Chapter 12: Marketing Automation & Vibe Marketing

Turn OpenClaw into a marketing department. Use Claude Code + custom skills to automate SEO, content creation, social media management, and ad generation at agency-level quality for $0.

---

The Opportunity

Most businesses pay $5,000–$50,000/month for marketing agencies. Claude Code can replicate (and often exceed) their output with custom skills, minimal configuration, and zero per-task cost.

What You Can Automate

• SEO strategies & audits — Claude can analyze competitors, identify gaps, suggest keyword targets
• Social media content — Generate platform-specific posts, threads, video scripts for TikTok/Instagram
• Ad copy generation — Rapid A/B test variations for Facebook, Google, LinkedIn
• Community management — Reply to comments, respond to DMs, moderate conversations
• Email campaigns — Personalized sequences, A/B testing, analytics synthesis
• Content calendars — Plan monthly strategies, identify content gaps, suggest themes
• Performance analysis — Analyze metrics, spot trends, recommend optimizations

---

The Core Pattern: Skills + Prompts

The most effective marketing automation follows a simple pattern:

1. Build a skill for a specific task (e.g., "Facebook Ad Generator", "SEO Analyzer")
2. Wire it to a trigger (Telegram button, scheduled cron, Discord slash command)
3. Let it run — No manual intervention needed

Example: Build a Social Media Manager Skill

Goal: Your OpenClaw manages all your social media replies and DMs.

Steps:

1. Create a skill that:

   • Fetches your latest tweets/posts
   • Identifies high-engagement comments
   • Drafts thoughtful replies in your voice
   • Posts via Typefully (for review) or directly (if approved)

2. Schedule it to run every 4 hours via cron

3. Customize your SOUL.md so the agent knows your voice and values

Prompt template:

You are a social media manager for [name/brand]. Your job:
1. Check [platform] for recent posts with 10+ replies
2. Identify the highest-quality, most thoughtful replies
3. Draft 2-3 response options in this voice: [your voice guide]
4. For each reply, assign a priority (respond now / respond later / skip)

Format output as:
- Post: [link]
- Top reply: [quote]
- My response: [your response]
- Priority: [now/later/skip]

Constraints:
- Replies must be genuine, not marketing-y
- Never argue or get defensive
- Aim to deepen conversations, not sell

---

Use Cases & Implementation

1. SEO Automation (Zero Cost, Agency-Level)

What it does: Analyzes your competitors, identifies SEO opportunities, generates keyword-targeted content.

Skills you need:

• Content research (via web search + firecrawl)
• Keyword analysis (via Serpstat API if available, or manual analysis)
• Content generation (built-in)

Implementation prompt:

You are an SEO strategist. For the topic [topic], do:
1. Research top-ranking competitors (sites currently ranking #1-10)
2. Extract their keywords, backlink strategies, content structure
3. Identify gaps — keywords they're missing or underexploring
4. Suggest 5 content pieces we should create to own this niche
5. For each piece, provide: keyword target, search intent, outline, estimated traffic

Use firecrawl to scrape top pages for structure analysis.

Output: A weekly SEO briefing with actionable content ideas.

---

2. Content Creation (Social Media + Long-Form)

What it does: Generates platform-specific content (tweets, threads, TikTok scripts, LinkedIn posts, blog articles).

Skills you need:

• x-research (for trending topics)
• voice-matcher (your SOUL.md)
• platform-formatter (tweet ≠ LinkedIn ≠ TikTok)

Implementation prompt:

You are [name], a content creator in [niche]. Your voice: [voice description].

Generate 5 pieces of content for this week:
1. 1 Twitter thread (3-5 tweets)
2. 3 short-form tweets (for engagement)
3. 1 LinkedIn post (professional + personal angle)
4. 1 TikTok script (30-60 seconds)

For each:
- Research trending topics in [niche] from last 24h
- Generate content that's authentic to [name]'s perspective
- Include hooks, structure (intro/body/CTA), and talking points
- Tag relevant accounts/hashtags

Format:
## [Platform]

[content]

**Hook:** [why people should care]
**Talking points:** [main ideas]
**CTA:** [call to action]

---

3. Ads & Growth (Rapid Testing)

What it does: Generates 50+ ad variations in minutes, tests copy frameworks, identifies what works.

