The blank page isn’t where you get stuck. The blank page is where you stay stuck.
Alex Lieberman knows this. He co-founded Morning Brew (sold for $75M), now runs Tenex, and still — the hardest part of content isn’t writing. It’s knowing what to say.
“The blank page is the real enemy of consistent content,” he told Lenny Rachitsky. “And an AI Oracle makes it disappear.”
His solution: a “Content Machine” built entirely in Claude Projects. Six components. Zero blank pages. 40% of inbound leads at his last company came from employee content this system produced.
Here’s how to build yours — no coding, no API, just Claude Pro and a weekend.
The Blank Page Is the Enemy — Not Writing
Most people think content problems are writing problems. They’re not.
Lieberman’s insight: “Writing is actually the easy part. Knowing what to say is where most people stall.”
His system solves the knowing what to say problem. Then it solves the saying it well problem. Then it solves the saying it better next time problem.
Six components. Each handles one failure point:
- AI Oracle — Finds ideas so you don’t have to
- Voice Guide — Makes it sound like you, not ChatGPT
- Interview Panel — Extracts your actual expertise
- Writer’s Council — Quality control that doesn’t sleep
- Reinforcement Loop — Gets smarter every publish
- Distribution — Your team becomes your channel
Meet the Content Machine (Morning Brew Founder’s System)
6 Components, One Goal: Zero Blank Pages
| Component | Job | Tool |
|---|---|---|
| Oracle | Scans Slack/Notion/Gmail/X for 15 ranked ideas/day | Claude Project |
| Voice Guide | Stores your hooks, structures, patterns | Markdown file in Project |
| Interview Panel | 6 personas ask follow-ups until specifics emerge | Claude prompts |
| Writer’s Council | 6 evaluators score drafts, demand 9/10 | Claude prompts |
| Reinforcement Loop | Compares draft→published, extracts lessons | Claude prompt |
| Distribution | Employee advocacy + Creator Cup incentives | Notion + manual (start) |
The result: Lieberman voice-to-texts answers via Wispr Flow, the system drafts against his Voice Guide, the Council scores it, and it doesn’t publish until it clears 9/10.
“AI slop is mostly a people problem, not a model problem,” he says. “The only time the Content Machine produces slop is when the person being interviewed doesn’t share interesting enough ideas. The model is shaping clay, not inventing ideas. If the clay is bad, the sculpture will be bad.”
Before You Build: Map Your Current Workflow
Lieberman’s first step: “Map your current workflow before you touch any AI tool. Draw out every step of content creation as it already exists, then rebuild it from scratch with no constraints.”
Don’t skip this. Grab paper. Draw your actual process:
Current: Idea → ? → Draft → ? → Edit → ? → Publish → ? → Distribute
Where are the gaps? Where do things stall? Where does quality drop?
My map showed: Idea generation (stall), Expert extraction (shallow), Quality control (inconsistent), Distribution (non-existent).
Your map will differ. Build for your gaps.
Component 1: The AI Oracle — Your Idea Radar
What It Does
Scans your inputs (Slack, Notion, Gmail, meeting notes, X/LinkedIn follows) and surfaces 15 ranked content spikes daily. Half from internal signals, half from external accounts you follow.
Internal signals: Customer questions, team debates, metric changes, product decisions, failed experiments.
External signals: Industry news, competitor moves, framework discussions, contrarian takes from accounts you trust.
Build It in Claude (Prompt Included)
Create a new Claude Project called “Content Machine.” Add this as the Project Instructions:
You are my Content Oracle. Your job: scan the inputs I provide and surface 15 ranked content ideas daily.
INPUTS I'll provide (paste into chat):
- Internal: Slack threads, Notion pages, Gmail threads, meeting transcripts, metric dashboards
- External: Tweets/LinkedIn posts from accounts I follow (I'll paste URLs or text)
OUTPUT format for each idea:
1. **Headline** (working title)
2. **Source** (internal/external + specific reference)
3. **Angle** (why this matters now)
4. **Format** (essay, framework, case study, tweet thread, video script)
5. **Urgency** (Evergreen / This Week / Today)
6. **Confidence** (1-10: how strong is the signal?)
RANKING criteria:
- Relevance to my audience (founders, builders, operators)
- Specificity (vague trends score lower)
- Actionability (can reader DO something with this?)
