30+ Best Claude Marketing Skills for Growth Teams (2026 Stack)

Claude marketing skills are reusable instruction files (a SKILL.md with YAML frontmatter) that teach Claude how to run specific growth jobs: CPA diagnostics, wasted-spend audits, creative fatigue detection, AEO content, and churn analysis. The open ecosystem now includes Corey Haines' marketing skills library at 44,100 GitHub stars, plus Metaflow's own 27-skill operator pack for paid search, paid social, SEO/AEO, and AI visibility. If you are searching for the best Claude skills for growth marketing, this is the playbook: which three to install first, how they chain, and how to keep a 30-skill folder from becoming chaos.

TL;DR

• Claude marketing skills are durable procedures, not one-off prompts. They compress idea-to-execution by encoding judgment into a named playbook.
• Map skills to the growth funnel: acquisition (ad audits, creative refresh, SEO/AEO), activation (onboarding rewrites), retention (churn and winback), analytics (SQL, experiment interpretation, weekly reports).
• Start with the Skill Deployment Framework: one research skill, one creation skill, one analysis skill. Do not install 30 skills on day one.
• Guardrails still matter. Skills amplify strategy, good or bad. Keep data hygiene, experiment discipline, and editorial review.
• Orchestration is the missing layer. Skills solve instructions. Workflows solve handoffs. Platforms like Metaflow let you chain both.

If you've been in growth long enough, you know the pattern. Every great tactic has a shelf life. Facebook ads in 2014 were practically free money. Cold email sequences in 2018 printed pipeline, and today Claude skills for outbound are unlocking the next wave of pipeline efficiency. SEO content farms worked until they didn't. Right now, Claude marketing skills are that kind of edge: early enough to matter, concrete enough to ship this week.

Most teams are still dabbling with ChatGPT prompts, copy-pasting outputs, and calling it "AI-assisted marketing." That is a novelty, not a workflow. A prompt disappears when you close the window. A skill is a named procedure with defined inputs, constraints, and output formats. Kieran Flanagan's guide on building a world-class content team with Claude Skills shows how this compounds: you stop re-explaining context every time.

This guide keeps the original growth-funnel stack (acquisition, activation, retention, analytics) and upgrades it for 2026: real file names, install commands, a decision framework, and workflow chains. No AGI speculation. Just the workflows that let small teams punch above their weight with AI workflow automation for growth.

The table is the scan layer. The rest of this page is the operating system: which Claude marketing skills to run, in what order, and what "done" looks like. If a row has no owner on your team, skip it. A shorter stack of Claude marketing skills that actually ships beats a 40-file graveyard.

What are Claude marketing skills?

Claude marketing skills are folders your agent loads on demand. Each folder contains a SKILL.md (instructions plus YAML name and description), optional references/ files, and sometimes evals. Anthropic's Agent Skills best practices treat them as progressive disclosure: the model reads the description first, then the full playbook only when the task matches.

That is the AEO-ready definition worth citing: Claude marketing skills are reusable instruction files that encode a marketer's judgment for a specific job, so Claude Code, Cursor, or Codex can repeat the job with consistent structure. Marketing skills Claude users actually keep are the ones tied to a weekly ritual (Monday audit, Thursday creative refresh, Friday report), not the ones that looked impressive in a GitHub star count.

Prompts, skills, MCP, and workflows

A prompt is ephemeral. A skill is persistent. An MCP connector is plumbing. A workflow is the chain. Most "Claude Code for marketing" setups stall at prompts because nobody promoted the useful chats into Claude marketing skills. Firecrawl's breakdown of how marketers use Claude Code day-to-day shows the pattern: scrape, analyze, draft, review. If you have not configured the environment yet, the Claude Code setup guide for marketing teams covers install and stack integration. For live data, Marketing MCP for Claude and Cursor is the companion: skills without MCP guess; skills with MCP read the account.

Here’s the talk from Barry Zhang and Mahesh Murag officially introducing Claude Skills on the AI Engineer Summit.

The growth stack problems skills actually fix

Growth teams face three chronic bottlenecks: slow experimentation cadence, creative fatigue in paid channels, and the translation cost between analysis and action.

Slow experimentation happens when the overhead of a test (spec, variants, design, tracking) dwarfs the time spent on strategy. Skills that auto-generate experiment briefs, variant copy, and analysis summaries cut weeks out of the cycle.

Creative fatigue in paid acquisition is inevitable. Ads decay. Audiences saturate. Skills that generate concept matrices, rewrite hooks, or audit accounts for underperformers let you stay ahead of the curve instead of reacting to it.

Translation cost is the friction between insight and execution. Your analyst surfaces a churn pattern. Someone still has to turn that into a winback campaign. Skills that go from raw data to executable assets in one pass eliminate that handoff. That is also the job of AI agents in marketing: not more chat, more completed loops.