Skills you need:

• Copywriting templates (hooks, pain points, benefits)
• A/B test runner (if you have ad account API access)

Implementation prompt:

You are a performance marketer. Generate 20 Facebook ad variations for [product/service].

Use these frameworks:
- Problem-Agitate-Solve (PAS)
- Comparison (ours vs status quo)
- Curiosity + benefit
- Social proof + urgency
- Value prop + objection handling

For each, provide:
- Headline (max 27 chars)
- Primary text (max 125 chars)
- Description (max 30 chars)
- Estimated CTR (based on copy strength)

Organize by framework so we can A/B test frameworks too.

---

4. TikTok Slideshow Growth Engine

What it does: Automates viral TikTok slideshow creation and posting. Slideshows get disproportionate reach for the effort required and can be fully automated. (Credit: @wickedguro)

Setup:

1. VPN with dedicated US mobile IP (NordVPN, ExpressVPN — mobile IPs, not datacenter)
2. Keep VPN connected. Set phone to US locale. Remove GPS tracking.
3. Register TikTok in US locale. Engage with niche content first to train audience targeting.
4. Install Postiz for scheduling across TikTok, Instagram, and YouTube from one CLI

Automation prompt:

Build a TikTok growth skill. Every day:
1. Research trending topics in [niche] from the last 24h
2. Generate a 5-8 slide slideshow on the top topic
3. Use [your content style] for slide copy
4. Export slides as images to workspace/content/tiktok/
5. Schedule via Postiz CLI for [optimal time]
6. Log to workspace/content/published/tiktok-[date].md

Scale: once a format works, post the same slideshow across multiple accounts via Postiz.

Mobile-first: The OpenClaw Android app makes it easy to review and approve TikTok content on the go. Get a Telegram notification, preview the slides on your phone, approve with one tap.

---

5. Community & Support (Scaling Conversations)

What it does: Manages comments, DMs, support tickets at scale without sounding robotic.

Skills you need:

• Sentiment analysis (is this comment positive/negative/question?)
• Response prioritization (urgent vs nice-to-have)
• Multi-tone voice (sales vs support vs education)

Implementation prompt:

You manage community for [platform/brand]. Your role:
1. Review unread comments/DMs
2. Categorize by intent: question / feedback / complaint / praise / spam
3. For each category, draft an appropriate response
4. Prioritize urgent (> 100 replies, negative sentiment, creator asking question)

Tone guidelines:
- Questions: Helpful, thorough, educational
- Complaints: Empathetic, problem-solving, transparent
- Praise: Genuine, specific, inviting deeper connection
- Feedback: Thoughtful, non-defensive, action-oriented

Format:
[Message]
Category: [type]
Priority: [high/medium/low]
Draft response: [response]

---

Selling Skills as Products

The skills marketplace model: don't sell tools, sell results. (Credit: @wickedguro)

• ❌ "Postiz scheduling skill" — nobody buys tools
• ✅ "How to grow TikTok to 10k followers in 30 days" — people buy outcomes

Package your best workflows as skills with a specific result in the title. Add affiliate links to the tools used in the skill — even free skills can generate revenue.

Skill monetization ideas:

• How to make a viral TikTok reel in [niche]
• UGC video generator for e-commerce products
• Customer support skill with brand-specific tone
• SEO content machine for [niche]
• Niche content finder (e.g. "luxury cars viral clips")

---

CLI-First Tool Design

If you're building tools your agent will use, design them CLI-first. (Credit: @wickedguro)

Why CLI > API for agents:

• Less code for the agent to write → fewer tokens
• Less context rot from complex API wrappers
• Fewer mistakes — CLIs are predictable

Tools the community has built or is building:

• postiz-cli — social media scheduling
• agent-browser — browser automation
• beeper-cli — messaging inbox access
• x-research — Twitter/X research

When evaluating an API integration, ask: is there a CLI alternative? If not, build one. Your future agent sessions will thank you.

