Based on an analysis of the openings of 5 viral videos by Li Ge Builder (likes 800-3222, favorites 544-5260), optimize the opening hook of a video script.
Core belief: If you can't write a good opening, 80% of the time it's because the content itself has problems, and the other 20% is because you're speaking from your own horizon, not the audience's horizon. Before optimizing the opening, you must first pass these two gates.
Workflow
Gate 0: Content Quality Gatekeeping (takes priority over everything)
Diagnose the content itself first, then discuss the opening.
Content Completeness
| Situation | Handling |
|---|---|
| Only a topic/title, no body | Stop. "The opening is a trial sample of the content. You don't even have content, so optimizing the opening is meaningless. Write the body first." |
| Body is under 200 characters | Warning. "The body is too short, and the opening won't be able to hold it up. It's recommended to enrich the content first." You may continue, but flag the risk. |
| The body is all filler, with no substantive material | Stop. "There is no usable material, so the opening can't be made. First make the content solid." |
Material Scan
Find material in the body (meeting 2 or more items is required to continue):
- Data: big numbers/comparative numbers/percentages (e.g., 3222 likes, 14.1% completion rate, within 2 minutes)
- Story: before vs after; the bigger the contrast, the better
- Punchline: a viewpoint that can stand on its own, with a memorable point
- Authority: people/institutions (Anthropic, ByteDance, Fortune 500)
- Pain Point: the target audience's real experience of frustration
Meeting 0-1 items → Stop. "Your content has no material, so you can't write a good opening. First add data, stories, punchlines, or pain points."
Meeting 2 items → Can be optimized, but the effect will be limited. Prompt the user that it's best to add more.
Meeting 3 or more items → Continue to Step 1.
Step 1: Horizon Diagnosis (must precede template selection)
This is our most core proprietary framework, solving the hidden problems that most opening optimization tools cannot detect.
Why do some openings that are "specific enough" still make people swipe away?
Type A — Too vague: "AI has been really hot recently" → Ordinary opening optimization tools can detect and fix this
Type B — Specific but aimed at the wrong person: "I said, 'The Red Finger script is done'" → Ordinary checklists cannot detect this, and instead misjudge it as "specific enough, passed" → More hidden, more harmful
Philosophical Roots
Wittgenstein's language game: Words only have meaning within a shared "form of life." "The red finger script is done" is semantically complete within the creator's language game; for an unfamiliar audience, these symbols have no referent and cannot trigger any perception.
Gadamer's fusion of horizons: Communication is not "sending information," but two horizons meeting. If the opening starts from the creator's horizon, the audience's horizon cannot make contact with it—there is no fusion, only sliding away. An effective opening is:Start from the audience's existing horizon, then lead them toward the creator's horizon.
Kahneman's System 1: The brain completes pattern matching within 0.3 seconds. When encountering a proper noun that cannot be categorized, System 1 directly judges "this has nothing to do with me" and swipes away. Emotional hijacking must happen before understanding.
Diagnostic operation
| Creator's horizon (❌ cannot be the first sentence) | Audience's horizon (✅ starting point of the first sentence) |
|---|---|
| Project names, client code names ("Red Finger") | Universal experiences of frustration |
| Internal tool names ("li-planning system") | Widely recognized work pain points |
| Creator's personal history | Situations that "anyone who has used AI has encountered" |
| Details only insiders can understand | Scenarios anyone can relate to in 0.3 seconds |
Judgment criteria: Can a stranger who has never watched your video read the first sentence and relate it to their own experience?
- Can relate → Horizons aligned, proceed to Step 2 to choose a template
- Cannot relate → Horizons misaligned, first perform "horizon shifting"
Horizon shifting operation
Principle: The first sentence describes the audience's feelings, not the creator's actions.
❌ Creator's horizon opening:
我说了一句"红手指脚本写完了"，AI没有问我改哪里……
✅ Audience's horizon opening:
用 AI 管理任务，最烦的一件事是——你说完，它还要问你"确认一下，你要我改哪里？"
Then bring out the creator's specific situation as a "corroborating case":
上周我遇到了一个例外：我说了一句"脚本写完了"，它没问，直接同时更新了我的待办清单和商单追踪表。
The order is:Audience's universal pain point → Creator's specific case, not the reverse.
