Claude Code for Marketers: 7 Workflows With Copy-Ready Prompts

Claude Code is useful for marketers because it works on real files on your computer: it reads a folder of CSV exports, visits competitor pages, drafts copy into documents and builds reports you open in a browser, then saves everything where you can find it. Below are seven marketing workflows, each with a copy-ready prompt, what a good result looks like and what to check before you use it. None of them requires coding, only clear inputs and the habit of checking numbers.

Where Claude Code beats a chat window

A chat window is built for one conversation at a time. Claude Code is built for work that involves files, steps and repetition, which describes a lot of marketing.

Set up a marketing workspace once

Create one folder for marketing work and open it in Claude Code: in the desktop app, pick the folder in the Code tab; in a terminal, move into the folder and type claude. This layout fits the workflows below:

marketing/
  CLAUDE.md
  brand/      voice.md, proof.md, utm-rules.md
  data/       exports you download each week
  research/
  briefs/
  report/
  seo/
  utm/

Then add a short CLAUDE.md, the file Claude Code reads at the start of every session:

# Marketing workspace
- We are [company]. We sell [offer] to [audience].
- Read brand/voice.md before writing any copy.
- Claims and numbers come only from brand/proof.md or files in data/. If something is missing, write [TBD].
- Never edit or overwrite files in data/. Save new work in the matching folder.

For fuller templates, see how to write a CLAUDE.md file. Depending on your permission mode, Claude Code asks before it fetches a website or runs a script for the first time, so approve what matches the task.

How the seven workflows feed each other

Each workflow stands alone, but the customer research makes every later step better:

Workflow 1: Competitor research with sources

Use it when you need a fair picture of three or four competitors without spending an afternoon on their sites.

Research these competitors: [SITE 1], [SITE 2], [SITE 3].
For each, read the home page, the pricing page and one product page.
Build a comparison table: who they target, main promise, prices (quote the page, or write "not listed"), offer or guarantee, main call to action, and the proof they show (reviews, client logos, numbers).
Then list three gaps we could own, each tied to something specific you saw.
Put the source URL under every claim. Save as research/competitors-[MONTH].md.

A good result: one column per competitor, prices quoted exactly as written, and gaps that each point back to a row in the table.

Check: open two source links and confirm the prices, and make sure missing data says "not listed" rather than a guess. For more than four sites, add "use a separate subagent for each competitor" so the research doesn't crowd the main conversation.

Workflow 2: Customer language from reviews or surveys

The words customers use are the best raw material for copy. The @ sign points Claude at a specific file, the way you would tag a colleague.

@data/reviews.csv has customer reviews with the columns date, rating and text.
Group the reviews into 6 to 10 themes. For each theme: a short name, how many reviews mention it, how many of those are 1 or 2 stars, and two quotes copied exactly.
Count with a script, not by estimate, and print the totals.
End with ten phrases customers use that we could use in our copy.
Save as research/customer-language.md.

A good result: a theme table, word-for-word quotes, and a phrase list that sounds like your customers rather than like marketing.

Check: search the CSV for one quote to confirm it is real. One review can mention several themes, so ask Claude to note that theme counts may add up to more than the number of reviews.

Workflow 3: An ad brief your designer can use

Briefs fail when they describe a mood instead of a decision. This prompt forces the decisions.

Read our product page at [URL], @brand/voice.md and @research/customer-language.md.
Write a creative brief for a designer making three static ads for [CHANNEL].
Include: the single idea, the audience and their main objection, three headline options under [40] characters with the count next to each, the visual direction in plain words, what must appear (logo, offer, call to action), and what to avoid.
Save as briefs/ads-[PRODUCT].md.

A good result: a one-page brief where every headline shows its character count and the objection comes from real customer language.

Check: count one headline yourself and confirm the offer matches the product page. Any claim in a headline must appear in proof.md.

Workflow 4: Landing page copy with a live preview

Copy reads differently on a page than in a document, so this workflow gives you both.

Write landing page copy for [OFFER] aimed at [AUDIENCE].
Sections: headline, subheadline, three benefits, how it works in three steps, objections with answers (use the themes in @research/customer-language.md), proof (only from @brand/proof.md), call to action.
Give two headline directions and one sentence on why each could work.
Then build preview/index.html, a plain mobile-friendly page showing the copy in context. No images, no tracking code. Open it in my browser when it's ready.

