X creator @yidabuilds (a “100-Year AI × Going Global” account) publicly shared the content production system behind his uninterrupted daily posting streak in a tweet. The prototype of this system comes from the Claude Code workflow shared by @dontbesilent, adapted for X long-form posts. As of writing, the system has supported him in publishing over a dozen long-form posts, averaging around 120K views per post, with 9000+ cumulative followers gained and a bookmark rate steadily at 0.5-1%.

This article is compiled from the full content of that tweet, preserving the core methodology and data details, for creators with similar needs to reference.

Core Idea: From Fragmented Questioning to a Closed-Loop System

The core problem @dontbesilent identified: most people use AI for content creation in a fragmented way — ask AI whenever an idea comes up, post whatever answer you get, forget about it, then start from zero next time.

His solution is to turn the whole process into a closed loop: ideas go into a topic bank → AI searches the material library for reusable things → write using proven frameworks → publish → review the data → distill effective patterns back into the methodology. Every piece of creation adds something to the system instead of reinventing the wheel each time.

@yidabuilds adopted this approach directly and made extensive modifications based on his own needs.

Four-Layer Knowledge Base: Obsidian + Claude Code

The whole system uses Obsidian to manage content and Claude Code to run automated workflows, split into four layers.

Layer 1: Corpus — Solving “AI Doesn’t Write Like You”

The biggest problem with AI writing long-form content is that it sounds too AI-generated. Long-form readers read word by word, and the moment it feels machine-written, something feels off.

@yidabuilds’s solution is to save everything he’s ever said: tweets, opinions from WeChat chat records, recordings, and random fragmented thoughts jotted down. From these he distills a writing style guide, for example:

• Prefers to state the conclusion first, then give reasons
• Prefers numbers over adjectives
• Likes explaining the current topic using logic from other industries
• Doesn’t wrap up with inspirational platitudes

Before each draft, AI reads this guide first, and the resulting first draft is at least 70-80% like the real person. After writing, he runs a “de-AI-ification” check to flag overly machine-sounding expressions for manual revision.

The detection rules include:

With these two steps added, the human feel of long-form posts improved noticeably.

Layer 2: Material Library — Choosing Your Path Standing on Others’ Data

The material library contains breakdowns of 47 similar accounts, data from 1100+ pieces of content, structural analysis of viral articles, and reusable concepts and quotable lines.

Before writing each new article, AI first goes through the library: who has written on this topic, what angle performed well in the data, what structure readers are willing to save.

After breaking down 47 accounts, a few findings directly shaped the topic strategy:

• Content reaching 1M+ views only falls into 5 categories: must-have tool tutorials, medical/health science explainers, AI + making money, persona analysis pieces, and resource roundups. Other types almost never break a million
• Bookmark rate and views aren’t necessarily correlated. Some articles have mediocre views but high bookmark rates, meaning the content has long-term value — readers save it for later use. This type of content is worth writing repeatedly
• Follower growth and views aren’t necessarily correlated either. A persona piece with 119K views gained 156 followers; a tutorial with 77K views gained only 25. Personas make people want to follow the person; tutorials get saved and left behind

Layer 3: Content Pipeline — Choosing Topics by Strategy, Not by Whim

The pipeline has four stages: topic pool → to be deepened → in progress → published. The pool permanently holds a dozen-plus topics ready to write at any time, plus a dozen-plus alternatives needing more material.

Topics rotate across several tracks: hands-on projects, AI money-making track breakdowns, grassroots low-barrier businesses, and new AI paradigm trends. Each track has a different topic strength:

• Hardcore tool tutorials: highest views
• Persona self-introductions: fastest follower growth
• Data reviews: narrower audience but better bookmark rates

Choose topics based on current goals: want views, post tutorials; want followers, post personas; want compounding, post reviews.

Layer 4: Methodology — Distilling Patterns from Your Own Data

What headlines work, what topics blow up, what structures get bookmarked — all distilled from your own publishing data.

Headlines are the most quantifiable part. After running a dozen-plus long-form posts, the headlines that generated data basically fall into four patterns:

Before publishing, run through a checklist: are there concrete numbers? Is there an identity tag? Is there contrast? Can readers tell from the headline what they’ll get? The more boxes ticked, the better the data.

Image Strategy: HERO Image, Headline, and Hook Carry Non-Overlapping Information

Images are crucial for X long-form posts. In the feed, the reader’s attention order is HERO image > headline > body. If the image is weak, no one clicks no matter how good the headline is.

Principle: the HERO image, headline, and hook deliver three layers of different information — no repetition.

• HERO image: tells you at a glance “what type of content this is”
• Headline: gives a data anchor that makes people stop
• First paragraph: unfolds the details

Two image styles, auto-selected by content type:

• Tutorials: infographic style — white background, light decorative bubbles, rounded cards, flat icons, large Chinese headline, like a clean SaaS-style hero banner
• Opinion pieces: concept poster — large type as the visual skeleton, characters and text interlocking, like an exhibition poster rather than a PPT

Each long-form post gets one cover plus two or three inner infographics. AI generates prompts based on the article content, calls the GPT Image 2 API to generate images, then downloads and crops to the needed aspect ratios. Where Canva used to take half an hour per image, now three images take 10 minutes.

Data Performance: 120K Average Views, 0.5-1% Bookmark Rate

The AI fortune-telling piece had the highest bookmark rate at 1.01% — the AI + making money + information asymmetry combo gets readers saving most actively.

Four Rules Grown Out of the Data

“Growing rules out of data” is @dontbesilent’s core methodology. Here are the concrete rules that came out of @yidabuilds’s own X long-form data:

Headlines must contain concrete numbers. “Made 100K in 4 months,” “$155 vs $15,” “452% return” — every long-form post that performed carried hard numbers. In the feed, numbers are the thing most likely to make people stop.

AI has to be the protagonist. AI tutorial long-form posts consistently clear 100K views; pure investing content struggles to pass 50K. People come to this account to see “how to use AI,” not “how to trade stocks.”

“Saving you time” is the underlying sharing logic. WeChat article scraping, Codex beginner guides, image production practice — the common thread across all viral long-form posts is “I’ve tried it, hit the pitfalls, and organized it for you — just follow along.”

The viral formula: hardcore tutorial or real experience + concrete data anchoring + a reproducible path. No viral headline was ever an abstract concept. They’re all “I did X, and here’s result Y” — sharing experience plus data, not teaching a class.

These rules get updated with every new post. The system is self-correcting.

How to Get Started: No Need to Do It All at Once

dontbesilent’s dbskill (4000+ stars on GitHub) is a great starting point. You can also do what @yidabuilds did — take the core ideas and adapt them to your own needs.

No need to go all-in at once. First build the topic pool and material library, run for two weeks, and let the data tell you which direction to adjust.

*This article is compiled from the tweet by X user @yidabuilds. Original link: https://x.com/yidabuilds/status/2052252982302200010