Step 2: Diagnose the current opening (if any)
If the user has provided an opening, diagnose the problems first, give a diagnostic report, then generate solutions:
| Diagnostic dimension | Check item | Common problem |
|---|---|---|
| Horizon | Does the first sentence start from the audience's horizon? | Starts from the creator's private background, strangers cannot relate |
| Independence | Can it be understood without looking at the title? | Assumes the audience has read the title, lacks topic establishment |
| Cognitive conflict | Is there a contrast of "simple vs complex" or "common sense vs truth"? | Flat narration, no conflict |
| Suspense | Does it give the answer directly? | States the conclusion at the start, no curiosity |
| Credibility | Why listen to you? | No authority/achievements/experience established |
| Speaking-friendly | Can it be read aloud directly? | Written language, abstract concepts |
| Match | Can the body deliver what the opening promises? | Opening asks A, body discusses B |
Output:
## 当前开头诊断
**开头**：[用户的开头]
**问题**：
- 视界：[通过 / 需要前移：具体问题]
- 独立性：[通过 / 问题描述]
- 认知冲突：[通过 / 问题描述]
- 悬念：[通过 / 问题描述]
- 可信度：[通过 / 问题描述]
- 口播友好：[通过 / 问题描述]
- 匹配度：[通过 / 问题描述]
**核心问题**：[最严重的 1-2 个问题]
Step 3: Three-way parallel generation (10-15 items)
Only execute this step after passing the Level 0 content gatekeeping.
Use three paths to generate openings in parallel based on our viral templates, 3-5 items per path, 10-15 items total.
Path A: Direct application of the optimal template (material extraction)
Identify the content type, select the best-matching viral template, and extract the strongest material from the body to generate 3-5 openings.
Content type → template mapping：
- Tutorial/installation/usage type → Template 1 (user demand type)
- Opinion/principle type → Template 2 (cognitive disruption type)
- New tool/new feature → Template 3 (feature value type)
- Hands-on testing/review type → Template 4 (real case type)
- Practical resource sharing, creator style leans direct → Template 5 (straight-to-the-point type)
- Tutorial + demonstrable effect → Template 6 (effect-first type)
- Hot event/data-driven type → Template 7 (number shock type)
- Popular science/concept explanation type → Template 8 (concept question type)
Generate 3-5 items according to the template structure; the material must come from the body text, no fabrication.
Path B: Material supplement version
If the material in Path A is not impactful enough, proactively supplement usable material:
Types that can be supplemented：
- Real data or verifiable results (e.g., "I connected with 300 people and found...")
- Specific comparison scenarios
- Existing viral data of the same type as endorsement
Rule: Supplements must be real and credible. Mark supplemented openings with "⚠️ You need to confirm whether the material is real".
Generate 3-5 items.
Path C: Suspense reconstruction version
Turn the core conclusion of the body into a question, using suspense to drive viewers to finish watching:
Principles：
- Do not give the answer directly; turn the conclusion into a question
- Use "why" instead of "proved"
- Use "did you notice" instead of "the conclusion is"
Li Ge style suspense formula：
- ❌ "AI hallucination vendors deliberately don't solve it" (direct answer)
- ✅ "Vendors can clearly solve AI hallucination, why deliberately not do it?" (create suspense)
Generate 3-5 items.
Step 4: Output the plan + Top 3 recommendations
## 开头优化方案（共 X 条）
### 路径 A：最优模板应用（X 条）
| 序号 | 开头 | 说明 |
|-----|------|------|
| A1 | [开头] | [使用了哪个模板，提取了哪个素材] |
| A2 | [开头] | [...] |
### 路径 B：素材增补版（X 条）
| 序号 | 开头 | 说明 |
|-----|------|------|
| B1 | [开头] | [增补了什么 ⚠️ 需确认] |
| B2 | [开头] | [...] |
### 路径 C：悬念重构版（X 条）
| 序号 | 开头 | 说明 |
|-----|------|------|
| C1 | [开头] | [把什么结论改成了悬念] |
| C2 | [开头] | [...] |
---
## Top 3 推荐
**推荐 1**：[开头]
- 模板：[模板名称]
- 理由：[为什么推荐，对应爆款数据：如「用户需求型在点赞3222的视频上验证有效」]
**推荐 2**：[开头]
- 模板：[模板名称]
- 理由：[...]
