the profile, testing it, and keeping it current is the subject of Brand Voice with Claude Skills.

Workflow 2: an SEO brief from your own exports

Put a Search Console export and a competitor list in data/, then ask seo-content-brief-writer for a brief. Because Claude Code can read the CSV directly and run a script over it, the brief is grounded in your queries rather than a guess about them.

Read data/gsc-last-90-days.csv. Find queries where we rank 8-20 with
at least 200 impressions. Pick the strongest cluster and use
seo-content-brief-writer to produce briefs/<cluster>.md: target query,
related queries from the export, search intent, outline, and the
questions a reader would still have. Cite the rows you used.

The end-to-end pipeline, from export to brief to draft to a pre-publish check, is in SEO Content Workflow with Claude Code. If your team already pays for a content-optimization tool, Surfer and Clearscope both export term lists that drop straight into data/; Surfer vs Clearscope is the head-to-head.

Workflow 3: repurposing across files

Repurposing is a folder problem: one source, many outputs, each with its own length and format rules. content-repurposer takes a long piece and produces the derivatives you specify; in Claude Code the same job is /repurpose, which reads the source from the folder and writes each output as its own file.

/repurpose content/blog/pricing-mistakes.md
Outputs: a 5-post LinkedIn sequence (one file), a newsletter section
under 200 words, and three tweet-length hooks. Apply brand/voice.md.
Do not invent examples that are not in the source.

For the video side, Descript turns a recording into an editable transcript and Opus Clip cuts it into short clips; the transcript is a text file, so it goes through the same repurposing pass. Gamma handles the deck when the output is a presentation rather than a post.

Workflow 4: a landing page without an engineer

A campaign page is files: index.html, styles.css, an image or two. That is the shape Claude Code was built for. landing-page-copywriter writes the copy blocks from a brief; the second prompt turns them into a page.

Use landing-page-copywriter for the Q4 webinar: audience is finance
leads at mid-market SaaS, offer is a live session on close automation,
CTA is registration. Write copy/webinar-q4.md. Then build
pages/webinar-q4/index.html from it: plain HTML and CSS, no frameworks,
mobile first, one registration form. Open it in my browser.

The follow-up email is the same pattern with email-sequence-drafter: a registration confirmation, two reminders, and a replay email, each written to its own file so whoever loads them into the ESP has one file per message. If that ESP is HubSpot, HubSpot Breeze is the assistant on the other side, and Claude Code can talk to HubSpot directly through the MCP server Anthropic's docs use as an example.

Workflow 5: research with Firecrawl and Exa

Claude Code reaches the web through MCP servers, which Anthropic's docs describe as the way it can "read your design docs in Google Drive, update tickets in Jira, pull data from Slack." Two servers cover most marketing research. Firecrawl exposes tools to scrape one page, map a site's URLs, search the web, parse a PDF, and crawl a section. Exa is a neural search engine. Each is one command:

claude mcp add --transport http exa https://mcp.exa.ai/mcp
claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp \
  --header "Authorization: Bearer $FIRECRAWL_API_KEY"

Then research becomes a file with receipts:

Use firecrawl to map acme-competitor.com and scrape their pricing,
features, and changelog pages. Use exa to find their last six months
of coverage. Write research/acme-competitor.md: positioning, pricing
model, recent changes, and a URL next to every claim. Mark anything
you are inferring rather than reading as an inference.

Anthropic's warning applies: "Verify you trust each server before connecting it. Servers that fetch external content can expose you to prompt injection risk." A scraped page can contain instructions aimed at the model; keep research sessions in a folder with nothing sensitive in it. Exa vs Tavily compares the search servers if you want a second option.

The official marketing plugin

Anthropic ships a marketing plugin in its open-source knowledge-work-plugins repository, "primarily designed for Cowork" and, per its README, also working in Claude Code. As of September 2026 the plugin folder holds eight skills: draft-content, campaign-plan, brand-review, competitive-brief, performance-report, seo-audit, email-sequence, and a background content-creation skill that holds templates and headline formulas. Its .mcp.json pre-configures connectors for Slack, Canva, Figma, HubSpot, Amplitude, Notion, Ahrefs, Similarweb, Klaviyo, and Supermetrics; each skill references a category like ~~SEO, and whatever you connect in that category fills the slot.

claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install marketing@knowledge-work-plugins

After that, /marketing:brand-review and the rest are slash commands. The README is candid that these are generic starting points to fork with your own voice and personas. What each skill produces, how it overlaps with the five skills above, and how to customize it is in Anthropic's Marketing Plugin: A Guide. The broader catalog is Claude's knowledge work plugins.

What can go wrong

Hallucinated statistics. A model asked for a persuasive paragraph will produce a persuasive number. The mechanical fix is the CLAUDE.md rule above, "no source, no stat," plus a pass by the content-editor agent, which reads a draft and returns every unsupported claim, every stat without a URL, and every sentence that violates the voice file. Run it before a human editor does, not instead of one.

Voice drift. Across a batch of thirty edits the voice loosens. The profile file and /brand-check exist for this; run the check on the batch, not just the first file.

Publishing AI slop. AI slop is the name for content produced at a volume no one asked for. Google's AI-content guidance says "appropriate use of AI or automation is not against our guidelines" and that using AI "doesn't give content any special gains." Its spam policy names "using generative AI tools or other similar tools to generate many pages without adding value for users" as scaled content abuse. Claude Code makes fifty pages an afternoon's work; it does not make them worth reading.

The search landscape those pages land in has changed as well: AI Overviews answer queries above the results, generative engine optimization and answer engine optimization are the names for writing to be cited by those answers, and llms.txt is a proposed file for telling models where your canonical content is. What is verified about each, and what is vendor noise, is in AI Content and Search in 2026.

Where it sits next to the other tools

Claude Code is not a writing assistant in the sense Jasper and Copy.ai are: those give you templates, a team workspace, and brand voice as a product feature, inside a browser. Claude Code gives you a general agent inside your files, and the voice, the templates, and the checks are things you write down. If your team wants the former, Best AI Writing Tools (2026) ranks them; if the question is the model rather than the wrapper, Claude vs ChatGPT for Writing is the comparison, and Best AI Tools for Marketers (2026) covers the whole stack from research to video.

Plans and usage

There is no separate Claude Code fee; it is included with Pro, Max, and Team, and Anthropic's support docs are explicit that "your usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit." Two things marketers do burn that pool faster than chat: long sessions over many files, and connectors, which the same article calls "token-intensive." Which tier fits a solo content marketer, a small team, or an agency is in Which Claude Plan Should a Marketer Pay For?; the figures live on the plans page it links to.

Where to go next

Install the five skills from Claude Skills for Marketers, in whichever surface you use. Build the voice profile first, because every other workflow imports it. Then pick the job from the list at the top that costs you the most hours and run it once, reading every proposed change before you approve it. Everything on this site for your role is collected at the marketers hub.