The Claude Marketing Stack

Read time: 16 min

Why does everyone install the marketing skills and get generic output?

Because the playbooks are only half a system, and the half that makes them specific to your business is a file most people never write.

The free half comes from Corey Haines, who runs the Swipe Files newsletter and an agency called Conversion Factory. He published a collection of marketing playbooks written specifically for AI to read. It is free, over forty three thousand people have saved it, and individual playbooks inside it show sixty to a hundred and six thousand installs each. It is the collection that started the category.

The barrier it removes is real. Claude already knew marketing in the abstract. What it lacked was a structured way to run a conversion audit or a cold email sequence the way a practitioner would, in the right order, asking the right questions first.

The barrier it does not remove is that the playbooks only advise. That is why this document exists.

Before you start: what all of this actually is

Five terms. Once you have these, nothing else here is complicated.

Claude Code. Claude running on your computer rather than in a browser tab. The chat version can talk to you. This version can open files, run programs and connect to your other software. It needs a paid Claude plan and there is no free tier for it. If you have never installed it, that is your first step, and Anthropic's own setup page covers it in a few minutes.

The terminal. A plain window where you type instructions to your computer instead of clicking. It looks intimidating and it does not matter, because everything in this document is a single line you copy and paste. You never need to understand what it is doing.

A repo. Short for repository. A folder of files somebody published publicly for anyone to download. The marketing playbooks live in one. Downloading a repo is no different from downloading anything else, it just happens through a typed line instead of a button.

A skill. A written playbook the AI reads before it starts working. Not code. Plain English instructions describing how a good marketer approaches a specific job, in what order, asking which questions. Installing forty eight skills means handing Claude forty eight documents it can consult.

A connector. Software that gives Claude permission to log into your other tools, like your ad accounts or your online store. You will sometimes see it called an MCP, which stands for Model Context Protocol, and that name does not matter. What matters is that a connector is the thing that lets Claude do something rather than just describe it. You authorise it the way you authorise any app to reach your Google account, by logging in once and approving.

That is the whole vocabulary. Playbooks tell Claude how to think. A connector lets Claude act. You need both.

What is actually in the collection

Forty eight playbooks across eight categories: SEO and content, conversion rate optimisation, content and copy, paid advertising and measurement, growth and retention, sales and go to market, strategy, and monetisation.

The ones worth knowing by name.

cro handles conversion work on your pages and forms, meaning the question of why visitors are not becoming customers.

copywriting rewrites your text, and it cross-references cro and ab-testing so the three work as a set rather than contradicting each other.

seo-audit goes through your site looking for what is costing you search traffic.

ai-seo is newer and worth understanding. It covers getting your business cited when somebody asks ChatGPT or Claude a question, rather than ranking in Google. Different game, growing fast.

programmatic-seo covers building large numbers of pages from a template and a data source, the way comparison sites do.

pricing, churn-prevention, cold-email, prospecting and marketing-psychology are self explanatory, and each is a full playbook rather than a tip.

Two are unusual enough to call out. marketing-council convenes a simulated board of advisors that argues with itself before recommending anything, which produces noticeably better thinking than asking for a single opinion. And marketing-loops sets up recurring workflows that run on a schedule rather than when you ask.

Installing it, step by step

Fifteen minutes, and only two steps involve typing anything.

Step one. Install Claude Code if you have not already. Anthropic's setup page walks through it.

Step two. Make a folder on your computer for your marketing work. Call it whatever you like. This is where Claude will keep your context file and any output it produces.

Step three. Open the terminal and move into that folder. If you do not know how, ask Claude Code itself once it is running, because it can give you the exact line for your machine.

Step four. Type claude and press enter. You are now talking to Claude Code.

Step five. Paste this in:

npx skills add coreyhaines31/marketingskills

That downloads all forty eight playbooks and makes them available. If you already use Claude plugins, this route is cleaner:

/plugin marketplace add coreyhaines31/marketingskills
/plugin install marketing-skills

Step six. Confirm it worked by asking in plain English:

List the marketing skills you now have access to and tell me
what each one does in one line.

Around forty eight come back and you are done.

