Claude Code Content System: 10M Views in 30 Days

Over the last month I went from nothing to 38,000 new YouTube followers, 50,000 new Instagram followers, and 11,000 new TikTok followers — driven by 10 million views across 90 pieces of content, all as a one-person operation with no editor and no VA. The system behind it is seven custom Claude Code skills mapped to four phases of content creation: research, ideation, scripting, and distribution. And the real metric isn't the 10 million views — it's the 90 videos in 30 days without a single one breaking 400k. This is small consistent jabs, not a lucky haymaker.

I'm going to break down the full workflow, the skills I built, and how each one plugs into a phase of the pipeline so you can copy the structure. You don't need my exact skills. You need the same shape of system tuned to your niche.

How does Chase's Claude Code content system actually work?

The system breaks content creation into four phases — research, ideation, scripting, and distribution — and assigns specific Claude Code skills to each phase. Some phases have multiple skills because there's more going on. Scripting, for example, covers hooks, outlines, titles, and thumbnail text. Every skill has one job.

The seven skills: YouTube pipeline research, ideation, hooks, outlines, titles, content cascade, and short-form repurposing. Plus two supporting pieces — a Twitter research web app and a trending-GitHub daily script — that feed the "step zero" of finding things worth talking about.

All the generated content lands in my Obsidian vault as markdown files, so I can see everything visually and reference back. That matters more than it sounds. When you're running this kind of volume, losing track of what you've researched and what you've published is a real risk, and Obsidian keeps it all organized.

Where do the ideas come from in the first place?

Before any of the phase-one skills kick in, you need step zero: a fountainhead of knowledge for your niche. For me — AI and tech — that's GitHub and Twitter. Ideas originate on Twitter or get released as GitHub repos, then eventually propagate to YouTube. If you can sit upstream of that flow, you're talking about things before most creators.

Two custom tools handle this:

Twitter research web app. Claude Code built me a scraping app that runs every 30-45 minutes (randomized timer), pulls 40-90 tweets via an API, scores them on five signals (velocity, authority, timing, opportunity, replyability), deduplicates against Supabase, applies a softmax probability to pick which ones surface, and pushes the picks to Telegram. It also has Grok wired in so I can send AI-assisted replies back. All replies get scored and stored to build a feedback loop over time.
Trending GitHub script. Runs every morning, pulls the top 10 trending AI repos created in the last seven days plus the top five for the month, and drops a markdown file in my Obsidian vault with stars, language, description, and its own recommendation.

You don't need my exact tools. Identify where ideas originate in your niche and have Claude Code build you the scraper that sits on top of it.

How does the YouTube pipeline skill work for research?

Once step zero surfaces an idea, the YouTube pipeline skill takes it from idea to deep analysis. This is the single most powerful skill in the system.

It's a "higher-order skill" — it calls another skill under the hood. The underlying piece is the NotebookLM-API CLI tool, which lets Claude Code talk to NotebookLM through the terminal. I can feed Notebook LM YouTube URLs, PDFs, documents — anything I could feed the web UI — but through Claude Code.

The pipeline skill wraps this. Given a topic, it sources relevant YouTube URLs automatically, pushes them into a new NotebookLM notebook, and runs analysis. The heavy lifting happens on Google's servers via Gemini, not on Claude Code tokens, so it's cheap. The output — summaries, slide decks, images, whatever NotebookLM deliverables I want — comes back into my vault as markdown.

This is how you do deep competitive research on a topic without either burning Claude Code tokens or manually clicking through NotebookLM for an hour.

How does ideation turn research into video concepts?

Research tells you what's out there. Ideation tells you where the gaps are and what would actually resonate with your audience. The ideation skill takes the research output from the vault and places it in a competitive landscape:

Saturated angles (what's been done to death)
Open gaps (what hasn't been covered)
Performance outliers (what blew up for someone else)
Video idea candidates with titles, angles, desire mapping, formats, and competitive gaps
A ranked list

I ran it recently for a "seven levels of Claude Code and RAG" concept. It came back with nine options with reasoning for each. I picked one and moved forward. Nine times out of ten I end up doing a variant of what it suggests — not the exact version, but the direction is usually right.

This is where Claude Code's role as collaborator matters more than "automator." You need to stay in the driver's seat. The skill does hours of competitive research in seconds; your job is to pick.

What do the scripting skills actually produce?

Three skills cover the scripting phase: hooks, outlines, and titles. I run them back-to-back on the same topic.

