# What 77 breakout videos have in common

**Part of The Viral Script System** · by Dan Shipped (@danshipped) · Last verified: October 2026

My radar (Shipyard OS) watches the channels of 27 AI and business creators. Every time a video does far more views than that creator usually gets, it gets flagged as a **breakout**. Between **July 5 and October 1, 2026** it flagged **77**. Together they pulled about **33.5 million views**.

The smallest breakout did **2.5×** its creator's usual views. The biggest did **12.8×**. The median breakout did **3.3×**. Half of them (50) cleared **3×**, and 16 cleared **5×**.

I sorted every title into a formula and asked one question: *which formulas show up most at the top?*

## Read this first (what this data can and can't tell you)

Deadly sincerity, before the patterns:

- **These are YouTube titles, not Reels or TikToks.** A title does the same job as a spoken hook (stop the scroll, open a gap), so the patterns transfer. The exact words don't. I translate each formula into a short-form hook below.
- **There's no control group.** I only have the winners, not the flops. So I can say "this formula shows up a lot among breakouts and has a high median multiplier". I can't say "this formula causes 4×". Treat it as strong hints, not laws.
- **Small groups.** Some formulas have 6–8 examples. One huge video can move a group's average, so I use the **median** (the middle video), which one outlier can't drag up.
- **Channel effect.** Creators with authority (big names, a known brand) get away with titles a newcomer can't. Keep that in mind on formula 3.
- **Model and product names** in the titles (GPT-6 Astra, Jev, Fable 5.1, etc.) are quoted from the titles as published. I'm not vouching for any claim inside them.

---

## The 6 findings that matter

### 1. Short wins. Almost always.
**75 of 77** titles are **12 words or fewer**. The median is **9 words**. The 16 biggest breakouts (5×+) have the same median: 9.

→ *Rule:* your spoken hook gets ≤12 words. Your on-screen text gets 3–6.

### 2. Numbers are everywhere.
**47 of 77** (61%) contain a number: hours, dollars, a count, a version, a year.

→ *Rule:* one real number per hook. "100 hours", "$18K/month", "15 real use cases", "24 minutes".

### 3. "Copy the expert" and "who you become" have the highest medians.

| Formula | Videos | Median × | Hit 5×+ | Best example |
|---|---|---|---|---|
| **Borrowed authority** ("X reveals his workflow", "just copy him") | 6 | **4.85×** | 3 of 6 | *Matt Pocock's Agentic Engineering Workflow (just copy him)* — 7.9× |
| **Identity / who you become** | 6 | **4.65×** | 3 of 6 | *How To Become Dangerously Self-Educated (with AI)* — 6.0× |
| It's here (arrival) | 7 | 4.3× | 2 of 7 | *This New AI Could Be the Biggest Breakthrough Since ChatGPT* — 5.1× |
| Short verdict on the new thing | 10 | 4.0× | 2 of 10 | *GPT 6 Astra, so good even OpenAI are worried* — 8.6× |
| Ranged list / levels | 6 | 3.05× | 2 of 6 | *GPT-6 Astra: 20 Real Examples From Useful to Almost Impossible* — **12.8×** |
| Compressed learning (time-box) | 6 | 3.2× | 1 of 6 | *1000+ hours of Learning Claude in 15 Minutes* — 6.5× |
| Hands-on effort proof | 8 | 2.95× | 1 of 8 | *I Spent 100 Hours Using GPT-6 Astra* — 8.4× |
| Money result | 7 | 3.1× | 1 of 7 | *This Insanely Simple App Makes $50K/Month* — 5.0× |
| Stakes / warning | 8 | 3.25× | **0 of 8** | *GPT-6 Escaped. This Is Worse Than You Think* — 4.2× |
| Comparison | 3 | 2.9× | 1 of 3 | *ChatGPT vs. Claude: Which Should You Use in 2026?* — 5.2× |
| Contrarian | 2 | 3.15× | 0 of 2 | *FORGET Loop Engineering. Agentic Engineering is about THIS* — 3.6× |
| **Plain how-to / guide** | 8 | **2.75×** (lowest) | **0 of 8** | *Build Anything with Jev, Here's How* — 4.2× |

Why this fits Schwartz: in a crowded market (and "AI tools" is very crowded), a plain promise ("how to use X") is stage 1–2 sophistication. It's been said a thousand times. **Authority** gives a new *reason to believe* (someone credible does it), and **identity** speaks to who the viewer wants to be. Both are stage 4–5 answers.