A good result: a copy document plus a simple page you can scroll through, with both headline options labeled.

Check: narrow your browser window to see the phone layout, and confirm every proof line exists in proof.md. When the copy is approved, build and publish the website with AI.

Workflow 5: A weekly report from CSV exports

Most weekly reporting is the same downloads, the same math and the same summary. Keep last week's exports in data/previous/ so Claude can compare.

This week's exports are in @data/: ads.csv, traffic.csv and email.csv. Last week's are in @data/previous/.
Step 1: describe each file (columns, date range, number of rows) and wait for my OK.
Step 2: write a script that calculates spend, clicks, conversions, cost per conversion and the change from last week, by channel.
Step 3: build report/week-[DATE].html with five summary bullets, one table by channel and one simple chart.
Step 4: print the script's totals here so I can compare them with each platform.
Flag any metric that moved more than [30]% from last week.

The pause in step 1 is deliberate. It catches a wrong or half-downloaded file before any math happens:

A good result: a report that opens in any browser, plus totals in the chat you can compare with each platform quickly.

Check: the totals match each platform's dashboard for the same dates. Anthropic's Claude Code best practices make the same point: give Claude a way to check its work and ask for the evidence.

Workflow 6: SEO tasks with Claude Code

Two jobs that take hours by hand. First, keyword clustering from an export:

@seo/keywords.csv is an export from my SEO tool with keyword, monthly volume, difficulty and the URL that ranks now.
Group the keywords into topic clusters by search intent. For each cluster, name one page that should rank for it: an existing URL from the file, or "new page" with a suggested title.
Flag keywords where two of our URLs compete for the same intent.
Keep volume and difficulty exactly as exported. Save seo/clusters.csv and a short seo/clusters.md summary.

Second, titles and meta descriptions in bulk:

For each URL in @seo/pages.csv, read the live page and write a new title (under [60] characters) and meta description (under [155] characters) around its main keyword.
Count the characters with a script and add the counts as columns.
Don't promise anything the page doesn't say. Save seo/meta-rewrites.csv with old and new versions side by side.

A good result: a clusters file that still contains every keyword from the export, and a rewrite file you can review row by row before changing your site.

Check: compare row counts with the original export, and read five rewrites against their live pages.

Workflow 7: UTM links and a tracking sheet

UTM parameters are the tags at the end of a link that tell your analytics where a visit came from. Small inconsistencies, like "Newsletter" in one link and "newsletter" in another, split your data.

Create tagged links for the [CAMPAIGN NAME] campaign pointing to [LANDING PAGE URL].
Channels: newsletter, Instagram bio, LinkedIn post and partner email.
Use lowercase and hyphens instead of spaces, and follow @brand/utm-rules.md if it exists.
Always set utm_source, utm_medium and utm_campaign. Add utm_content when a channel has more than one link.
Output utm/[CAMPAIGN].csv with the channel, the full link and each parameter in its own column, then confirm every link opens.

A good result: one row per link, identical campaign values everywhere, and a confirmation that each link loads.

Check: parameter names must be spelled exactly. Google's URL builder help page lists the parameters Google Analytics reads, including the three it says to always use.

Make it repeatable

Once a workflow has run twice, stop retyping it. Three features take you from a good prompt to a routine:

How to create a Claude skill walks through building one for campaign briefs, and ready-made marketing skills such as Competitive Brief and Copywriting are in the ai school skills directory. HubSpot, Canva and Google Drive all have official connectors in Anthropic's connector directory; start with what an MCP server is for the safety basics, and browse the ai school MCP directory for more servers.

If your goal is a steady flow of social posts rather than reports, see how to automate social media posts with AI.

Check before anything goes live

What to do next

Set up the marketing folder today, add voice.md and proof.md, and run Workflow 2 on your latest reviews or survey export. Its output feeds the brief, the landing page and the ads, so it pays off three times.

When one workflow works, turn it into a skill before you start the next. If you also want to produce the visuals and automate the follow-up, the AI marketing course pairs the Creators course with early access to Autopilot.