- Uniqueness (is everyone already writing this?)
Give me the top 15, ranked. Number 1 = strongest signal.
How to use it: Every morning, paste 3-5 items from your inputs. Takes 5 minutes. Get 15 ranked ideas. Pick one. Done.
Beginner adaptation: No Zapier/Notion API? Just copy-paste. It works. The constraint forces curation — you only feed it what’s actually worth reading.
Component 2: Your Voice Guide — The Force Multiplier
What Goes In It
This is the single highest-leverage file in the system. Lieberman calls it “your voice lives in a Markdown file, and that file changes everything.”
Your Voice Guide contains:
– Top 10 performing posts (full text or links + why they worked)
– Hook formulas (your 3-5 go-to opening patterns)
– Content structures (your recurring frameworks: Problem→Solution→Proof, etc.)
– Language patterns (phrases you use, words you avoid, sentence rhythms)
– Hot takes (opinions you’re willing to publish)
– Anti-patterns (what makes you cringe — “leverage,” “deep dive,” “in today’s world”)
Template to Copy
Create voice-guide.md in your Claude Project knowledge:
# [Your Name] Voice Guide
## Top 10 Posts (Why They Worked)
1. [Title] - [Link] - Worked because: [specific reason]
2. [Title] - [Link] - Worked because: [specific reason]
...
## Hook Formulas (My Go-To Openers)
1. **Stat + Tension**: "X% of [audience] struggle with [problem]. Most [common solution] doesn't work."
2. **Contrarian**: "Everyone says [common advice]. Here's why it's wrong for [specific context]."
3. **Personal Stakes**: "I lost $[amount] / [time] making this mistake. Here's what I'd do differently."
4. **Question + Gap**: "You know [thing]. But do you know [deeper layer]? Most people miss it."
5. **Scene Setter**: "It's [time/context]. You're [situation]. Here's what happens next."
## Content Structures
- **Framework Essay**: Problem → Existing Solutions (Why They Fail) → My Framework → Application → Proof
- **Case Study**: Context → Challenge → Approach → Results → Lessons → Template
- **Hot Take**: Common Belief → Why It's Wrong → Better Model → Evidence → Implications
- **Tutorial**: Goal → Prerequisites → Step 1...N → Common Pitfalls → Next Level
## Language Patterns
**Use:** "Here's the thing," "Sound familiar?", "Short answer:", "I'll be honest,"
**Avoid:** "In today's world," "It's worth noting," "Comprehensive guide," "Leverage," "Seamless"
## Hot Takes (Opinions I'll Defend)
- [Your contrarian view 1]
- [Your contrarian view 2]
- [Your contrarian view 3]
## Anti-Patterns (What Makes Me Cringe)
- Corporate speak ("synergy," "paradigm shift")
- Fake humility ("I'm no expert but...")
- Filler transitions ("Furthermore," "Moreover," "Additionally")
Pro tip: Spend 2 hours on this. Paste your actual best posts. Extract real patterns. This file is what makes the output sound like you.
Component 3: The Interview Panel — Extract Your Expertise
6 Personas That Ask the Right Questions
Lieberman uses six interviewer personas. Each has a distinct style. Together, they extract specifics you’d miss alone.
| Persona | Style | Best For |
|---|---|---|
| Tim Ferriss | Deconstructs tactics, asks “how exactly?” | Tactical frameworks |
| Joe Rogan | Curious, follows tangents, “wait, what?” | Story extraction |
| Larry King | Simple questions, “why?” “then what?” | Clear explanations |
| Howard Stern | Provocative, pushes uncomfortable | Hot takes, vulnerabilities |
| Barbara Walters | Emotional core, “how did that feel?” | Personal narrative |
| Michael Barrow | Business focus, “what’s the metric?” | Case studies, ROI |
Prompt Templates for Each
Add these as separate prompts in your Project (or keep in a prompts/ folder in Project knowledge):
Tim Ferriss Interviewer:
You are Tim Ferriss interviewing me for content.
Style: Deconstruct tactics. Ask "how exactly?" "What's the step-by-step?" "What would you tell your 25-year-old self?"
Goal: Extract a repeatable framework or process.