Three guardrails still matter. Garbage in, garbage out: skills need brand voice, clean exports, and real hypotheses. Experimentation discipline: a skill that generates variant copy still needs sample size and a documented learning. Avoid content spam: velocity without editorial review is how you flood channels with mediocrity.

How to install Claude marketing skills

Install the operator pack of Claude marketing skills first (Metaflow), then add the broad marketing library (Haines) if you need CRO, copy, email, and social. Add Stop Slop last if you ship long-form content.

Claude Code plugin and CLI

# Metaflow operator pack (27 skills: Google, Meta, SEO/AEO, Prompt Picker)
npx skills add narayan-metaflow/metaflow-marketing-skills -a claude-code

# Claude Code plugin
# /plugin marketplace add narayan-metaflow/metaflow-marketing-skills
# /plugin install metaflow-marketing-skills

# Complementary ecosystem pack (CRO, copywriting, emails, social)
npx skills add coreyhaines31/marketingskills -a claude-code

# Content quality companion
git clone https://github.com/hardikpandya/stop-slop.git ~/.claude/skills/stop-slop

If you run npx skills add from inside an agent session, pass -a claude-code explicitly. Otherwise the CLI may write to .agents/skills/, which Claude Code does not read. A failed install is the most common reason teams decide Claude skills for marketing "don't work."

Cursor, Codex, and MCP connectors

# Cursor: copy skill folders into the project skills directory
git clone https://github.com/narayan-metaflow/metaflow-marketing-skills.git
cp -r metaflow-marketing-skills/skills/* .cursor/skills/

# Optional: Haines pack into the same layout
npx skills add coreyhaines31/marketingskills -a cursor

Codex and other Agent Skills clients follow the agentskills.io folder spec. Clone the repo, keep skills/<name>/SKILL.md, and point the client at that tree. MCP is separate: Claude Desktop or Claude Code connectors for Google Ads, Search Console, or your warehouse. Skills tell the model how to think. MCP tells it what to read. You need both for a live audit.

Browse the same playbooks as product skills in the Metaflow skills directory. The GitHub repo is the source of truth for SKILL.md files you can fork.

The 30+ skill Claude stack for growth teams

Here is the distilled list. Items 01 to 15 are the original growth-funnel stack. Items 16 to 32 are the Metaflow operator files plus Stop Slop. Each line is what the skill does, why it pays off, and who owns it, so this map of Claude marketing skills is usable without opening every file. Usage, prompts, and links for the week-one set are in the next section.

1. Google Ads Account Audit: Reads a campaign export and returns ICE-ranked wasted-spend fixes, so a four-hour teardown becomes a 15-minute review. For paid media leads.
2. Meta Ads Account Audit: Flags audience overlap, fatigue, and bad budget split, so you stop pouring spend into dead ad sets. For paid media leads.
3. Meta Ads Creative Engine: Builds a concept-vs-variation matrix and a 7-day rotation, so "we're out of ideas" stops killing the account. For paid media or brand.
4. SEO Keyword Research and Intent Mapping: Clusters a seed keyword and competitor URLs by intent, so the calendar maps to jobs instead of a dump. For content or SEO leads.
5. Outline-to-draft + Stop Slop: Expands an approved brief, then strips AI cadence, so you ship ski-ramp structure that still sounds human. For content leads.
6. Onboarding flow rewrite: Teardowns emails and empty states against time-to-first-value, so new users complete one job instead of bouncing. For lifecycle or growth PMs.
7. In-product messaging: Drafts tooltip, email, and push for one real user state, so inbox and in-app stop contradicting each other. For product marketing or lifecycle.
8. Experiment design: Turns a Slack hypothesis into a spec (metric, sample, instrumentation), so tests ship with a decision rule. For growth PMs or analysts.
9. Churn analysis: Clusters a churn CSV into reasons, so you treat cohorts instead of reading 200 rows. For lifecycle or analysts.
10. Winback campaign generator: Writes different resurrection plays for 30-day vs six-month churn, so you stop sending one generic discount. For lifecycle.
11. NPS/survey clustering: Themes open-ended feedback into a ranked action list, so roadmap debates use frequency not anecdotes. For product or growth PMs.
12. Upsell messaging: Turns pricing tiers and usage patterns into expansion copy, so upgrades match how people already use the product. For lifecycle or sales ops.
13. SQL query generator: Turns a warehouse question into a query plus data-quality flags, so non-technical operators can pull numbers without a ticket. For analysts or growth PMs.
14. Experiment results interpreter: Summarizes control vs variant, significance, and next tests, so the learning is documented before the next launch. For analysts or growth PMs.
15. Weekly growth report: Turns latest numbers into what moved, what is concerning, and what to do Monday. For growth leads.
16. Prompt Picker: Builds a 20 to 40 prompt library across the buyer journey, so AI search tracking is a sample not one ChatGPT screenshot. For SEO or growth.