Two notes. This also works in Codex, Cursor and Windsurf, so you are not locked into Claude Code. And if you installed an earlier version, version 2.0 renamed seventeen playbooks and merged two older ones into a single cro. You will end up with duplicate folders and Claude will read two conflicting versions of the same playbook. The collection ships a cleanup script for exactly this, so ask Claude to run it.

The file that decides everything

This is the most useful thing in this document and almost no coverage mentions it.

One playbook in the collection, product-marketing, is the foundation. Every other playbook reads it first, before doing anything else. So you describe your product, your customer and your positioning once, and the other forty seven inherit that understanding automatically.

Skip it and every playbook runs on guesses. That is the real reason people install this collection, get bland output, and conclude it is overrated. The playbooks are fine. They were briefed on nothing.

It works exactly like hiring. A good marketer on their first day, given no context, produces confident generic work. The same marketer after a proper onboarding produces something useful. This file is the onboarding, and it takes twenty minutes once.

How to fill it in

Ask Claude:

Open the product-marketing skill and walk me through filling in
the context file. Ask me one question at a time and write my
answers into it as we go.

It will interview you. Answer properly rather than briefly, because everything downstream is capped by this.

What good answers look like

A worked example for an imaginary business, so you can see the level of specificity that matters.

What you sell. Weak: "project management software." Strong: "a project tracker built for architecture firms that bill by phase, so partners can see which projects are running over before the invoice goes out."

Who buys it. Weak: "small businesses." Strong: "the managing partner at a five to thirty person architecture practice, usually the same person who does the invoicing and hates it."

What they did before you. Weak: "spreadsheets." Strong: "a shared spreadsheet one person maintains and everyone else forgets to update, plus a time tracking tool nobody fills in until the end of the month."

What they compare you to. Name real competitors, including the option of doing nothing. "They look at Monday and Asana, decide both are too general, then usually do nothing for another six months."

The objection that kills deals. The single most valuable line in the file. "They think switching means re-entering three years of project history, and they will not start unless somebody tells them it does not."

Your price and why it is that number. "Two hundred a month per practice, priced against the four hours a month the managing partner currently spends reconciling timesheets."

How your best customers describe you. Their words, not your marketing. "They say it stops projects quietly going over budget."

Seven answers. Twenty minutes. It changes the output of all forty eight playbooks at once.

Why the plans stop at a plan

Run any playbook now and you get a document. A good one, structured the way a practitioner would structure it, and then nothing happens.

That is deliberate rather than broken. The playbooks are advisory. Claude reads them, thinks like a marketer, hands you the plan and stops, because it has no way to see your ad account, your analytics or your store.

To make it act you need a connector. That is the paid half, and there are three routes.

Choosing your connector

The two official ones are genuinely free. Meta's launched in April 2026, and Google's is free but read only, meaning Claude can analyse your Google Ads account and cannot change anything inside it. If you run one platform and mainly want analysis, start there and pay nothing.

Hyper is the paid multi-platform option at forty nine dollars a month after a seven day trial. What you get is everything through one login rather than four separate setups, approval gates on every action, and separate workspaces if you handle more than one business.

Most operators run more than one platform, which is why the paid route usually wins. Be clear about what you are buying though. The playbooks are the thinking and they are free. The connector is the hands, and it is the only piece that can actually spend your money.

Connecting it, step by step

Twenty minutes, mostly spent waiting for login screens.

Step one. Pick a route from section six. If you are unsure, start with the free official connector for whichever platform you spend the most on.

Step two. Add it. In Claude Code this is one line, and every connector publishes its own on its setup page. It starts with claude mcp add followed by the connector's address. Copy it from their page rather than typing it from memory.

Step three. Authorise it. A browser window opens asking you to log into your ad account and approve access. Same flow as connecting any app to your Google account. No passwords go into Claude and no keys get copied around.

Step four, and do not skip this. Turn approval gates on, meaning Claude has to ask before anything goes live. Every connector worth using supports it. Ask directly:

Confirm that approval is required before any action that
publishes, spends money, or changes a live campaign.

Step five. Test with something read only:

Show me my ad spend for the last thirty days broken down by
campaign. Do not change anything.