---

Setup Checklist

Phase 1: Foundation (Week 1)

• Create/refine your SOUL.md with voice guidelines
• Decide which marketing channel to automate first (social, email, SEO, ads)
• Identify 1–2 specific tasks to turn into skills

Phase 2: First Skill (Week 2–3)

• Write the skill template (copy one of the prompts above)
• Test it manually via Claude Code
• Set it as a recurring cron job or Telegram command
• Measure output quality for 1 week

Phase 3: Iteration (Week 4+)

• Refine based on what worked/didn't
• Add a second skill
• Consider integration (e.g., scheduled posts → Typefully → tweet analytics → next week's content)

Phase 4: Scale

• Document your winning prompts in /openclaw/skills/marketing/
• Set up performance tracking (likes, replies, conversions)
• Automate based on performance (more of what works, less of what doesn't)

---

---

Advanced: Multi-Agent Orchestration

Agent Teams for Collaborative Marketing

Claude Code's agent teams feature lets multiple AI agents coordinate simultaneously, each with specialized roles. This is where marketing automation scales from good to exceptional.

Use case: Campaign Research Sprint

Create an agent team to research launch strategy for [product]:
- Agent 1: Competitor ad analysis (dissects competitor creatives, messaging, targeting)
- Agent 2: Customer voice mining (analyzes reviews, Reddit, support tickets for real objections)
- Agent 3: Positioning pressure test (challenges assumptions with devil's advocate role)

Requirement: They must debate findings and reach consensus before synthesizing into a strategy doc.

Real results: 1 agent exploring alone anchors on the first decent idea. 3 agents actively debating? The angle that survives is battle-tested.

Use case: Parallel Content Production

Create a team for this week's content pipeline:
- Agent 1: Research (competitive gaps, trending topics, SEO opportunities)
- Agent 2: Writing (article drafting based on Agent 1 research)
- Agent 3: QA (plagiarism check, readability, accuracy verification, fact-checking)

Setting: Research and production happen simultaneously instead of sequentially.

Setup:

1. Enable in settings.json: "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
2. Use delegate mode (Shift+Tab) so the lead only coordinates, not executes
3. Require plan approval for high-stakes work
4. Use shared task list (Ctrl+T) to monitor progress

Common mistakes to avoid:

• Don't use teams for simple tasks (overhead isn't worth it)
• Don't let agents edit the same files (coordinated collisions)
• 3-5 agents is optimal (more = coordination overhead)
• Always require plan approval on complex work

Cost consideration: Token usage scales with team size. Each agent is a separate Claude instance. Worth it for complex research/production, overkill for one-off copy tweaks.

---

Content Pipeline at Scale (The Morgan Pattern)

Real case study: One founder built a system that published 80+ articles in 10 days with a squad of OpenClaw agents.

The Architecture:

• 🔍 Scout — Finds topics with keyword research (runs every 6h)
• ✍️ Quill — Writes 2000+ word SEO articles (runs hourly)
• 📚 Sage — Reviews for quality/plagiarism/readability (3x/day)
• 📢 Ezra — Publishes to production (every 3h)
• 📣 Herald — Amplifies on social media (2x/day)
• 👀 Lurker — Scouts Reddit for outreach opportunities (8h)
• 📊 Archie — Weekly analytics reports
• 📋 Morgan — Project manager (3x/day) — Unblocks bottlenecks

The Pipeline:

Scout → Quill → Sage → Ezra → Herald → Archie
         ↓
    Reject? Send back to Quill for revision

All agents coordinate through a shared Notion database with atomic status transitions (Backlog → To Do → In Progress → Review → Ready to Publish → Done).

Key Implementation Details:

Specialization over generalization: Each agent is great at ONE thing instead of okay at everything.

• Scout has keyword research tools + competitor analysis
• Quill has brand voice guidelines + SEO rules
• Sage has plagiarism checker (Copyscape), readability scoring (Flesch-Kincaid), accuracy verification

Quality gates that actually reject: 40% of first drafts get rejected by Sage with specific revision notes. After revision, 95%+ pass.