If a line has no owner on your team, skip that file. A shorter set of Claude marketing skills that actually ships beats collecting all 32. These thirty-plus Claude marketing skills cover the high-frequency jobs that slow growth teams. They are foundational, not exhaustive. For CRO, copywriting, cold email, and social, add Corey Haines' marketing skills library. A printed list is not a stack. Next: where each job sits in the funnel, then how to invoke the file on a real export.

How the Claude marketing skills fit in the growth funnel

The most useful way to organize Claude marketing skills is still by funnel stage: acquisition, activation, retention, expansion, and resurrection. Each stage has distinct jobs, metrics, and skill requirements. The 2026 upgrade is that acquisition now has real file names you can clone, while activation and retention still need custom skills (templates later in this page).

Acquisition skills get the right people into the top of the funnel: paid optimization, SEO/AEO production, landing page iteration. Metrics: CAC, impression share, CTR, cost per qualified lead.

Activation skills move signups toward first value: onboarding, email, in-product messaging. Metrics: activation rate, time to first value.

Retention skills keep users engaged: lifecycle, feature nudges, re-engagement. Metrics: DAU/MAU, cohort curves, churn.

Expansion skills grow revenue from existing users: upsell, pricing, expansion plays. Metrics: expansion MRR, ARPU, net revenue retention.

Resurrection skills win back churned users: winback campaigns, churn-reason clustering. Metrics: reactivation rate, resurrected LTV.

How to deploy these skills with examples

A skill on disk does nothing. Deployment is: pick the file, attach a real artifact (export, URL, CSV), paste a prompt that names the skill, then review the output before it ships. The cards below are that loop for paid, SEO, and lifecycle. File names live in the prompts. Product pages live in the Metaflow skills directory.

Paid acquisition Claude marketing skills

Paid channels are the fastest testing ground because feedback loops are short and the data is structured. Start with the three Claude marketing skills for Google Ads and Meta Ads below. File names live in the prompts. Product pages live in the Metaflow skills directory.

Google Ads Account Audit

A four-hour account review compressed into a scored punch list. Feed an export. Get wasted spend, structure, and tracking issues ranked by ICE, not a vibe check. Modeled time: 15 minutes of review instead of a half-day teardown.

Usage

• Run the 70-checkpoint audit on a campaign export, search terms, Quality Score, and auction insights
• Rank fixes by impact, confidence, and effort so the media lead knows what to do Monday
• Catch tracking and bidding problems before you add more budget

Prompt

Use google-ads-account-audit on this export [attach].
Score: tracking, structure, keywords, ads, bidding, audiences.
Output ICE-ranked fixes. Do not declare winners under 50 conversions.

Google Ads Account Audit · SKILL.md

Meta Ads Account Audit

Same job as the Google audit, different failure modes: audience overlap, frequency, and CAPI gaps. Rising CPMs with a flat CTR is fatigue, not "the algorithm." Pair with Meta Ads Tracking Setup if Pixel or CAPI is the real leak.

Usage

• Flag overlap and budget piled into fatigued ad sets
• Separate tracking problems from creative problems
• Produce a client-ready health snapshot after a messy handoff

Prompt

Use meta-ads-account-audit on this account export [attach].
Check: audience overlap, frequency, CPM vs CTR, budget split, CAPI/Pixel.
Output: what to pause, what to keep, what to investigate in tracking.

Meta Ads Account Audit · SKILL.md

Meta Ads Creative Engine

This is the "we're out of ideas" skill. Describe the offer, audience, and current concepts. You get a concept-vs-variation matrix, fatigue flags (frequency above 4), and 20 to 30 hooks to cherry-pick. Concept tests before color tests.

Usage

• Refresh a fatigued ad set without guessing which variable to change
• Build a 7-day rotation: pause 1-2, add 2-3, keep a winner in rest
• Produce hook / visual / copy modules a designer can actually use

Prompt

Use meta-ads-creative-engine.
Offer: [x]. Audience: [x]. Current concepts: [paste]. Frequency: [n]. CTR/CPA trend: [x].
Output: fatigue flags, concept vs variation matrix, next 7-day rotation.
Isolate one variable per test. Captions required on video.

Meta Ads Creative Engine · SKILL.md

Install the audit first. Add builder, optimizer, and reporter only after the weekly ritual exists. Those files are in the operator pack section below.

SEO, GEO, and content acquisition

SEO is high-leverage and slow. The right Claude skills for SEO cut keyword research to a brief a human can edit. Animalz's no-code Claude Code setup is a useful companion for the environment, not a substitute for the brief skill.