### 4. "How to" alone is the weakest formula.
The 8 plain guides/tutorials have the **lowest median (2.75×)** and **zero** reached 5×. They still broke out (they're in this list), but they never broke out *big*.

→ *Rule:* never let "how to [use tool]" be the whole hook. Add a who (authority), a who-you-become (identity), a number (effort/time), or a range.

### 5. Fear gets a click, not a breakout.
The 8 stakes/warning titles ("…is getting ugly", "The End Is Near…", "Worse Than You Think") have a decent median (3.25×) but **none** reached 5×.

→ *Rule:* a warning hook works best as a door to a *solution* you show right away (Schwartz: attack only if you show the alternative). Pure doom doesn't build follows.

### 6. Ranges and effort numbers produce the biggest single spikes.
The #1 breakout (12.8×) is a **ranged list**: "20 Real Examples **From Useful to Almost Impossible**". The #3 (8.4×) is **effort proof**: "I Spent **100 Hours** Using…". A range promises a ladder (the viewer wants to see the top rung). An effort number says "I did the boring part for you".

→ *Rule:* if you make a list, make it a ladder ("from easy to almost impossible", "level 1 to level 10"). If you tested something, say how much.

---

## Two places the data argues with my rules (and what I do)

**1. Intensity words.** 16 titles use words like *insane, freak, ridiculous, incredible, so good*, and 6 of them are in the 5×+ group. On YouTube that works because it's a **verdict** from a creator the viewer already trusts, about news the viewer already cares about.
On a short video to a cold stranger, the same word reads like an ad (Schwartz: adjectives the reader notices lower belief). **What I do:** replace the adjective with the evidence. Not "this AI is insane" → "it wrote 2,000 lines and none of them broke." If you must use one, use one, and prove it in the next 2 seconds.

**2. Tool name first.** 36 titles open with a product or model name. That breaks my "start from the desire" rule, except it doesn't: when a model launches, the whole market becomes **product-aware overnight**. People already know the name and are looking for a take. **What I do:** tool-first hooks only on news days (first 72 hours of a big launch) or for followers. Every other day, desire or problem first.

---

## The formulas, translated to short-form hooks

Each one: the YouTube pattern → a template you can say in under 4 seconds → an example. (Numbers in examples are placeholders. Use your real ones.)

| # | Formula | Short-form template | Example |
|---|---|---|---|
| 1 | Borrowed authority | How [EXPERT ROLE] [DOES RESULT]. Just copy this. | How senior engineers review AI code. Just copy this. |
| 2 | Borrowed authority | [EXPERT] showed me [PROCESS]. It's [SURPRISING TRAIT]. | A 20-year copywriter showed me her hook process. It's 3 questions. |
| 3 | Identity | How to become the [IDENTITY] who [RESULT]. | How to become the freelancer who never chases invoices. |
| 4 | Identity | You're not behind. Yet. Here's [FIRST STEP]. | You're not behind on AI. Yet. Here's your first agent. |
| 5 | Identity | Once you [MILESTONE], [ACTION] these [NUMBER] things. | Once you land your first client, automate these 3 things. |
| 6 | Ranged list | [NUMBER] [THINGS], from [EASY] to almost impossible. | 7 AI agent uses, from easy to almost impossible. |
| 7 | Levels | [NUMBER] levels of [SKILL]. Most people stop at [N]. | 5 levels of prompting. Most people stop at 2. |
| 8 | Compressed learning | [BIG NUMBER] hours of [SKILL] in [SMALL] seconds. | 1,000 hours of using Claude, in 40 seconds. |
| 9 | Effort proof | I spent [TIME] [TESTING/USING] [THING]. Here's the verdict. | I spent 100 hours testing AI editors. Here's the verdict. |
| 10 | Effort proof | My [WORKFLOW] after [BIG NUMBER] [SESSIONS]. | My content workflow after 400 videos. |
| 11 | Money result | This simple [THING] makes [REAL $]/month. Here's how. | This simple Notion template makes $3K a month. Here's how. |
| 12 | Price swap | [EXPENSIVE THING] for $0. Here's the catch. | My $20-a-month AI setup, for $0. Here's the catch. |
| 13 | Comparison | [A] vs [B]: which should you use in [YEAR]? | ChatGPT vs Claude: which should you use in 2026? |
| 14 | Verdict (news days) | [NEW THING] just [DID WHAT]. Here's what it means for you. | That new model just shipped. Here's what it means for you. |
| 15 | Contrarian | Forget [POPULAR THING]. [REAL THING] is about this. | Forget prompt tricks. Good AI output is about this. |

