Deep Content Writer v2
A structured system for writing content that readers actually finish. v2 improvement: Depth-first, structure-second.
The structure keeps readers reading. The depth makes them share.
Core Belief
A great article is NOT a template filled with words. It's a specific claim backed by specific evidence, told through story.
Depth comes from three things, in this order:
- Specific research (studies, data, names, dates)
- Cross-domain connections (linking this topic to an unexpected field)
- Structural clarity (making all the above easy to read)
The 6-part arc only solves #3. The skill MUST solve #1 and #2 first.
Phase 0: Knowledge Reconnaissance (MANDATORY)
Before writing a single word, do this:
Step 0a: Deep Dive (self + search)
Ask yourself (the agent/LLM):
- "What do I ALREADY know in my training data about this topic?"
- List 3-5 concrete studies, experiments, historical events, or named researchers related to this topic
- If you can't name specifics, search the web now
Then search the web for:
- Recent studies on this topic (last 5 years preferred)
- Counterintuitive data points — numbers that surprise people
- Named researchers or institutions doing work in this area
- Real-world case studies with specific companies, people, or events
- Public debates or controversies around this topic
Step 0b: Find the ANTI-Intuition
Every good article challenges a common belief. Find it:
| Question | Why it matters |
|---|---|
| What does "everyone know" about this topic? | That's the straw man to knock down |
| What's the counterintuitive truth? | That's your article's reason to exist |
| What would surprise a smart person about this? | That's your hook |
| What other domain has a parallel pattern? | That's your analogy |
If you can't find at least ONE anti-intuition, the article doesn't need to exist yet. Wait, research more.
Step 0c: Specificity Check
For every major claim you plan to make, ask:
Claim: "People are bad at estimating their own competence."
→ Specific? No.
→ Fixed: "In 1999, Kruger and Dunning found that the bottom quartile
of test scorers rated themselves in the top 60% — a 40-point gap."
→ Now it has names, year, and a number. Ready to write.
Rule: Every paragraph must have at least one specific: a name, a date, a number, a place, or a quote. If a paragraph has none, it's filler — cut or research it.
Phase 1: Writing — The 6-Part Arc
Once you have your research assembled, structure the article:
1. HOOK — Open with a concrete, relatable moment
- A specific person in a specific situation
- A surprising statistic
- A short anecdote
Don't say: "Communication is hard." Do say: "A Stanford researcher once asked people to tap a song on a table. The tappers thought listeners would guess the song 50% of the time. The real number: 2.5%."
The hook must be SPECIFIC to earn its place. No generic openings.
2. DEFINE — State the concept in one compelling sentence
Define through contrast, not dictionary:
"It's not [what people think]. It's [what it actually is]."
Example:
"It's not that experts are bad teachers. It's that once you know something, you literally cannot imagine what it's like not to know it."
Then expand for 1-2 paragraphs with the research foundation.
3. THE MECHANISM — Why does this happen?
This is where your research pays off:
- Which cognitive bias or psychological mechanism is at work?
- Who discovered/studied it? When? What did they find?
- What evolutionary or structural reason explains it?
If you don't have a named study or researcher here, stop and search. This section is the backbone of the article's authority.
4. EXAMPLES — 2-3 concrete scenarios
Each example must be a MINI-STORY:
- Example 1: Everyday life (reader thinks "that's me!")
- Example 2: Professional/work context (reader thinks "that's my coworker!")
- Counter-example: What happens when someone does it RIGHT
Each example: 1-2 paragraphs, with specific details (names, quotes, dialogue, situations).
5. TAKEAWAY — What the reader can DO
3-5 actionable, specific pieces of advice.
Each takeaway must pass the "So what?" test:
- ❌ "Communicate better."
- ✅ "Before explaining anything, ask yourself: what does this person already know? Start there, not at the beginning."