SEO Keyword Research and Intent Mapping

Stops "we should rank for this" from becoming 40 unrelated drafts. Cluster a seed keyword and competitor URLs by intent so the content calendar maps to jobs, not a keyword dump.

Usage

• Turn one seed plus 5 competitor URLs into intent clusters
• Separate informational, commercial, and transactional work
• Hand the cluster to the brief skill instead of writing from a spreadsheet row

Prompt

Use seo-keyword-research-intent-mapping.
Seed: [keyword]. Competitors: [URLs].
Cluster by intent. Flag cannibalization. Output: primary + 5 supporting terms per cluster, and which page type to build.

SEO Keyword Research and Intent Mapping · SKILL.md

SEO and AEO Content Strategy

Turns the cluster into a ski-ramp brief: answer first, question-based H2s, entities, information-gain angle. This is the intake for the content factory. Human editing still required.

Usage

• Write a publishable brief from a target keyword
• Front-load the claim so AI Overviews and humans get the answer in paragraph one
• Specify internal links and primary-source citations before anyone drafts

Prompt

Use seo-aeo-content-strategy for [keyword].
Intent: [informational/commercial]. Competitors: [URLs].
Output a ski-ramp brief: first-paragraph claim, question H2s, entities, information-gain angle, 3 internal links, 3 primary sources.

SEO and AEO Content Strategy · SKILL.md

Outline-to-draft and Stop Slop

Outline-to-draft is still custom for most teams. Stop Slop is the quality gate after it. Ski-ramp structure without Stop Slop ships as formulaic cadence. That is how pages rank for a week and get ignored.

Usage

• Expand an approved brief into a draft, then strip AI tells
• Catch banned phrases, parallel section rhythms, and empty transitions
• Keep the ski-ramp H2s while making the sentences sound like an operator wrote them

Prompt

Draft from this brief [attach]. Then run stop-slop.
Keep question-based H2s and the first-paragraph claim.
Kill stock phrases, identical section lengths, and "in today's landscape" openers.
Match this voice sample: [paste 2 paragraphs].

Stop Slop · brief starter is Template 2 below

Prompt Picker

AI search tracking without a prompt library is one ChatGPT screenshot. Prompt Picker builds 20 to 40 prompts across awareness, consideration, and brand evaluation, branded vs non-branded, so you measure a journey instead of a vibe.

Usage

• Build the question set buyers actually ask ChatGPT, Gemini, and Perplexity
• Split branded vs non-branded into separate tracking projects
• Hand the library to AI Search Visibility for multi-run measurement

Prompt

Use prompt-picker for [product] vs [competitors].
ICP: [x]. Journey stages: awareness, consideration, brand evaluation.
Output 30 prompts, labeled branded/non-branded, with persona variants.

Prompt Picker · SKILL.md

Programmatic SEO still fails when the pages are thin. Use it only with differentiated data. On-page, technical, BOFU, PR, local, and SEO Max files are in the operator pack below. Run Prompt Picker, then AI Search Visibility: 60 to 100 runs per platform, not one answer.

Activation, retention, and analytics

Activation is where SaaS leaves money on the table. Users sign up, poke around, and leave before value. These Claude marketing skills are not in the paid/SEO repo yet. Fork them as custom skills. Load Template 1 (brand voice) first or the copy will sound like every other product.

Onboarding flow rewrite

Most onboarding copy was written at launch and never matched the activation metric. This skill teardowns emails, tooltips, and empty states against "time to first value," then rewrites toward one completed job.

Usage

• Audit the signup-to-value path for friction, unclear CTAs, and mismatched promises
• Rewrite the first three emails plus the empty state for the same job
• Align in-app and inbox so they do not contradict each other

Prompt

Activation metric: [users who complete X in 24h].
Here are current emails, tooltips, and screens [attach].
Flag friction. Rewrite the first 3 messages toward that metric. One CTA. No new claims.

Custom. Start from Template 1.

In-product messaging

"Generates contextual nudges" is not a skill. A skill takes a real user state and drafts the email plus the in-app pair together, against brand voice, so the tooltip and the inbox say the same thing.

Usage

• Draft tooltip + email + push for one incomplete setup state
• Rewrite a feature announcement for users who skipped the tour
• Build a 3-message activation sequence from current screens, not from a blank prompt

Prompt

User: signed up 3 days ago, logged in twice, has not completed [setup step].
Activation metric: [x]. Brand voice: [attach Template 1].
Draft: 1 tooltip, 1 email, 1 push. Same CTA. No new claims. No "just checking in."

Custom. Owner: lifecycle or product marketing.

