How to Install and Use the marketing-skills Bundle with Claude Code (2026)

# How to Install and Use the marketing-skills Bundle with Claude Code (2026)

Anthropic shipped marketing-skills in late 2025 as a bundle of 36 sub-skills covering the work that used to need a small marketing agency: CRO copy, landing-page audits, paid-ad headline writing, email sequences, SEO briefs, brand voice work, and competitor analysis. One install, all 36.

This is the install guide and first-run walkthrough I wish I'd had. By the end of this you'll have run an actual landing-page audit and an actual ad-copy session against your own product.

For the broader picture of what skills do for solo operators, see [our list of AI agent skills for freelancers](/blog/best-ai-agent-skills-for-freelancers-2026).

What you need

• Claude Code CLI installed and signed in (Pro or Max plan)
• Terminal access (any OS: macOS, Linux, WSL on Windows, or native Windows PowerShell)
• 5 minutes for the install
• A real landing page or product to test against

If you don't have Claude Code yet, install it first:

``bash npm install -g @anthropic-ai/claude-code claude login `

That's the prerequisite. The skill install assumes you can run claude from your shell.

Step 1: Install the skill

`bash claude skill add anthropic/marketing-skills `

This downloads the bundle from the Anthropic skills registry, stores it in your local skills directory (~/.claude/skills/ on macOS/Linux, %USERPROFILE%\.claude\skills\ on Windows), and registers it for autoloading.

You should see something like:

` Installed marketing-skills v1.4.2 (Anthropic) 36 sub-skills available: cro-copy, landing-page-audit, ad-headlines, ... Skill will autoload when relevant prompts are detected. `

Step 2: Verify the install

Run:

`bash claude skill list `

You should see marketing-skills in the output. If you want to inspect what was installed:

`bash claude skill info marketing-skills `

This dumps the skill's manifest: the list of sub-skills, the descriptions, and the trigger phrases that load them.

Step 3: How sub-skills load

The marketing-skills bundle uses a router pattern. The top-level SKILL.md is small and tells Claude which sub-skill to load based on what you ask. You don't pick a sub-skill manually. You ask a marketing question and the right sub-skill comes in.

For example:

• "Audit my landing page at example.com" → loads landing-page-audit
• "Write three Facebook ad headlines for our pricing page" → loads ad-headlines
• "Draft a 4-email welcome sequence for new signups" → loads email-sequences
• "What's our brand voice?" + brand doc attached → loads brand-voice

This pattern matters because it keeps your context window small. You don't pay tokens for the 35 sub-skills you didn't need.

Step 4: Run your first audit

Open Claude Code in a project directory:

`bash cd ~/my-project claude `

Then prompt:

` Audit the landing page at https://your-real-product.com. Identify the top 5 conversion blockers and rank by impact. Cite specific copy or design choices in your reasoning. `

Claude should respond with the landing-page-audit sub-skill loaded. Expected output: a numbered list of 5-7 issues, each with a quote from the page, an impact estimate, and a specific fix. Should take 30-60 seconds to generate.

If it gives you a vague "consider your headline" answer, your prompt was too vague. Try again with the URL spelled out and the request for specific quotes.

What a good audit output looks like

I ran this against my own freelance landing page and got back issues like:

• "Hero subhead reads 'Strategic AI consulting' — vague benefit, no specific outcome named. Fix: 'Cut your team's manual reporting from 6 hrs/week to 30 minutes.'"
• "CTA button says 'Get in touch' — generic. Test 'Book a 20-min audit call.'"
• "Pricing section absent on the page. Even a 'starting at' anchor reduces sales-call no-shows."

That's the level of specificity you want. Vague output means vague prompt. Re-prompt with concrete examples.

Step 5: Run an ad-copy session

` I'm running Facebook ads for our SaaS product. Product: [your one-line description] Target audience: [your audience] Pain point: [specific pain] Write 5 ad headlines + 5 ad body variants. Each headline should be under 40 characters. Each body should be under 125 characters. Include one variant using social proof and one using urgency. `

The ad-headlines sub-skill loads. You'll get back 5x5 = 25 combinations, with character counts shown so you don't have to verify them manually.

Two quick tips:

• Always specify the channel (Facebook, Google, LinkedIn) because the constraints differ
• Always give a real pain point, not a generic one. "Slow to ship features" beats "wants productivity"

Step 6: Tell Claude to remember your brand

If you have a brand voice doc, put it somewhere Claude can read it:

`bash cd ~/my-project mkdir -p .claude/context cp ~/Documents/brand-voice.md .claude/context/ `

Then in your Claude Code session:

` Read .claude/context/brand-voice.md before answering any marketing prompt this session. Apply that voice to every output. `

The brand-voice sub-skill picks this up. From here on, ad copy, email sequences, and audits will reflect your brand's tone, not generic Anthropic-default voice.