Use the hook scorer (`tools/hook-scorer.html`) on your version. Anything under 70, rewrite.

---

## How to rebuild this list for your niche (15 minutes a week)

1. Pick 10–20 creators in your niche (YouTube is the easiest place to get numbers).
2. For each one, look at their last ~15 videos and write down the **middle** view count (sort the numbers, take the middle). That's their "usual".
3. Any video at **3× their usual or more** goes in your swipe file with its multiplier.
4. Every Friday, tag each title with a formula from the table above. After 4 weeks, you'll see which formulas your niche rewards.

The full manual method (YouTube, TikTok and Instagram) is in the README, chapter 10: "The outlier method".

---

## Appendix: all 77 titles, by formula

Multiplier = the video's views ÷ the creator's usual (median) views, as computed by the Shipyard OS radar. Views are rounded as captured.


### 1 · Borrowed authority (6)

| × | Title | Channel | Views |
|---|---|---|---|
| 7.9× | Matt Pocock’s Agentic Engineering Workflow (just copy him) | David Ondrej | 450,000 |
| 5.8× | Ex-NASA dev reveals his Agentic Engineering Workflow | David Ondrej | 333,000 |
| 5.6× | Google Just Dropped a Masterclass on Agentic Engineering (It's SO Good) | Cole Medin | 163,000 |
| 4.1× | Pi Agent dev reveals his Agentic Engineering Workflow | David Ondrej | 236,000 |
| 2.8× | L8 Principal's Agentic Engineering Setup (just copy him) | David Ondrej | 159,000 |
| 2.6× | Anthropic Engineer Explains: What to Build Instead of AI Agents | Nate Herk / AI Automation | 349,000 |

### 2 · Identity / who you become (6)

| × | Title | Channel | Views |
|---|---|---|---|
| 7.7× | Install DLSS 5 EARLY & your Games will Look DISTURBINGLY Real. | MattVidPro | 120,000 |
| 6.0× | How To Become Dangerously Self-Educated (with AI) | Dan Martell | 1,200,000 |
| 5.7× | GPT-6 Astra will Make you Think BIGGER. | MattVidPro | 88,000 |
| 3.6× | You’re Not Behind (Yet): How to Build Your First AI Agent (Full Guide) | Dan Martell | 709,000 |
| 2.8× | If you use AI, switch to Omarchy immediately | Alex Finn | 243,000 |
| 2.5× | Use Claude Better Than 99% of People (Beginner to Pro) | Futurepedia | 253,000 |