Experiment design

This is where AI agents for marketing earn their keep: a hypothesis becomes a spec instead of a Slack thread. Null, alternative, metric, sample size, instrumentation.

Usage

• Turn "we should test the headline" into a documented test
• Force a primary metric before anyone opens the A/B tool
• Hand the spec to the results interpreter after the run

Prompt

Hypothesis: [x] will lift [metric] for [audience] because [reason].
Write the spec: null/alternative, primary + guardrail metrics, sample size, instrumentation, ship criteria.
Do not invent a sample size. Flag what data is missing.

Custom. Owner: growth PM or analyst.

Churn analysis and winback

One generic winback to every cohort is how you train people to ignore you. Export recently churned users. Cluster why they left. Write different plays for under-30-day vs six-month cohorts.

Usage

• Cluster churn reasons from a CSV instead of reading 200 rows
• Draft onboarding-help copy for early churn, "what changed?" plus a bounded offer for late churn
• Feed themes into NPS clustering if the reasons are messy

Prompt

Here is a churn export [attach]. Cluster reasons.
Split: under 30 days vs 6+ months. For each cluster: likely cause, message, offer bounds, success metric.
No blanket discount.

Custom. Pair with Template 1.

The rest of activation and analytics (NPS clustering, upsell, SQL, experiment results, weekly report) follows the same shape: real artifact in, decision-ready artifact out. Those are custom until you fork them. Paid weeks of the weekly report should call Paid Media Reporter. Organic weeks should call SEO Reporting. That AI-powered GTM workflow is the Monday ritual, not a dashboard screenshot.

Launch workflow, kept from the original: audit competitor ads (Google or Meta audit) → creative matrix (Meta Ads Creative Engine) → landing variants → experiment spec. Growth PM reviews and ships. These AI-powered marketing automation loops are the point of Claude marketing skills: a named chain, not four unrelated chats.

The operator files below are the rest of the pack: invoke them after the audit or brief, once the weekly ritual exists.

Paid operator pack

Use these after the audit. Ad Spend Allocator answers "which channel deserves the next dollar" using MER, marginal ROAS, and incrementality, not last-click vanity. Prompt: Use ad-spend-allocator on [platform exports]. Compare MER vs platform ROAS. Recommend a 10-15% shift. Do not move budget if tracking is broken. Campaign Analyzer is the cross-channel diagnosis when Google looks efficient and Meta looks expensive, or the reverse. Paid Media Reporter turns that into a stakeholder narrative the CMO can read.

Usage

• Optimizer files (Google Ads Optimizer, Meta Ads Optimizer): in-platform next actions after the audit. Prompt: Use google-ads-optimizer (or meta-ads-optimizer) on the ICE list. Output this week's changes only.
• Builders (Google Ads Campaign Builder, Meta Ads Campaign Builder): structure, ads, and launch checklists so a new campaign is not a blank UI.
• Query and automation (Google Ads Keyword Engine, Google Ads Scripts): n-grams, negatives, and scripts. Prompt: Use google-ads-keyword-engine on this search terms report. Output negatives + n-grams, no new campaigns.
• Hygiene (Meta Ads Tracking Setup, Google Merchant Center): CAPI/Pixel and feed issues. Run tracking before allocator.
• Playbooks (Paid Search Ads Playbook, Meta Ads Playbook): the longer operating manual when one SKILL.md is not enough.

SEO and AI search operator pack

AI Search Visibility is the measurement half of Prompt Picker. One ChatGPT answer is not a sample. Prompt: Use ai-search-visibility-aeo-geo-llmo. Run the prompt-picker library 60-100 times per platform. Output citation rate, crawler access, and which pages to fix. On-Page SEO and AEO makes a URL extractable. Technical SEO and AI Crawler Audit checks robots, JS, and llms.txt so the model can see the page at all.

Usage

• BOFU SEO and AEO Strategy: comparison and alternatives pages. Prompt: Use bofu-seo-aeo-strategy for [us] vs [competitor]. Output page outline, proof points we can cite, claims we must not invent.
• Link Building and Digital PR: distribution for information-gain assets, not guest-post spam.
• Local SEO and Google Business Profile: local pack and GBP when the funnel is geo-bound.
• SEO Reporting and Measurement: organic plus AI-visibility KPIs for the Monday review.
• SEO Max KT: long-form methodology when one SKILL.md cannot hold the whole system.

These thirty-plus Claude marketing skills cover the high-frequency jobs that slow growth teams. They are foundational, not exhaustive. For CRO, copywriting, cold email, and social, add Corey Haines' marketing skills library. Deploy incrementally. A printed list of Claude marketing skills is not a stack. A stack is three skills, an owner, and a weekly trigger.