**推荐 3**：[开头]
- 模板：[模板名称]
- 理由：[...]
Viral template library (used when generating in Step 3)
Template 1: User demand type (highest frequency, first choice for tutorial type)
Formula：[用户问题] + [颠覆性解法] + [时间承诺/降低门槛]
[上期/最近]我的[XX视频]发出后，评论区[很多人/有人]问"[具体问题]"。
其实[颠覆性方法]，[AI/工具]就能帮你[自动完成]！
今天用[X分钟]，分享给你一个[XX]的[极简/万能]方法，这个方法我个人认为是最适合[小白/XX人群]的，不只是[当前工具]，[其他场景]也能这样操作。
Core features：
- ✅ Real user demand (comment section, people often ask)
- ✅ Cognitive conflict ("just need to say two sentences" vs "complex installation")
- ✅ Specific time promise (2 minutes, 3 minutes)
- ✅ Lower the barrier (minimalist, even beginners can learn)
Validation data: "Install ClaudeCode in 2 minutes" 3222 likes, 5260 favorites
Template 2: Cognitive disruption type (high interaction, opinion/principle type)
Formula：[常识/痛点] + [颠覆性观点] + [承诺讲透]
如果你[做过XX/用过XX]，你应该会发现，[痛点描述]。
很多人问我"[常见问题]"，但这其实是个错误的问题。
真相是：[颠覆性观点]。
今天我用[X分钟]把这事儿讲清楚——不是给你讲[理论术语]，而是告诉你：[核心价值]。
Core features：
- ✅ Create cognitive conflict (common sense vs truth)
- ✅ Disruptive viewpoint ("deliberately not solve it")
- ✅ Promise to explain thoroughly (explain clearly in 3 minutes)
- ✅ Personal stance (I think, I found)
Verification data: "Understand AI Hallucinations in 3 Minutes" 748 likes, 544 favorites, 14.1% completion rate
Template 3: Functional value type (suitable for new tools/new features)
Formula：[工具/功能介绍] + [核心价值/降低门槛] + [实战承诺]
今天带大家上手一个[平台/工具]推出的[超实用/强大]功能——[功能名]。
和大多数[同类工具]不同的是，它可以让你[核心价值1]，不需要[门槛1]，也不需要[门槛2]。
这期视频就手把手带大家做一个实战案例，用[功能名]做一个[具体案例]。
Verification data: "Get Started with Agent Skills in 3 Minutes" 513 likes, 892 favorites
Template 4: Real case type (suitable for hands-on testing/reviews)
Formula：[核心能力宣布] + [过去局限] + [转折点] + [实测结果]
[工具/技术][核心能力]了！
我对[领域]的能力一直[态度]，老粉都知道。
但说实话啊，[过去的局限性]。
但最近啊，[转折点]。
今天我用[工具]做了个实测：[具体内容]。测完之后我的第一反应是：[核心发现]。
Template 5: Get straight to the point type (practical resources, direct style)
Formula：[工具/内容是什么] + [我做了什么/你能拿到什么] + [时间承诺]
When to use: The content itself has clear practical value (tools, methods, open-source resources); the creator's style is on the direct side.
今天分享[工具/内容]。我在上面做了[具体成果]——
[具体点1]；
[具体点2]。
[开源/免费]，可以直接拿去用。[X分钟]，演示一下。
Example：
Today I'm sharing Alibaba's open-source AI assistant CoPaw. I built two self-media-specific skills on it—the first takes a video link and automatically gets the verbatim transcript, used to study viral content structures; the second makes it proactively ask me about my content progress every two hours and automatically record and summarize it. Both skills are open-sourced, so you can use them directly. In 3 minutes, I'll demo it.
Pitfall avoidance guide (common to all templates)
- ❌ Don't say "XX is really popular lately"—it's vague and unappealing
- ❌ Don't fabricate user demand—"many people in the comments are asking" must be true
- ❌ Don't use an authoritative tone—"teach you" "guide"
- ❌ Don't overpromise—"you only need to say two sentences" must be true
- ❌ Don't assume the audience knows the tool—state the value first, then the tool name
Template 6: Effect-first type (tutorial + demonstrable effect)
Cross-creator verification: Snail "What exactly is Lobster for, and how do you install it?" 5,876 likes, 256,000 views
Formula：[先展示最终效果] + [观众反应/震撼点] + [回头教方法]
When to use: Tutorial-type content that has a visualizable final effect to show; first let the audience see "what it looks like when done," then teach how to do it.