### 3 · Short verdict on the new thing (10)

| × | Title | Channel | Views |
|---|---|---|---|
| 8.6× | GPT 6 Astra, so good even OpenAI are worried | AI Explained | 853,000 |
| 6.5× | GPT 6 Astra is a freak | AI Search | 1,500,000 |
| 4.8× | GPT-6 Astra Is Finally Here (And It’s REALLY Good) | Matt Wolfe | 500,000 |
| 4.4× | Googles New Gemini 4 Argon is Now The Worlds Smartest AI | TheAIGRID | 109,000 |
| 4.2× | Jev + Treg is a crazy combo for automation... | AI Jason | 146,000 |
| 3.8× | Claude Opus 5.5 is ridiculous | AI Search | 886,000 |
| 3.7× | Meta Muse is an INCREDIBLE AI agent | Alex Finn | 324,000 |
| 3.4× | Deepseek just did the impossible | AI Search | 787,000 |
| 3.0× | Claude Opus 5 is a freak | AI Search | 691,000 |
| 2.9× | I Didn't Expect YuE2 to Sound THIS Good | MattVidPro | 45,000 |

### 4 · It's here (arrival) (7)

| × | Title | Channel | Views |
|---|---|---|---|
| 5.1× | This New AI Could Be the Biggest Breakthrough Since ChatGPT - Jev | TheAIGRID | 128,000 |
| 5.0× | Jev is HERE. How to use it | Greg Isenberg | 744,000 |
| 4.3× | Claude Can Do So Much More For You Now... | The AI Advantage | 132,000 |
| 4.3× | The Biggest ChatGPT Update Yet: Meet dots | Skill Leap AI | 166,000 |
| 3.7× | Finally, an Open Standard for the Karpathy LLM Wiki is HERE | Cole Medin | 108,000 |
| 3.3× | ASTRA IS HERE (GPT-6 RELEASED) | Matthew Berman | 351,000 |
| 2.8× | OpenAI Just Revealed Dots… This Changes ChatGPT Forever | TheAIGRID | 69,000 |

### 5 · Ranged list / levels (6)

| × | Title | Channel | Views |
|---|---|---|---|
| 12.8× | GPT-6 Astra: 20 Real Examples From Useful to Almost Impossible | The AI Advantage | 389,000 |
| 7.0× | Once You Make Money, Upgrade These 7 Things ASAP | Dan Martell | 1,400,000 |
| 3.2× | 10 Levels of Jev For Agentic Engineers | IndyDevDan | 114,000 |
| 2.9× | 28 Insane Things Astra Can Do (2 Hour Course) | Riley Brown | 259,000 |
| 2.6× | 11 Tiny Coding Agent Fixes With A Stupid Amount Of Payoff | Cole Medin | 76,000 |
| 2.5× | 14 Insane Things ChatGPT Work Can Do (Automate Anything) | Riley Brown | 225,000 |

### 6 · Compressed learning (time-box) (6)

| × | Title | Channel | Views |
|---|---|---|---|
| 6.5× | 1000+ hours of Learning Claude in 15 Minutes (Beginner to Pro) | Dan Martell | 1,300,000 |
| 3.5× | Obsidian in 24 Minutes | Tina Huang | 706,000 |
| 3.3× | Every Size Local AI In 24 Minutes | Tina Huang | 679,000 |
| 3.1× | How to Use New ChatGPT Work in 12 Minutes | The AI Advantage | 94,000 |
| 3.0× | Hermes Agent Fundamentals In 29 Minutes | Tina Huang | 604,000 |
| 2.6× | Learn 10X Faster In 10 Minutes | Tina Huang | 532,000 |

### 7 · Hands-on effort proof (8)

| × | Title | Channel | Views |
|---|---|---|---|
| 8.4× | I Spent 100 Hours Using GPT-6 Astra (This Feels Like AGI) | Riley Brown | 745,000 |
| 3.8× | I Turned GPT-6 Astra Into a 24/7 Stock Trader (tutorial) | Nate Herk / AI Automation | 519,000 |
| 3.3× | GPT-6 Astra Made This Entire Video | Nate Herk / AI Automation | 456,000 |
| 3.0× | Jev (Fully Tested) + Browser Use: FASTEST AI Agent I'VE TRIED YET! | AICodeKing | 40,000 |
| 2.9× | I Tested GPT-6 Astra vs Fable 5.1 on 15 Real Use Cases | Nate Herk / AI Automation | 395,000 |
| 2.8× | My Agentic Engineering Workflow (after 6,775 sessions) | David Ondrej | 161,000 |
| 2.7× | ChatGPT Dots: How I Spent My First 24 Hours | The AI Advantage | 81,000 |
| 2.6× | I Tested Jev on 12 Real Use Cases. My Honest Thoughts. | Nate Herk / AI Automation | 354,000 |