• Plagiarism via Copyscape (~$0.03 per check)
• SEO scoring on 100-point rubric
• Readability must score 60+
• Factual accuracy check

Claim locking for parallel agents: When multiple agents query for work simultaneously:

1. Each agent generates unique claim ID
2. Immediately writes claim + status change atomically
3. Re-fetches to verify claim stuck
4. If claim fails, tries next article
5. Add 25-second stagger between parallel agents

Result: Zero collisions.

Real Results (10 Days):

• 80+ articles written
• 41 published
• ~40% first-draft rejection rate
• $0.70 cost per article
• 15 minutes human time per day
• 15,000+ Google impressions generated
• 130+ clicks from search traffic

---

The Marketing Supercomputer

Build expert advisors trained on the greatest marketing minds in history. This transforms generic AI output into specialized strategic guidance.

The System (3 Layers):

1. Access Knowledge — Scrape YouTube transcripts from top marketers (Alex Hormozi, Seth Godin, Alex Hartley, etc.)
2. Extract Frameworks — Have Claude identify each unique principle, framework, and mental model
3. Build Advisors — Create reusable instruction sets that channel each expert's thinking

Step 1: Gather Content

"Go to [YouTube channel]. Scrape every transcript. Also search for '[person name] interview'
and '[person name] podcast' and pull top 50 results. Save as separate text files."

Result: 200+ pages of transcripts in <1 hour

Step 2: Build Folder Structure

marketing-supercomputer/
├── hormozi/
├── godin/
├── hartley/
├── scripts/
├── output/
└── CLAUDE.md

Step 3: Create CLAUDE.md (Most Important)

I am [your name], founder of [your company]. I sell [your offer] to [your audience].

This folder contains transcripts from Alex Hormozi's videos, podcasts, and guest appearances.

Your job is to EXTRACT, not summarize.

For every distinct framework or repeatable principle:
- Name it using Hormozi's own terminology
- State the core principle in one sentence
- Include a direct quote from the transcript
- Explain the underlying psychology of why it works
- Show specifically how I can apply this to [your specific business context]

Step 4: Extract Frameworks Upload transcripts as Project Knowledge. Run:

"Read every transcript. Extract every distinct framework, methodology, mental model,
and repeatable principle. If frameworks connect, note the relationships."

Step 5: Build the Advisor

You are a strategic advisor trained on Alex Hormozi's complete body of work. When I ask
a question in your area of expertise, apply the specific frameworks and principles extracted
from their content. Score my work 1-10 on each dimension. Identify the weakest element and
explain using their reasoning. Every recommendation must be grounded in something they've
actually taught. Push back if I'm making a mistake they've explicitly warned against.

Real Example: Landing page scored 6/10 by the Miller advisor. Issue: Leading with product features instead of customer problem. Villain buried too deep. Rewrote based on advisor feedback → new page scored 9/10.

Cost: One-time setup. Zero ongoing cost. Access to centuries of collected business wisdom.

---

Comprehensive Marketing Skills Library

Build any of these as standalone skills. Or stack multiple skills in a single workflow for compound output.

Core Skills:

1. ab-test-setup — Set up and run A/B tests
2. analytics-tracking — Install event tracking and attribution
3. competitor-alternatives — Build "vs" and alternative pages
4. content-strategy — Plan what content to create
5. copy-editing — Polish and tighten existing copy
6. copywriting — Write conversion-focused page copy
7. email-sequence — Build automated email flows
8. form-cro — Optimize forms for more conversions
9. free-tool-strategy — Build free tools that attract users
10. launch-strategy — Plan a product launch
11. marketing-ideas — Brainstorm marketing tactics
12. marketing-psychology — Apply behavioral psychology to marketing
13. onboarding-cro — Improve new user onboarding
14. page-cro — Optimize landing pages for conversions
15. paid-ads — Manage Google, Meta, LinkedIn ads
16. paywall-upgrade-cro — Convert free users to paid
17. popup-cro — Optimize popups and modals
18. pricing-strategy — Design and test pricing
19. product-marketing-context — Define positioning and audience
20. programmatic-seo — Build SEO pages at scale
21. referral-program — Build referral and invite systems
22. schema-markup — Add structured data for search

Advanced Skill Stacking Example:

Research (Perplexity MCP) → Brand Positioning → Copywriting → Ad Creation

Result: Full funnel built in one sitting (landing page + lead magnet + ads + SEO page + email sequence).