Rules:
- One question at a time
- Push for specificity ("Give me an example")
- Ask for the template/checklist/process
- Stop when you have enough for a 1500-word tactical guide
Joe Rogan Interviewer:
You are Joe Rogan interviewing me.
Style: Genuinely curious. Follow interesting tangents. "Wait, what do you mean by that?" "That's wild — tell me more."
Goal: Extract stories, unexpected insights, human moments.
Rules:
- One question at a time
- React to answers ("No way," "That's fascinating")
- Go deep on surprising threads
- Stop when you have 3-5 story beats worth expanding
Larry King Interviewer:
You are Larry King interviewing me.
Style: Simple, direct. "Why?" "And then what happened?" "What did you learn?"
Goal: Extract clear, jargon-free explanations anyone can understand.
Rules:
- One question at a time
- No jargon allowed in your questions
- If I use jargon, ask "What does that mean in plain English?"
- Stop when the concept is explainable to a smart 15-year-old
Howard Stern Interviewer:
You are Howard Stern interviewing me.
Style: Provocative. Pushes uncomfortable. "Be honest — did you screw that up?" "What are you afraid people will think?"
Goal: Extract vulnerabilities, mistakes, contrarian truths.
Rules:
- One question at a time
- Challenge safe answers
- Ask for the version I wouldn't tell my mom
- Stop when you have a "hot take" or vulnerable moment worth publishing
Barbara Walters Interviewer:
You are Barbara Walters interviewing me.
Style: Emotional core. "How did that feel?" "What was the moment you knew?"
Goal: Extract the human story behind the achievement.
Rules:
- One question at a time
- Focus on turning points and feelings
- Ask for the scene (sights, sounds, internal monologue)
- Stop when you have an emotional arc worth building a narrative around
Michael Barrow Interviewer:
You are Michael Barrow (business operator) interviewing me.
Style: Metrics-driven. "What's the number?" "What did it cost?" "What's the ROI?"
Goal: Extract business case studies with numbers.
Rules:
- One question at a time
- Every claim needs a number
- Ask for the spreadsheet/dashboard
- Stop when you have a complete before/after with metrics
How to use: Pick ONE interviewer per piece. Paste the prompt, then answer via voice (Wispr Flow) or text. The interviewer asks follow-ups. You answer. When they stop, you have raw material.
Component 4: The Writer’s Council — Quality Control
6 Evaluators, One Score (Must Hit 9/10)
Lieberman’s Writer’s Council: six writer personas score every draft. Aggregate must hit 9/10. Revision loop until it clears.
The 6 Evaluators:
| Persona | Focus | Pass Threshold |
|---|---|---|
| Clarity Editor | “Can a tired reader understand this in one pass?” | 9/10 |
| Hook Surgeon | “Does the opening stop the scroll?” | 9/10 |
| Evidence Auditor | “Every claim backed? No hand-waving?” | 9/10 |
| Voice Guardian | “Does this sound like the Voice Guide?” | 9/10 |
| AI Slop Allergist | “Any generic filler? Corporate speak? Fake transitions?” | 9/10 |
| Action Architect | “Does the reader know exactly what to DO next?” | 9/10 |
Scoring Rubric
Each evaluator uses this scale:
– 10 = Perfect, ship it
– 9 = Minor tweak, then ship
– 8 = Needs revision, specific feedback required
– 7 = Major structural issue
– ≤6 = Rewrite from scratch
Council Prompt (single prompt runs all 6):
You are the Writer's Council. Evaluate this draft against the Voice Guide.
DRAFT: [paste draft]
VOICE GUIDE: [reference voice-guide.md in project]
Score each evaluator (1-10) with specific feedback:
1. **Clarity Editor** - Can a tired reader understand in one pass?
2. **Hook Surgeon** - Does the opening stop the scroll?
3. **Evidence Auditor** - Every claim backed? No hand-waving?
4. **Voice Guardian** - Sounds like the Voice Guide?
5. **AI Slop Allergist** - Any generic filler, corporate speak, fake transitions?
6. **Action Architect** - Reader knows exactly what to DO next?
OUTPUT:
- Individual scores (1-10 each)
- Aggregate score (average)
- Specific feedback per evaluator (2-3 bullets each)
- VERDICT: PASS (≥9 aggregate) or REVISE (specific changes needed)
Run this after every draft. If aggregate <9, apply feedback, re-run. Repeat until PASS.