Hooks skill. Generates five hook variations per video. Each hook has three aligned pieces: the spoken line, the visual hook, and a text overlay option (mainly for short-form). I don't script whole videos — I script the first 30 seconds because the hook is the single highest-leverage piece of any content. Everything after is bullets and improvisation.

The hook framework leans heavily on Kallaway's desire-based framework. I trained the skill on his videos and my own high performers, so the output stays aligned with what actually works for my niche.

Outline skill. Given the topic and hook, produces a full video outline: target length, relevant related documents from my vault, section-by-section talking points, visual aid suggestions (including Excalidraw diagrams if useful), and source material references. Big picture — not a word-for-word script.

Titles skill. Pulls from my existing top-performing titles as reference data and generates title options in tiers. Tier 1: safe, data-backed titles with "this previous video got X views, here's why this title would work." Tier 2: calculated risks — weirder titles worth testing for variance. Each set also comes with thumbnail text options.

By the end of those three skills I have 90% of the video packaged — hook, outline, title, thumbnail text. I just need to film and edit.

How does the content cascade handle distribution?

Distribution has two layers: platform posting and repurposing across formats. Platform posting I mostly do manually — CapCut pushes to YouTube and TikTok, Instagram reels I post by hand because automating Reels trials is a pain and not worth the effort for me.

Repurposing is where the content cascade skill earns its place. Given a YouTube video, it:

Grabs the transcript automatically.
Generates a full blog post and auto-posts it to my Chase AI blog (with embedded YouTube video, SEO metadata, and written in my voice — I fed the skill a pile of my own writing so it avoids the ChatGPT-isms like "it's not X, it's Y").
Generates a Twitter thread with seven follow-up replies. Auto-posts after I approve it.
Generates LinkedIn post variations (I don't auto-post LinkedIn — I use Lead Shark for the lead magnet flow there).

One video becomes a blog post, Twitter thread, and LinkedIn post without me touching it beyond approval.

How does the short-form skill turn long videos into Reels?

The short-form skill is essentially a condensed version of the hooks + outline skills tuned for 30-60-90 second clips. Given a long-form YouTube video, it identifies clip-worthy moments, generates hooks for each, suggests on-screen captions, and outputs structured clip prompts.

This is what lets me get 60 short-form pieces out of ~30 long-form videos in a month. One YouTube video becomes a blog post, a Twitter thread, a LinkedIn post, a YouTube Short, an Instagram Reel, and a TikTok — six platforms from one source. That's where the "content cascade" name comes from.

Why Claude Code beats fully automated content bots

The instinct with AI content is to build the fully automated pipeline — research to script to published, no human touch. Don't do that. Your content will suck.

Claude Code works as a collaborator because at every step, I'm there to redirect. "Don't go with that angle, go with this one." "I don't like this hook, give me five more variations on this specific frame." "Cut the third section, it's a tangent." If I let the system run end-to-end with no input, the output would be generic AI slop and the audience would feel it instantly.

The system buys you the leg work, not the judgment. All the competitive research, hook drafts, outline bullets, and distribution formatting gets offloaded. You stay in charge of taste and voice. That's the only way this math works — 90 videos in 30 days as one person — without the quality falling off a cliff.

FAQ

How many Claude Code skills does the full content system use?

Seven main skills: YouTube pipeline research, ideation, hooks, outlines, titles, content cascade, and short-form. Plus two supporting tools — a Twitter research web app and a trending GitHub daily script — that feed idea discovery at the "step zero" layer.

Can Claude Code fully automate content creation?

Technically yes, practically no. Running a skill chain end-to-end with no human input produces generic, voice-less content that audiences reject. The system works because Claude Code handles the leg work — research, drafts, formatting — while a human stays in the driver's seat for taste and direction.

What does the YouTube pipeline skill do?

It's a higher-order skill that sources relevant YouTube URLs on a topic, pushes them into a NotebookLM notebook via the NotebookLM-API CLI, runs analysis on Google's servers, and returns summaries and deliverables into an Obsidian vault as markdown. It moves the compute cost off Claude Code tokens and onto Gemini.

How does the content cascade repurpose a single video?

Given a YouTube video, the content cascade skill grabs the transcript, generates a full blog post with SEO metadata and auto-publishes it, produces a Twitter thread with seven replies for auto-posting after approval, and outputs LinkedIn post variations. One video becomes four text formats automatically.

Do you need to automate posting to social media?

Not really. Platform posting is often faster to do manually — CapCut publishes straight to YouTube and TikTok, and Instagram Reels trials are hard to automate cleanly. Focus automation on research, scripting, and repurposing where the time savings are real. Publishing is usually the cheapest step.

If you want to go deeper into building your own Claude Code content system, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.