### 8 · Money result (7)

| × | Title | Channel | Views |
|---|---|---|---|
| 5.0× | This Insanely Simple App Makes $50K/Month | Starter Story | 385,000 |
| 3.6× | How This SaaS Hit $69K/Month In Just 2 Months | Starter Story | 272,000 |
| 3.6× | Claude Design Now Builds Beautiful $10,000 Websites (NO AI Slop) | Jack Roberts | 243,000 |
| 3.1× | Start a $10,000/mo Solo AI Creative Agency (Higgsfield + Claude) | Liam Ottley | 129,000 |
| 3.0× | DeepSeek Harness = Claude Code for $0 | Jack Roberts | 204,000 |
| 2.9× | Build a $5,000 AI Datacenter at Home, Here’s How | David Ondrej | 164,000 |
| 2.9× | This Simple App Makes $18K/Month | Starter Story | 220,000 |

### 9 · Stakes / warning (8)

| × | Title | Channel | Views |
|---|---|---|---|
| 4.2× | GPT-6 Escaped. This Is Worse Than You Think | TheAIGRID | 106,000 |
| 3.7× | We need to talk about Jev... | Matthew Berman | 397,000 |
| 3.5× | Did OpenAI actually build AGI? GPT-6 Astra first look | Fireship | 4,200,000 |
| 3.4× | The most important video you'll ever watch in the age of AI | Nick Saraev | 213,000 |
| 3.1× | The End Is Near for Claude Code AI Operating Systems... | Liam Ottley | 127,000 |
| 2.9× | Opus 5.5: How Close Are We to Automated AI Research? | AI Explained | 286,000 |
| 2.8× | OpenAI's biggest math breakthrough is getting ugly... | Fireship | 3,400,000 |
| 2.6× | Claude JUST found hidden DNA... | Wes Roth | 331,000 |

### 10 · Comparison (3)

| × | Title | Channel | Views |
|---|---|---|---|
| 5.2× | ChatGPT vs. Claude: Which Should You Use in 2026? | The AI Advantage | 160,000 |
| 2.9× | GPT-6 Astra (Fully Tested & Side by Side comparison with Fable 5.1): ONE is a CLEAR WINNER! | AICodeKing | 39,000 |
| 2.9× | Hands on: Fable 5 makes GPT 5.5 feel like a "toy" | MattVidPro | 45,000 |

### 11 · Contrarian (2)

| × | Title | Channel | Views |
|---|---|---|---|
| 3.6× | FORGET Loop Engineering. Agentic Engineering is about THIS | IndyDevDan | 130,000 |
| 2.7× | I'm Obsessed With Local AI. Here's Why | Greg Isenberg | 401,000 |

### 12 · Plain how-to / guide (8)

| × | Title | Channel | Views |
|---|---|---|---|
| 4.2× | Build Anything with Jev, Here’s How | David Ondrej | 240,000 |
| 3.8× | Building a GPT-6 Kalshi AI Trading Bot From Scratch (full guide) | All About AI | 23,000 |
| 3.1× | Ultimate Guide To Claude Skills | Skill Leap AI | 118,000 |
| 2.8× | Ultimate Guide To ChatGPT Codex for Everyday People | Skill Leap AI | 106,000 |
| 2.7× | How to Use ChatGPT Sites to Build Anything (Insane) | Riley Brown | 243,000 |
| 2.7× | Build Anything with Kimi K3, Here’s How | David Ondrej | 152,000 |
| 2.7× | Agentic AI Trading For Beginners: A New Money Making Era Is Here | All About AI | 16,000 |
| 2.5× | CLAUDE CODE MARKETING FULL COURSE (6 HOURS) | Nick Saraev | 158,000 |

---
Made by Dan Shipped — @danshipped