Component 5: The Reinforcement Loop — Gets Smarter Every Time
The Prompt
After publishing, run this:
Compare the ORIGINAL DRAFT (pre-Council) vs PUBLISHED VERSION.
ORIGINAL: [paste]
PUBLISHED: [paste]
Extract:
1. What changes did the Council force that improved the piece?
2. What did I change on my own instinct?
3. What patterns emerge? (e.g., "I always bury the lead," "My hooks are too long")
4. What ONE lesson should be added to the Voice Guide permanently?
Output:
- 3-5 specific observations
- 1 permanent Voice Guide addition (exact wording)
This is how the system compounds. Every piece makes the next one easier.
Component 6: Distribution — Your Team Is Your Channel
Lieberman’s hardest-learned lesson: “Your employees are your most underleveraged marketing channel, and most companies are actively suppressing them.”
At Storyarb: “Own the Internet” program → employees post → 40% of inbound leads.
At Tenex: Creator Cup ($5K prizes) → posting becomes team sport.
Beginner version (solo/小团队):
– Share every draft with 2-3 trusted peers before publishing
– Ask: “Would you share this? Why/why not?”
– When you publish, send link to 5 people: “Thought you’d find this useful”
– Track: which pieces get shared organically? Do more like those.
Scale version (team):
– Monthly Creator Cup: best post wins $500
– Criteria: Council score + engagement + peer votes
– Public leaderboard in Notion
– Quarter-end: $2K grand prize
Putting It All Together: Your First Run This Weekend
Minimal Viable Version (3 Components)
Saturday morning (2 hours):
1. Create Claude Project “Content Machine”
2. Build Voice Guide (paste 5 best posts, extract patterns)
3. Add Oracle prompt to Project Instructions
Saturday afternoon (1 hour):
4. Feed Oracle 3 inputs from this week → get 15 ideas
5. Pick #1 idea
6. Run Interview Panel (pick ONE persona) → answer via voice
Sunday morning (2 hours):
7. Draft from interview transcript
8. Run Writer’s Council → revise until PASS
9. Publish
10. Run Reinforcement Loop → update Voice Guide
Total: ~5 hours. One published piece. System live.
Full Version (All 6 Components)
Add Friday: Distribution setup (Creator Cup, peer review list)
Add Monday: Reinforcement Loop on Monday’s publish
Add ongoing: Weekly Oracle feed, monthly Voice Guide audit
Pro Tips from the Trenches
1. Voice Guide is the force multiplier
Lieberman: “Your voice lives in a Markdown file, and that file changes everything.” Spend disproportionate time here. Update weekly.
2. Map your workflow before touching AI
“Most companies discover huge inefficiencies in this step that have nothing to do with AI at all. AI just forced them to look.”
3. The interview panel is the step most people skip
“Six interviewer personas ask follow-up questions until they’ve extracted enough specifics to write something real.” Don’t skip. The raw material quality determines everything downstream.
4. AI slop is a people problem
“The only time the Content Machine produces slop is when the person being interviewed doesn’t share interesting enough ideas.” Feed it better clay.
5. Your employees are your distribution
“The fastest way to lose a great employee is to make it impossible for them to build a personal brand.” Enable them. Incentivize them.
6. Bootstrapped beats VC-backed on distribution
“Tenex can’t rely on TechCrunch coverage. The team’s content IS the visibility strategy. If Creator Cup gets one great engineer, $5K pays back many times over.”
Your Next Step
This weekend:
1. Create the Claude Project
2. Build the Voice Guide (2 hours — use the template)
3. Add the Oracle prompt
4. Feed it 3 things from this week
5. Pick one idea, run one interview, write, Council, publish
That’s it. The system works because it’s a system — each component covers the previous one’s weakness.
The blank page doesn’t stand a chance.
Source: Lenny’s Newsletter “How the founder of Morning Brew built a Claude content machine that never runs out of ideas” — lennysnewsletter.com/p/how-i-ai-how-the-founder-of-morning
Detailed workflow walkthroughs: chatprd.ai/how-i-ai/alex-liebermans-6-step-workflow-to-beat-ai-slop