Real numbers come back and you are connected.

Your first three runs, in order

Do these in sequence over one afternoon rather than picking whichever sounds most exciting. The first two need no connector, so you find out whether the playbooks produce anything useful on your business before you pay for anything.

Run one, free, analysis only.

Use the seo-audit skill on [YOUR WEBSITE]. Read the
product-marketing context first. Give me the five issues costing
the most traffic, in priority order, with exactly what to change
on each one and roughly how long each would take.

Run two, still free.

Use the cro skill on [YOUR HIGHEST TRAFFIC PAGE], then use the
copywriting skill to rewrite the sections it flagged. Keep the
positioning from the product-marketing context and do not
invent claims I cannot support.

Run three, connector required. This is the one that justifies the subscription.

Pull the last thirty days from my ads account. Use the paid and
measurement skills to tell me which campaigns to cut and why,
then build the replacement campaign structure and leave it
paused for my review.

Never let a first connected run publish anything. Keep approval gates on until you have watched it work three or four times and you trust what it does.

The three artifacts you need

One, the completed product-marketing context file. Section four. Everything inherits it and nothing works properly without it.

Two, a connector with approval gates on. Whichever route you took, configured so nothing publishes without you.

Three, a written positioning brief, 300 to 600 words, and almost nobody has one. This is what you paste into the context file, and it deserves to be a separate artifact because most founders have never written their positioning down anywhere. Not the tagline. The actual thing: the specific customer, the specific problem, the alternative they would use if you did not exist, and the reason they choose you over it. Write it once and it feeds not just this setup but every piece of marketing you ever commission, from an agency to a freelancer to an AI. The quality of everything in this document is capped by this one file, which is why it deserves an afternoon rather than ten minutes.

The compounding layer

This setup improves with use in one specific way, and it is not the way people expect.

When the output misses, the instinct is to write a longer prompt. Resist that. Ask instead what the AI did not know about your business, then add that to the context file. The next run of all forty eight playbooks improves at once, rather than one prompt improving once.

After a few weeks that file becomes the most valuable thing in the setup. It is also portable, since it is plain text describing your business rather than anything technical, so it survives whatever replaces these tools next year.

The marketing-loops playbook is where this pays off properly. Once the context is good, you can set workflows that run on a schedule and produce work you never have to re-brief.

If something goes wrong

Honest limits

The playbooks are advisory, and this is the thing to internalise before installing anything. Forty eight playbooks do not connect you to Google Ads. They give the AI frameworks, and you still need the plumbing, which is either the free single-platform route, a free tier, or a subscription.

The recommendation to pair these two things specifically comes from Hyper's own comparison, and Hyper sells the paid half. The logic holds independently, since an advisory playbook genuinely cannot log into an ad account, but you should know who wrote the argument you are being persuaded by.

The free collection is the top of a funnel, and there is nothing wrong with that. Its front page points at the author's agency, his paid course, his community, and a paid autonomous marketing agent built on these same playbooks. Free playbooks as distribution for a paid product is the cleanest go-to-market in this category right now. Worth studying rather than resenting, and worth knowing that the free thing is the demo.

Agent skills as a category have a security problem. An audit of over two thousand eight hundred skills in early 2026 found 41.7 percent contained substantive vulnerabilities, with 30.6 percent rated high or critical. That is a category finding rather than a knock on this collection, and the practical lesson is to install from named people with visible reputations rather than from whatever is trending this week.

And the honest bottom line on the output. These playbooks encode how a good marketer approaches a job. They do not know your market, they cannot tell you what is genuinely wrong with your business, and they will produce a confident, well-structured conversion audit of a page whose real problem is that nobody wants the product. Judgment stays yours.

What this changes

Marketing has been unbundling for a decade, from agency to freelancer to contractor to tool. What changed this year is that the frameworks stopped being the expensive part and the execution stopped being the manual part, in the same twelve months.

What is left is the thing that was always actually scarce, which is knowing what to point it at. Forty eight playbooks will happily run a flawless campaign for a positioning you have never thought through. So the leverage moved to that six hundred word file, and the people who write it properly will get compounding returns from tools everyone else has already installed and given up on.