The most important and valuable full-stack Claude marketing skills

Which Claude marketing skills are actually worth installing first? Not the 32-file catalog. The most valuable full-stack set is three jobs that cover the loop: research, creation, and analysis.

Research finds what to work on. Creation produces the artifact. Analysis tells you whether it worked. Skip any one and you get either busywork or unmeasured output. We call this the Skill Deployment Framework: after you know what the files do, pick the three that make a full stack for your role.

The 3-skill minimum

Research without creation is a slide. Creation without analysis is spam. Analysis without research is rearranging last month's numbers. The 3-skill minimum is how Claude marketing skills stay attached to a job instead of a folder.

Give each person one week with those three. At the end of the week, run a retro: what felt clunky, what still takes too long, what output you actually shipped. That retro is how Claude marketing skills become operating rhythm instead of a bookmark. The same three-skill rule is how you keep this Claude marketing skills stack from rotting in a folder.

Claude marketing skills by team and channel

The implication is boring on purpose: install fewer Claude marketing skills, assign an owner, and attach them to a real export. Modeled operator time, not a survey: a 70-checkpoint Google Ads audit that used to take four hours becomes a 15 to 20 minute review of ICE-prioritized output. Anthropic's own growth team cut ad creation time from 30 minutes to 30 seconds with Claude Code workflows for paid media. That is the shape of leverage, not a promise that every account will 20x overnight.

If you are starting from zero, walk through how to create Claude skills before you scale the folder across the team.

Here’s how Gracie Leung is using Claude Skills for marketing. Gracie is a digital growth consultant and educator. Follow her newsletter and community if you want more, or watch the video below.

Workflow blueprints: chaining skills into end-to-end processes

A skill is a step. A blueprint is the job. Output of A is input of B. That dependency chain is what listicles skip. Claude skills marketing pages that only name files leave you with a directory. These three chains are the jobs growth operators actually run.

Blueprint 1: SEO content factory. SEO Keyword Research & Intent Mapping → SEO & AEO Content Strategy (brief) → outline-to-draft → Stop Slop → On-Page SEO & AEO Optimization → SEO Reporting & Measurement. Failure mode: skipping Stop Slop and shipping ski-ramp structure with slop cadence. Success: a brief with an information-gain angle and a draft a human can edit in one pass.

Blueprint 2: Paid ads diagnostics. Google Ads Account Audit or Meta Ads Account Audit → Meta Ads Creative Engine or Google Ads Optimizer → Ad Spend Allocator → Paid Media Reporter. Failure mode: reallocating budget before fixing tracking. Success: ICE-ranked fixes, then a 10 to 15 percent budget shift, then a report the CMO can read.

Blueprint 3: Churn response. CSV export → churn analysis → winback campaign generator → NPS follow-up clustering. Failure mode: one generic winback to every cohort. Success: under-30-day users get onboarding help; six-month users get a "what changed?" survey plus a bounded offer.

Document the sequence in a wiki: "When we launch a paid campaign, we run audit → creative → experiment spec. Each output feeds the next." Better: encode it as an AI workflow with human review gates so the chain is infrastructure, not a Slack habit.

Quick-start SKILL.md templates

The full playbooks live in the Metaflow repo. These three starters match that YAML shape so you can paste them into .claude/skills/<name>/SKILL.md today. Fork, then replace the stubs with your brand facts. Do not treat the stubs as the 70-checkpoint ads audit or the full ski-ramp brief. They are on-ramps: enough structure that Claude marketing skills you write yourself do not start as a blank page. Brand voice loads first. The brief skill is the content factory intake. The creative reviewer is the paid-media gate before you spend another week on fatigued ads.

Template 1: Brand voice guide (load this before copy or lifecycle skills).

name: brand-voice-guide
description: "Apply when writing or editing marketing copy, emails, ads, or in-product messaging. Encodes voice, banned phrases, proof points, and CTA rules so outputs stay on-brand."

# Brand voice

Voice:

- We sound like: [peer operator, not a vendor]

- We never sound like: [hype, guru, corporate filler]

Proof we can cite:

- [metric + source]

- [customer outcome, anonymized if needed]

Banned phrases:

- game-changer, seamless, leverage, delve

CTA rules:

- Primary: [Get a demo / Start workspace]

- Do not invent discounts or SLAs

Template 2: SEO content brief (trimmed from seo-aeo-content-strategy).

name: seo-content-brief
description: "Use when turning a target keyword into a publishable brief. Requires search intent, ski-ramp H2s, entities, and an information-gain angle."

