The Content Engine That Added $151K MRR in 87 Days

We rebuilt our entire content engine inside Claude Code.

In 87 days, this system added $151K MRR with 24 people posting across LinkedIn, Twitter, newsletters, blogs, and video scripts. It started as a small experiment with one person, then two, then three. Mai-Lan Khong turned the experiment into a replicable system that runs on 27 skills inside one terminal.

The output sounds nothing like a marketing team. It sounds like 24 individuals writing in their own voice, at scale.

Here is the exact architecture behind it.

1. The Foundation

Every person who joins the engine goes through a one-time setup. Skip this step and the engine produces polished content that all sounds the same. That is the trap.

Voice profile

We run a 25 question conversation with each person, not a form. The goal is to capture how someone actually talks before any AI touches their content. Word choice, sentence rhythm, pet phrases, the things they refuse to say.

ICP document

Three buyer tiers, mapped to the exact language each tier uses to search for solutions. The same pain point reads differently to a founder than to a Head of Growth than to an enterprise buyer. The ICP doc captures that language gap.

Content pillars

Three to five themes every post traces back to. Without pillars, posts drift into trending topics that have nothing to do with your positioning. With pillars, every post compounds.

People skip this layer and jump straight to writing. Then they wonder why everything their AI produces sounds identical to everyone else's AI output.

2. The Research Layer

The research layer feeds the production line with validated ideas and audience language. It runs weekly across five sources.

Apify scrapes LinkedIn for what is performing right now in your niche. It surfaces the hooks, angles, and formats earning impressions this week, not last year.

Reddit pulls the real audience pain language straight from the communities your buyers participate in. People say things on Reddit they would never say in a gated survey.

YouTube surfaces frameworks worth adapting. Long-form creators stress-test ideas in 30-minute formats, which means their frameworks are usually battle-tested before you see them.

X and Twitter track live debates in your space. What is getting quote-tweeted, where the industry disagrees, which takes are building momentum.

Fireflies.ai transcripts pull insights from real client calls. The exact words prospects use to describe their problem become hooks, headlines, and pillar topics.

Five sources, one weekly run. The output is a library of validated ideas pre-matched to each person's voice profile and pillar list.

3. The Production Line

Once an idea is validated, the production line turns it into a finished post. Three skills handle the core draft, then visuals get added on top.

Hook generator

50 templates, 20+ variations per idea. The generator produces a menu of openers ranked by hook strength, so the writer picks one instead of staring at a blank page.

Copy developer

A full draft written in the person's documented voice. The voice profile built in step one is the guardrail here. AI drafts, the writer sharpens.

Post grader

A 5-dimension rubric scored out of 50. Any post below 38 goes back to rewrite. No "good enough" passes. The grader enforces the quality bar without needing a human editor in every loop.

For visuals, Gemini handles first drafts and Figma plus Canva refine them into brand-aligned assets.

You can preview how your LinkedIn content will look before publishing, for free:

LinkedIn Post Previewer Tool

4. The Repurpose Engine

One LinkedIn post that performs well is a validated idea. A validated idea deserves more than one format.

The repurpose engine rebuilds the same core idea for:

→ X and Twitter threads

→ Newsletters

→ Blog articles

→ Video scripts

→ Carousels

Each format is rebuilt from scratch. A LinkedIn hook that works in the feed falls flat in a newsletter. A newsletter intro that earns opens reads like filler in a thread.

Repurposing is restructuring. One idea, many formats, each one rebuilt for how people consume content on that platform.

5. Refresh and Maintain

A content engine that never self-corrects drifts fast. Four automated loops keep the system sharp.

Feedback capture

Every writing session gets logged automatically. What the writer changed, what they left alone, what broke the voice profile. That log feeds pattern recognition.

Pattern recognition

Runs every 5 sessions. The system scans recent edits for recurring rewrites. If the same phrase keeps getting cut, that is a voice drift signal. If the same hook keeps underperforming, that is a template retirement signal.

Monthly voice refresh

New client call transcripts get pulled in and compared against the original voice profile. People change how they talk over time. The refresh catches that drift before it shows up in the content.

Content audit

Runs quarterly. Are your pillars still aligned with your positioning? Has the ICP language shifted? The audit forces a strategic check-in, not just a tactical one.

6. Delivery

Once a post is ready, three tools handle handoff, scheduling, and performance feedback.

ClickUp handles the client handoff. Drafts move from the writer through review into a scheduling queue the client can see.

Taplio schedules posts across personal LinkedIn accounts with native-feeling timing and analytics.

Performance data flows back into Claude Code for the next cycle. What performed, what flopped, what the algorithm rewarded this week. The engine learns from its own output.

7. Why This Works at Scale

27 skills. One terminal. Each skill does one job well, and none of them try to replace the human in the loop.

The shortcut nobody talks about is this:

→ AI drafts, humans sharpen

→ The voice profile is the guardrail

→ The grader is the quality bar

→ The repurpose engine multiplies distribution

→ The refresh loop prevents drift

That is how 24 people sound like themselves at scale, rather than sounding like the same AI output wearing different names.

Your folder structure becomes your content brain. Each person's voice profile, ICP doc, and pillar list lives alongside the skills that use them. The system scales because the architecture is clean, not because AI replaced anyone.

Shoutout to Pilar Varela for designing the cheat sheet that documents this system.

You can see how your full GTM motion performs across content, outbound, and conversion, for free:

GTM Reports Tool

---