Claude Marketing Skills: 98 Free Skills, Ranked by Job (2026)

What are Claude marketing skills?

Anthropic calls the format Agent Skills. A skill is a folder containing a SKILL.md with YAML frontmatter (a name and a description that says when the skill applies), then the instructions, and optionally supporting files: checklists, benchmark tables, scripts. A marketing skill is that format pointed at one marketing job: audit a search terms report, diagnose a CPA spike, write the Monday digest, tear down a competitor’s landing page.

The difference from a good prompt shows up on the tenth use, not the first. A prompt lives in a paste buffer and drifts each time someone retypes it. A skill is versioned, shared, and produces the same table with the same columns every run. When an agency runs the same audit across 14 client accounts, the skill guarantees account 14 gets the same rigor as account 1. Claude reads only each skill’s name and description up front (progressive disclosure), so a library of 50 skills does not eat your context window.

Skills decide how Claude works. They do not grant data access: a skill on its own works with what you paste or upload. Connect Claude to your ad accounts through an MCP connector and the same skills run against live campaign data instead of Friday’s export. Nothing in either library executes changes in an ad account; that is a deliberate boundary, covered in the stack section.

Which Claude marketing skill for which job

If you install five skills and nothing else, install the first five rows. The rest map the most common marketing jobs to the one skill that handles them best, from the two libraries covered below. Sources: Ryze marketing-skills (48 files, paid ads, SEO, AI visibility) and coreyhaines31/marketingskills (50 skills, full marketing stack).

The pattern behind the first five: diagnostics before optimization, and one reporting skill from day one so improvements are visible.

How to install Claude marketing skills

The same SKILL.md folder installs in three places, and where you install it changes what it can do. Steps verified against Anthropic’s help center and the Claude Code docs, September 2026.

In claude.ai (web and mobile)

Skills are available on Free, Pro, Max, Team and Enterprise plans once code execution is on. Go to Settings › Capabilities and enable “Code execution and file creation”. Then open Customize › Skills, click +, Create skill, Upload a skill, and upload the skill folder as a .zip. Toggle it on. From then on there is nothing to invoke: ask “audit this search terms report for wasted spend” and, if a skill’s description matches, Claude uses it and says so. Uploaded skills are private to your account; on Team and Enterprise the owner enables Skills under Organization settings and can install them org-wide. The Ryze repo ships an all-marketing-skills.zip for exactly this step.

In Claude Code (the terminal)

The fastest route is one command: npx skills add coreyhaines31/marketingskills. Add --skill cro ads to install only the skills you name, or -a claude-code to target Claude Code specifically. The plugin route is /plugin marketplace add coreyhaines31/marketingskills then /plugin install marketing-skills. Manually, Claude Code reads two directories: ~/.claude/skills/ for personal skills available in every project, and .claude/skills/ inside a repo for team skills that ship with the code. Drop each skill in as a folder containing its SKILL.md; no restart needed. Invoke one directly with /skill-name or let Claude trigger it. This is where skills become pipelines: pull data through an MCP connector, run the analysis, write the CSV or HTML report, commit it, in one command. The Claude Code marketing workflows guide walks through five end to end.

In Claude Desktop and Cowork

The desktop app shares your claude.ai skills and adds the piece chat cannot: scheduling. In Cowork, a skill plus a cadence becomes a standing job: the Monday-morning account summary that runs before you are at your desk, the creative-fatigue check every Thursday. Pair a reporting skill with an ads MCP connector and the report arrives generated, not requested. See Claude Cowork for Meta Ads for the scheduled-run setup.

Other agents read the same format. The marketingskills repo installs into Cursor, Codex and Windsurf with the same npx skills add command; the paths differ (~/.cursor/skills/, ~/.codex/skills/). We keep separate lists for Cursor and OpenClaw.

The best Claude marketing skills in 2026, ranked by job

Ninety-eight skills is a library, not a shortlist. Below, every skill worth installing, grouped by the job people actually search for, with what it does, which repo it comes from, and how to install it. Rankings inside each group are editorial, weighted by how much money the skill’s output moves per minute of setup. Ryze publishes one of the two libraries, so read the paid-ads group with that in mind; the other library is not ours and we rank its skills where they land.