# SEO content brief

1. Primary keyword + 5 supporting terms
2. Intent: informational / commercial / transactional
3. First-paragraph claim (front-load the answer)
4. Question-based H2s, one topic each
5. Entities to include (brands, tools, numbers, dates)
6. Information-gain angle (what no ranking page has)
7. Internal links: pillar + 2 related clusters
8. External citations: primary sources only

Template 3: Ad creative reviewer (trimmed from meta-ads-creative-engine).

name: ad-creative-reviewer
description: "Use when reviewing ad concepts or fatigued creative. Isolate one variable, flag frequency above 4, and output a test matrix rather than opinions."

# Ad creative reviewer

Inputs:

- Offer, audience, current concepts, frequency, CTR/CPA trend

Rules:

- Isolate one variable per test

- Concept tests before variation tests

- Do not declare a winner under 50 conversions per variant

- Captions required on video (most Feed/Reels plays are silent)

Output:

- Fatigue flags

- Concept vs variation matrix

- Next 7-day rotation (pause 1-2, add 2-3, keep a winner in rest)

For the 70-checkpoint Google Ads version, do not paste a stub. Clone skills/google-ads-account-audit/SKILL.md from the repo. That file is the actual IP.

Compatibility, models, and skill stack management

Read the matrix as a constraint, not a shopping list. Claude marketing skills are portable markdown. Live data and multi-step jobs are not. If the client cannot load a SKILL.md folder, you are back to prompts. Pick the runtime you already use, then add MCP only when the audit needs live account data.

Model selection is a cost and quality trade, not a brand loyalty test. Use a fast model for SQL drafts, n-gram negatives, and report scaffolding. Use a stronger model for account audits, creative strategy, and anything that will be customer-facing. If the skill encodes the rubric (ICE, ski ramp, fatigue thresholds), a cheaper model can still follow it. If the skill is a blank "be a senior marketer," you are paying for improvisation.

Skill stack management for 30-plus Claude marketing skills:

• Keep one repo. Fork Metaflow marketing skills for operator skills. Add a custom/ directory for onboarding, churn, and brand voice. Do not clone three overlapping CRO skills.
• Name folders after the job (google-ads-account-audit), not the person (sarahs-audit).
• Prefer fork over copy-paste when you will pull upstream. Prefer a custom skill when the playbook is your positioning, not a platform checklist.
• Deduplicate by description: if two SKILL.md files both claim "optimize Meta ads," delete one. Claude will otherwise load the wrong one.
• Version the files. A skill that still says "ECPC" after the March 2025 deprecation is a liability.

The 7-day plan to become an expert with Claude marketing skills

You do not become fluent by collecting 30 files. You become fluent by running the same three Claude marketing skills on live work for a week, then writing the chain so the next person can repeat it. This is a practice plan, not a reading list. Each day has one job and a definition of done.

Day 1: Install and pick three. Load the Metaflow pack, then choose one research, one creation, and one analysis skill from the Skill Deployment Framework. Do not install the rest yet.

• Run npx skills add narayan-metaflow/metaflow-marketing-skills -a claude-code (or copy skills/* into .cursor/skills/ for Cursor).
• Confirm the three folders landed in .claude/skills/ or .cursor/skills/. The wrong directory looks like a failed install.
• Write the three skill names, the owner, and the weekly trigger in a shared note.

Day 2: Research on a real artifact. Run only the research skill. The input must be an export, a URL, or a CSV you already use on Monday, not a toy prompt.

• Paid: drop a Google Ads or Meta export into Google Ads Account Audit (or the Meta audit).
• Organic: run SEO Keyword Research & Intent Mapping or Prompt Picker on a live URL and a real competitor set.
• Done means a ranked list of issues or opportunities, not a recap of the skill's own instructions.

Day 3: Creation from yesterday's output. Feed Day 2's packet into the creation skill. Do not start a new chat with a blank brief.

• Paid: Meta Ads Creative Engine or Google Ads Optimizer against the ICE-ranked fixes.
• Organic: SEO & AEO Content Strategy (or a winback draft for lifecycle).
• Done means an artifact a human can ship or reject in one pass: concepts, a brief, or a page outline.

Day 4: Analysis on the same job. Run the analysis skill on what you created, plus the original export. The question is "did this change the number," not "does the copy sound smart."

• Paid: Campaign Analyzer or Ad Spend Allocator.
• Organic: SEO Reporting & Measurement.
• Done means a decision: keep, kill, or reallocate, with the metric you will check next week.

Day 5: Write the workflow sentence. Turn the three runs into a chain someone else can follow. Output of A is input of B.

• Example: audit → creative → report or keyword map → ski-ramp brief → Stop Slop.
• Put the sentence in the same wiki as Day 1. If it needs a Slack explanation, it is not a workflow yet.
• Done means a new teammate could run the chain without asking you what "the packet" is.

Day 6: Add the quality gate. If you ship copy or long-form, install Stop Slop or your brand-voice skill and run it on Day 3's artifact. If you do not ship copy this week, skip this day and tighten Day 5 instead.