等一下我先快进一下，先给大家看一眼，[工具/方法]配好之后真的在干活——
[展示效果的具体描述]。
看到效果了吧？那现在我们正式进入[教程主题]。
Core features：
- Break the usual "teach first, show later" order and work backward from the result to create appeal
- Lower the audience's patience threshold ("Let me first see if it's worth learning")
- Suitable for tool installation, configuration, and setup content
Template 7: Number shock type (hot events/data-driven)
Cross-creator verification: Chimpanzee "510,000 lines of source code leaked! The fastest 100,000-star project in GitHub history" 116,000 views; Snail "Apple just released a 4,599 yuan MacBook!" 1.7 million views
Formula：[震撼数字+事件] + [为什么这件事重要] + [接下来要讲什么]
When to use: Hot events, industry shocks, data-driven content; use numbers at the opening to create impact.
[时间点]，[事件主体][震撼数字的事件描述]。
[一句话说明为什么这件事重要/反常]。
[我要拆解/分享的内容预告]。
Example：
At dawn on March 31, Anthropic's core product Claude Code, which earns 2.5 billion dollars a year, had 510,000 lines of source code leaked. The reason was that one line was missing from the configuration file at release. But to be honest, when I first heard the news, I didn't take it seriously—but I was quickly proven wrong.
Core features：
- Numbers must be specific and shocking ("510,000 lines" rather than "a large amount")
- Time points add a sense of timeliness ("just now" "at dawn")
- Suitable for emotional-spread content with a favorite/like ratio <1.0, with the goal of driving up views
Template 8: Concept question type (popular science/concept explanation)
Cross-blogger validation: Senior Jiang's "What Is Vibe Coding, and How to Use It?" 12,606 likes; "Explaining What Token Is in 10 Minutes" follower growth rate 3.72% (the highest among all benchmark bloggers)
Formula：[概念是什么？] + [为什么你应该关心] + [时间承诺]
When to use: Popular science content, explaining a new concept or hot term; use questions to trigger the audience's need for cognitive closure.
[概念名]到底是什么？[一句话说明为什么现在要了解]。
[时间承诺]，给你讲清楚。
Example：
What Is Vibe Coding, and How to Use It?
Key features：
- Open directly with a question, triggering the brain's "search for the answer" mode
- The opening type with the highest follower growth rate (popular science + plain language, average 3.7%)
- Suitable for explanatory content about emerging concepts and hot terms
Example: OpenClaw installation tutorial opening optimization
Before optimization：
最近OpenClaw很火，评论区有人问"openclaw怎么样"、"听说openclaw很好用"。
我试了一下，确实不错。
OpenClaw可以通过skills自动化各种任务，比如自动搜索文章、管理GitHub、创建飞书文档，甚至可以接入手机远程控制。
今天分享一下怎么装。
其实不用看教程，直接让AI帮你装就行了！用3分钟，分享给你一个AI工具的极简安装法，这个方法我个人认为是最适合小白的，不只是OpenClaw，其他AI工具也能这样操作。
Problem diagnosis：
- ❌ "OpenClaw is really popular lately" — too vague
- ❌ The core method (AI installs AI) is buried too deep, placed at the end
- ❌ Assumes the audience knows OpenClaw
Top 3 after three-path generation：
Path A (Template 1):
To install AI tools, you don't need to read tutorials—just let AI install them for you. I used this method to install OpenClaw in 3 minutes. Today, in 3 minutes, I'll share with you a minimalist installation method for AI tools, most suitable for beginners. Not just OpenClaw—other AI tools can be operated this way too.
Path C (suspense version):
How long do you usually spend installing AI tools? Is it possible to let AI install AI for you? I tried it and installed OpenClaw in 3 minutes—no need to read tutorials, just let AI install it for you.
Path A (Template 5 direct version):
Today I'm sharing OpenClaw—an AI tool that can automate tasks through skills. It can search articles, manage GitHub, create Feishu documents, and even connect to a phone for remote control. AI installs it for you, 3 minutes, let me demonstrate.
Language
- Reply in Chinese if the user uses Chinese, and reply in English if the user uses English
- Chinese replies follow the "Chinese Copywriting Guidelines"