Common pitfalls

Pitfall 1: expecting it to write the whole campaign at once. The bundle is best at one task at a time. "Audit my page, write ads, draft emails, and design my landing page" all at once will return shallow output on each. Run one task per prompt.

Pitfall 2: using it without a brand doc. Default output reads like every other AI marketing tool. Feed it your voice doc first.

Pitfall 3: not iterating. Marketing copy is iterative. The first ad batch is rarely the winner. Treat the first output as a starting point, push back ("more concrete," "less corporate," "lean into the urgency angle"), and run again. The skill responds well to specific feedback.

Pitfall 4: ignoring the SEO sub-skills. The bundle includes SEO briefs and on-page audits, but they overlap with the dedicated [claude-seo skill](/blog/claude-seo-skill-review-worth-installing-2026). For serious SEO work I run claude-seo. For occasional SEO touches inside a marketing project, marketing-skills' SEO sub-skills are enough.

What's inside the bundle (sub-skill list)

A condensed view of the 36 sub-skills:

CRO and conversion

• landing-page-audit
• pricing-page-optimization
• cta-testing
• form-friction-audit
• checkout-funnel-review

Paid ads

• ad-headlines (Facebook, Google, LinkedIn variants)
• ad-body-copy
• creative-brief
• ad-image-prompt-generator

Email

• email-sequences (welcome, onboarding, re-engagement)
• broadcast-email
• subject-line-testing
• transactional-email-tone

SEO

• keyword-research-light
• on-page-audit-light
• meta-tag-optimization

Brand and voice

• brand-voice-extraction (reads a doc, codifies the voice)
• competitor-voice-comparison
• tagline-generator

Strategy and analysis

• competitor-positioning-analysis
• icp-development (ideal customer profile)
• value-prop-canvas
• positioning-statement
• messaging-house

The remaining sub-skills cover content strategy, partner/affiliate copy, PR pitches, sales-page structures, and a handful of vertical-specific helpers (B2B SaaS, e-commerce, agencies).

Updating the skill

Anthropic ships updates roughly monthly. To pull the latest:

`bash claude skill update marketing-skills `

Run this every 4-6 weeks. Updates have added new sub-skills (the messaging-house sub-skill was added in March 2026) and tuned existing ones.

Performance tip: keep your context window clean

The bundle is 36 sub-skills, but Claude only loads the relevant one per prompt. That said, if you run a 90-minute marketing session in one conversation, your context fills with prior outputs. Two practices keep things sharp:

1. Start a new Claude Code session for each task. Audit the landing page in one session, write ad copy in another. The audit context shouldn't leak into the ad copy. 2. Save outputs to files instead of re-asking. When Claude generates 25 ad variants, save them to ads-v1.md, then start fresh. Asking "remember those ads from earlier?" pulls 4,000 tokens of context for no reason.

I learned this the hard way after a 3-hour session that produced confused outputs because Claude was juggling a brand-voice doc, a competitor analysis, six rounds of ad copy, and an email sequence in the same context. Splitting the work across four sessions, each with one job, fixed it.

When to use marketing-skills vs other tools

• You want a comprehensive marketing co-pilot: marketing-skills wins
• You only do SEO content: claude-seo is more focused
• You only do landing pages and need design too: add frontend-design alongside marketing-skills
• You want to keep brand voice consistent across long deliverables: add brand-guidelines alongside marketing-skills

The skills are designed to compose. I run marketing-skills + brand-guidelines + claude-seo simultaneously on most client work. They don't fight each other.

FAQ

Is marketing-skills free? The skill itself is free. You pay for Claude Code (Pro $20/month or Max $200/month). There's no separate fee for the bundle.

Can I uninstall it if I don't like it? Run claude skill remove marketing-skills` and the bundle, including all 36 sub-skills, is gone.

Does it work with claude.ai or just Claude Code? Just Claude Code. Skills are a CLI feature. The web claude.ai doesn't load skills.

How do I know which sub-skill is loaded? Run any prompt, then ask "which sub-skill did you use for that response?" Claude will tell you.

Can I edit the sub-skills? Yes. They're files in your local skills directory. Editing them is supported, but updates from Anthropic will overwrite your changes unless you fork the skill. If you want a permanent custom version, copy the bundle, rename it, and install your local version.

What if my industry isn't covered? Most sub-skills are industry-agnostic. The vertical-specific ones (B2B SaaS, e-commerce) are starting points. You'd extend them with your own context. For very specialized verticals (healthcare regulated copy, finance compliance), you'd want to add a brand-guidelines doc with the regulatory constraints baked in.