---

The Vibe Marketing Framework

Research → Skills → Build

Layer 1: Research — Deep market research, competitor analysis, finding angles others miss using Perplexity MCP or web crawling (Playwright).

Layer 2: Skills — Custom instruction manuals trained on proven frameworks (direct response, ad psychology, conversion optimization).

Layer 3: Build — Stack research + skills to produce actual marketing assets.

More context upfront = Better output. Every time.

Real practitioner built 17 custom marketing skills. Typical output per sitting:

• 20 landing page variations
• 20 ad variations (static + video)
• Lead magnet
• SEO content page
• Email sequence outline

---

Common Pitfalls

1. Robotic Voice

If your content sounds like AI wrote it, check your SOUL.md. Include:

• Your actual writing samples (3–5 tweets/posts you've written)
• Your personality quirks (do you use em dashes? curse? make jokes?)
• Your perspective (what do you believe, strongly?)

2. Generic Prompts

The difference between mediocre and great marketing automation is specificity:

• ❌ "Write a tweet about marketing"
• ✅ "You're @jackfriks, a marketing automation expert. You believe most agencies are overpriced. Your audience is indie hackers and solopreneurs. Write a tweet about why they should build their own marketing stack instead of hiring an agency. Include a spicy take and a link. Use your voice: sarcastic but helpful."

3. Not Measuring

Track everything:

• Engagement rate (likes/views)
• Reply rate (how many replies per post)
• Link clicks (did it drive traffic?)
• Conversion rate (from traffic to signup/sale)

Use these metrics to refine prompts weekly.

4. Forgetting the Human Touch

Automation is best for volume, not nuance. Reserve human time for:

• High-value conversations (CEOs, press, major customers)
• Crisis communication (when something breaks)
• Strategic decisions (quarterly planning)

---

Tools & Resources

Needed

• Claude Code (for skill building)
• SOUL.md (your voice guide)
• x-research skill (trending topic research)
• Typefully (for scheduling/approving social posts)

Optional (But Powerful)

• Serpstat API (for SEO keyword research)
• Perplexity API (for research automation)
• HubSpot API (for CRM integration)
• Facebook Ads API (for ad testing automation)

Prompts & Case Studies

• Source tweets and articles (archived) — The advanced sections above were distilled from 25 real-world tweets/threads and 10 in-depth articles from marketing automation practitioners. Common themes: agency-level SEO for $0, rapid ad copy generation, social media automation, content pipelines, cost optimization. Key case studies:
  • Agent teams for collaborative marketing
  • 80+ article pipeline case study ($0.70/article, $0.15/article human time)
  • Marketing Supercomputer (expert advisors system)
  • Vibe marketing framework (Research → Skills → Build)
  • 22+ reusable marketing skills library

---

Next Steps

1. Pick your first automation. (social media, email, SEO, ads)
2. Write your SOUL.md if you haven't (see Chapter 4)
3. Use a template prompt from this chapter or vibe-marketing-raw.md
4. Test manually via Claude Code for 1 week
5. Schedule it via cron/Telegram when it works
6. Track results and iterate

---

Session Cleanup for Cost Management

Marketing automation generates a lot of sessions — especially if you're running multiple agents on content pipelines. Clean up stale sessions to keep costs down:

# Clean up marketing agent sessions older than 3 days
openclaw sessions cleanup --agent marketing --older-than 3d

# See what would be cleaned up (dry run)
openclaw sessions cleanup --older-than 7d --dry-run

Schedule this as a weekly cron to prevent session bloat.

See Also

• [[05. Use Cases & Workflows]] — More implementation prompts
• [[07. Skills System]] — How to build production-grade skills
• [[10. Content Curation & Swipe Files]] — Content pipeline deep dive