6. CLOSE — Circle back, provoke
- Reference the hook
- Leave the reader with a question or challenge
- Invite discussion with a specific question
Phase 2: Depth Techniques (The real skill)
These techniques separate good AI writing from great writing. They MUST be applied during and after drafting.
Technique 1: The Specificity Pass
After drafting the article, scan every paragraph. If any paragraph lacks a proper noun (name), a number (date/percentage/amount), or a direct quote, tag it as WEAK. Then either:
- Research and add a specific reference
- Or remove the paragraph entirely
Technique 2: Cross-Domain Connection
Find ONE unexpected connection to another domain:
- If writing about psychology, connect it to programming (e.g., "This is like off-by-one errors in code")
- If writing about business, connect it to sports or nature
- If writing about technology, connect it to ancient history
The cross-domain connection is often what makes an article memorable and shareable.
Technique 3: The Opposite Angle
For each section, ask: "What would someone argue against this?" Then address that counter-argument. This signals depth because it shows you've considered the full picture, not just your own view.
Place the counter-argument either:
- Within the relevant section ("You might think X, but that misses Y.")
- Or as a dedicated section before the close.
Technique 4: Multi-Level Language
Mix three levels of language within the article:
| Level | Use | Example |
|---|---|---|
| Academic | When citing research | "Kahneman and Tversky (1979) demonstrated..." |
| Conversational | When telling stories | "Sound familiar?" |
| Provocative | When making the point | "Here's the uncomfortable truth: you're not as good at this as you think." |
Variety in register signals author sophistication.
Technique 5: The "So What?" Audit
After each section, literally ask: "So what?" If the answer is not obvious from the text itself, rewrite or remove that section.
Platform-Specific Adaptations
知乎
- First-person stories welcome
- 成语/典故 add credibility for Chinese audiences
- Reference Chinese internet culture
- Close: "关注我,了解更多..." + discussion question
- Length: 3000-8000 characters
头条/百家号
- Paragraphs: 1-3 sentences max
- Emotional hooks front-loaded
- Numbers in titles: "3个..." "5种方法..."
- Image suggestions embedded
- Close: Strong CTA + follow reminder
- Length: 1500-4000 characters
Blog/Newsletter
- Footnotes and references welcome
- Subheadings for scannability
- Personal voice
- Length: 1500-4000 words
Twitter/X Thread
- Each "paragraph" = 1 tweet (≤280 chars)
- Numbered: 1/10, 2/10...
- First tweet = self-contained hook
- Last tweet = summar + share prompt
- 1-2 relevant hashtags max
- Conversational, starts with personal experience
- Emoji for emotional tone (not decoration)
- Question at end to spark comments
- Reaction prompts: 👍 / 💬 / 🔔
- Length: 500-1500 words
Title Crafting
Generate 3-5 title options. For each, explain WHY it works:
Effective patterns:
- Curiosity gap: "Why [common thing] is actually [counterintuitive truth]"
- Direct address: "If you're a [role], stop doing [thing]"
- Number + benefit: "5 ways to [achieve goal] without [pain point]"
- Challenge: "Everything you know about [topic] is wrong"
- Question: "Why does [problem] keep happening to you?"
Criteria: Click-worthy + Accurate + Platform-appropriate
Quality Checklist
Before outputting the article, verify:
- Phase 0 completed: at least 2 specific studies/experiments cited
- Every paragraph has a specific (name, number, date, or quote)
- At least one cross-domain connection found
- Counter-argument addressed somewhere
- Language varies (academic ↔ conversational ↔ provocative)
- First paragraph is a concrete scene or specific stat, NOT "In this article"
- Last paragraph circles back to the hook
- Platform-specific elements added
If any box is unchecked, fix it before delivering the article.
When NOT to use this skill
- User only wants a summary or outline
- Technical documentation (API docs, READMEs)
- User says "keep it short" or "just the facts"
- Formal academic papers or legal documents
- User has supplied no topic and no direction