• Run Stop Slop after outline-to-draft, not instead of the brief skill.
• Cut stock phrases and keep the structure (ski ramp, ICE, fatigue rules).
• Done means the draft no longer sounds like a model reciting a template.

Day 7: Retro and scoreboard. Review the week against three numbers, then decide whether to add a fourth skill.

• Time saved: minutes for the same audit, brief, or report versus last month.
• Throughput: experiments, campaigns, or posts you actually shipped.
• Metric impact: CAC, activation, or organic. Not every skill shows lift in seven days. The weekly report is an efficiency play. Experiment design compounds over quarters.

If Day 7's retro is "we installed files," you practiced collecting, not operating. If the retro is "we ran three skills on live data and wrote the chain," you have a Claude marketing skills stack you can keep.

Growth teams running Claude marketing skills today are not just saving time. They are operating at a different experiment velocity: campaign launches in an afternoon, churn analyses between meetings, skill stacks assembled without a quarter-long roadmap. That is the new floor.

The uncomfortable part is coordination. A folder of 30 skills still leaves you copy-pasting outputs between tools, re-explaining context, and losing the packet of "what we decided last Tuesday." Skills encode judgment. They do not, by themselves, keep discovery and execution in one place. That is the gap an AI marketing workspace is for: agents and flows that can load the same playbooks, pass structured output downstream, and leave a trail you can debug.

Metaflow sits at that handoff. You can browse the same operator playbooks in the skills directory, then solidify the chain as an agent or a flow instead of a pile of markdown. If Claude marketing skills are how you stop repeating yourself, orchestration is how those skills become a growth system.

Frequently Asked Questions

What are the best Claude skills for marketing?

The best Claude marketing skills depend on the job, not a universal top 10. For paid growth, start with Google Ads Account Audit, Meta Ads Creative Engine, and Campaign Analyzer. For organic and AI search, start with Prompt Picker, SEO & AEO Content Strategy, and Stop Slop. For lifecycle, keep custom onboarding, churn, and winback skills. Metaflow's 27-skill pack plus the Haines ecosystem cover most of those jobs if you follow the Skill Deployment Framework instead of dumping the whole catalog.

How do I use Claude for marketing strategy?

Treat strategy as a research-plus-analysis loop, not a brainstorm chat. Load Prompt Picker or SEO & AEO Content Strategy for the question set, run Google Ads Account Audit or Ad Spend Allocator on real exports, then write decisions into a product-marketing context file the other skills can read. Claude for marketing strategy works when the skill has a rubric (ICE, MER, ski ramp). It fails when you ask it to "be the CMO" with no inputs. In Metaflow, that loop is an agent with the same skills attached, so the strategy artifact persists across sessions.

What are the top 10 skills for Claude?

If you want a short list: Google Ads Account Audit, Meta Ads Account Audit, Meta Ads Creative Engine, SEO Keyword Research & Intent Mapping, SEO & AEO Content Strategy, Prompt Picker, Ad Spend Allocator, onboarding rewrite, churn analysis, and weekly growth report. That set covers paid, organic, GEO, budget, activation, retention, and the Monday review. Expand from there with optimizer, reporter, and Stop Slop skills rather than collecting 40 files you will never invoke.

How do I install Claude marketing skills?

For Claude Code, run npx skills add narayan-metaflow/metaflow-marketing-skills -a claude-code or add the marketplace plugin. For Cursor, clone the repo and copy skills/* into .cursor/skills/. Then optionally add npx skills add coreyhaines31/marketingskills -a claude-code for CRO and copy, and clone Stop Slop for long-form. Install three, not thirty. Confirm the files landed in .claude/skills/ (Claude Code) vs .agents/skills/ (universal), because the wrong directory looks like a failed install.

What is the Stop Slop skill?

The Stop Slop skill is a SKILL.md pack that teaches the model to catch AI tells: stock phrases, formulaic section rhythms, and robotic transitions. It is not a marketing channel skill. It is a quality gate you run after outline-to-draft, especially on SEO pages. Pair it with SEO & AEO Content Strategy so you keep the ski-ramp structure and lose the slop. Teams that skip it ship faster and rank worse, because readers and models both punish generic cadence.

Do Claude skills work in Cursor and Codex?

Yes, if the client supports the Agent Skills folder spec (SKILL.md plus optional references). Claude marketing skills from the Metaflow repo and the Haines repo are markdown playbooks, not Claude-only binaries. Cursor reads .cursor/skills/. Codex and other agents typically read .agents/skills/. Claude Desktop is weaker for large libraries unless you attach files or use connectors. Live account data still needs MCP. Multi-step jobs still need a workflow layer, which is where Metaflow agents and flows sit on top of the same skill files.