Claude skills for paid ads (Google Ads and Meta Ads)

This is where skills compound hardest, because the job is already made of repeatable numeric decisions. The Ryze files are written per platform; the marketingskills ads skill is a generalist that covers Google, Meta, LinkedIn and X in one file. Full platform lists with every prompt: Claude skills for Google Ads and Claude skills for Meta Ads.

Claude skills for marketing strategy

Strategy resists automation less than people assume, because the inputs to the thinking are automatable: channel P&Ls, cohort curves, last quarter’s tests. The marketingskills approach is to write positioning once into a shared .agents/product-marketing.md file, which every other skill reads, so a plan, a launch and an ad all respect the same audience and claims. Install product-marketing first; the rest inherit it.

Claude skills for performance marketing

The discipline these enforce is thinking in marginal CPA. A campaign with a great average CPA can still be the wrong place for the next dollar if its last increment of spend converted poorly. Skills that always compute the marginal view keep that mistake out of the Monday budget meeting. The wider PPC workflow around them is in how to use Claude AI for PPC management.

Claude skills for digital marketing: SEO and AI visibility

The workhorse SEO skills are audits, clustering and programmatic page builds. The newer half of the job is getting cited by ChatGPT, Perplexity and AI Overviews, which the Ryze repo splits into four skills of its own. The full workflow is in Claude Fable 5 for SEO and GEO; a starter set is in Claude skills for SEO.

Claude skills for social media marketing and content

The repeatable social job is not writing one clever post. It is turning every substantial asset into a channel-correct distribution kit without the voice drifting, which is exactly the fixed-output task skills exist for: the voice rules live in the file once, not in someone’s head.

Claude skills for market research and audience analysis

Research is where a skill saves the most re-explaining: the questions to ask a customer interview, the fields a competitor profile must have, the segments an audience breakdown must always include. Write them once and every study comes back in the same shape.

Claude skills for brand strategy and competitive research

Claude skills for marketing agencies

Agencies get more from skills than any in-house team, for a structural reason: they run the same workflows across many accounts. A skill written once is quality control across the whole book of business, and a new hire’s first audit reads like a senior’s because the senior’s process is in the file. Keep the skill folders in a shared repo so every account manager runs the identical version, and version them in git next to client configs.

Claude skills for CRO and landing pages

Paid traffic quality is half the CPA equation; the landing page is the other half. These give Claude a fixed teardown order (message match, headline, friction, mobile, proof) so the review is complete every time. Shopify stores get a dedicated set in 10 Claude skills for Shopify CRO.

Not covered here: the retention, growth-engineering, PR, events and lifecycle skills in the marketingskills repo. They exist and they are good; they are outside this page’s scope. Email flows get one clause: the emails skill and the Ryze Email Sequence Writer both exist.

Skills vs prompts vs MCP connectors vs execution

These four get blurred together in every “Claude for marketing” conversation, and they solve different problems. A prompt is what you say. A skill is a saved way of working. An MCP connector is a pipe to live data. An execution layer applies the change. They stack rather than compete.

That last column is the boundary the rest of this guide keeps returning to. Claude with skills and MCP is a formidable analyst: it reads the account, finds the waste, writes the plan. It does not log in at 3am and move the budget. Some teams close that gap by hand, which is fine at one account and painful at ten. The alternative is an execution layer like Ryze AI, which runs the same analyze-decide loop and then applies the changes, with an approval step before anything goes live. A sane adoption path: conversation first, then the five shortlist skills, then MCP when export round-trips are the bottleneck, then execution when reviewing recommendations becomes the slowest step.

Claude Code marketing skills: what changes when the same skill runs in the terminal

Claude Code marketing skills are the same SKILL.md folders as everywhere else; the runtime is what changes. In claude.ai a skill works on whatever you paste or upload. In Claude Code it has a working directory, so it can read the search terms export you dropped into exports/, run a Python pivot on it, write the result to reports/2026-09-08-wasted-spend.md and commit the file. The output stops being a chat message and becomes a file with history.

That is why the marketing skills for Claude Code that pay off first are the ones with a fixed output contract: Wasted Spend Finder, Weekly Account Summary, Client Report Narratives. Each run produces the same columns, so last week’s file diffs cleanly against this week’s. Keep them in .claude/skills/ inside a repo per client or per brand, with the exports folder next to them, and a new team member gets the whole process with git clone.

Two honest limits. It is a terminal, and most marketing teams have one person comfortable there, so let that person harden the skills and have everyone else run them in claude.ai or Cowork. And a skill that needs live account data still needs an MCP connector attached in Claude Code, the same as anywhere else. The terminal does not grant access on its own.

Marketing skills for Claude by team size: what to install solo, in-house and at an agency

The right set of marketing skills for Claude depends less on your channel mix than on how many people will run them. A founder doing their own ads needs five: Wasted Spend Finder, Weekly Account Summary, product-marketing (so every other skill knows the positioning), marketing-plan and cro.

An in-house team of three to eight adds the skills that keep people consistent with each other: the naming-convention and UTM generators, A/B Test Analyzer, Creative Fatigue Detection and the Content Repurposer. Put them in .claude/skills/ in a shared repo rather than in personal folders, so a change to the output format reaches everyone at once and nobody runs last quarter’s version.

An agency installs for onboarding and reporting first: the Google Ads and Meta Ads audit skills for the first week of any new account, Client Report Narratives for the monthly deck, Budget Scenario Planner for the renewal conversation, and ICP Research Assistant for pitches. Skills suit agencies better than anyone because of scale: one marketing skill for Claude runs identically across forty accounts.

Whatever the size, keep each person under ten active skills and write the trigger descriptions in the words your team uses, not the words the repo author used.

Claude skills for marketing: how the two libraries are organized, so you can find a skill without reading all 98

Every Claude skill for marketing in this guide comes from one of two repos, and they are organized differently. coreyhaines31/marketingskills (50 skills, MIT licence) is arranged by discipline with lowercase, hyphenated names: product-marketing, marketing-plan, copywriting, seo, social, cro, ab-testing, analytics, pricing, launch. The Ryze marketing-skills repo (48 skills) is arranged by platform and job with Title Case names: Google Ads, Meta Ads, SEO and AI visibility, reporting. If the skill you want is about a channel and money, look in the Ryze repo; if it is about positioning, content or the site, look in marketingskills.

The two overlap in three places: SEO, analytics and ad creative. Install one skill per job, not both, because overlapping descriptions are the most common reason the wrong skill triggers. Read the description line in each SKILL.md frontmatter before installing; that line, not the folder name, decides when Claude loads it.

One naming point. “AI marketing skills” in a job description usually means the human ability to use AI tools in marketing, and courses sell it. Claude skills for marketing run the other direction: files that give the model your process. Being good at the first is what makes the second worth writing.

A weekly operating cadence with Claude marketing skills

Skills earn their value on a schedule. The failure mode is installing fifty and running them whenever something already feels wrong, which is reactive firefighting with better tooling. Give each skill a slot so the diagnostics run before anything hurts.

The cadence is also the honest test of whether skills are working for you. If a slot keeps getting skipped, either the output is not driving decisions (rewrite its output contract) or the slot is wrong (change the slot, not the habit).

How to write your own marketing skill (SKILL.md anatomy)

The public libraries cover the common cases; your best skill will be the workflow only your team runs. Writing one is not a development task. A SKILL.md has four working parts:

• Frontmatter — YAML with a name and a description. The description is the trigger. Write it as “does X. Use when the user asks for Y or Z”; it is the highest-leverage line in the file.
• Role and steps — who Claude is and the steps in order, including the checks a careless analyst would skip.
• Input contract — exactly what data the skill expects and what to do when something is missing: ask, do not guess.
• Output contract — the exact format of the result: columns, sort order, severity scale. This is what makes run 40 comparable with run 1.

Two habits separate skills that work from skills that disappoint. Tune the description until the skill triggers on the phrasing your team actually uses; vague descriptions are the number-one reason a skill never fires. And keep SKILL.md short: move benchmark tables, brand voice docs and example outputs into separate files inside the folder, which Claude pulls in when needed. Build order for your first one: pick the task you did three times last month, write it from your own notes, run it against a task you already did by hand, and diff the results.

Troubleshooting: when Claude skills don’t work

Nearly every skill failure is one of five patterns, and all five have boring fixes.

Related guides