The 11 Hooks That Stop the Scroll

What you'll be able to do after this lesson: name all 11 hook types, write a hook in any of them from a fill-in template, and use AI (ChatGPT or Claude) to generate and A/B-rewrite hooks for your niche — so you never again post a video that dies in the first 3 seconds.

Watch the class

Primary video: How I Use ChatGPT Projects to Write TikTok Hooks in Seconds — YouTube watch?v=589opy7VZTI · duration TBD [PENDING EMBED VERIFICATION]

Fallback video: Sabrina Ramonov — How do I make awesome ChatGPT prompts? — YouTube watch?v=b7UUcbKlTLM [PENDING EMBED VERIFICATION]

What the video teaches: how to set up a reusable ChatGPT workspace loaded with your brand context so it writes hooks in your voice on demand.

The text is a complete standalone manual; the video just shows the workflow live.

The concept: why the first 3 seconds decide everything

As a working benchmark, most viral short-form videos land their hook inside the first 3 seconds, and strong hooks run about 10-14 words (benchmark — Verified: Jul 2026; a hypothesis to test against your own numbers, not a law).

A hook is not clickbait. Clickbait promises what the video won't deliver, and platforms punish the drop-off. A hook is an honest promise made fast: it tells the right viewer "this is for you, stay."

A hook works on two channels: the verbal hook (the first sentence you speak or caption line — this lesson's focus, the part AI drafts in seconds) and the visual hook (the first frame: movement, an unexpected image, on-screen text, a result shown early). The strongest videos fire both, so pair every template below with a matching visual. And rotate: creators who vary 5-10 hook types keep their audience guessing; lean on one and you become predictable, and predictable gets scrolled.

The 11 hook types

For every type: why it works, a fill-in template with [VARIABLES], and an example for Maya, a home-fitness creator teaching busy beginners to train at home.

1. Direct question — a question opens a mental loop the brain wants to close. Why does [FRUSTRATING RESULT] happen even when you [THING THEY DO RIGHT]? Maya: "Why aren't your abs showing when you work out five days a week?"

2. Bold / contrarian claim — contradicting a held belief creates friction; they stay to defend or update it. [POPULAR ADVICE] is actually [KEEPING YOU FROM / RUINING] [GOAL]. Maya: "Daily cardio is the reason you're not losing fat."

3. Contrast / before-after — the brain is wired to notice change; a then/now gap begs the question what happened between? [TIME AGO] I was [BEFORE STATE]. Today I [AFTER STATE]. Here's what changed. Maya: "A year ago I couldn't do one push-up. This morning I did fifty. Here's what changed."

4. Curiosity gap / open loop — you reveal valuable info exists but withhold the detail; the gap is uncomfortable. There's one [MISTAKE / HABIT / SETTING] that's quietly killing your [GOAL] — and almost nobody checks it. Maya: "There's one form mistake quietly wrecking your squats — and almost nobody checks it."

5. Confession — admitting a flaw is disarming; vulnerability reads as honesty. I [EMBARRASSING TRUTH ABOUT YOUR FIELD] for [TIME PERIOD] — until [TURNING POINT]. Maya: "I taught spin classes for two years while secretly hating cardio — until I found this."

6. Negativity / warning — loss aversion: "stop doing X" outpulls "start doing Y" because it threatens something they have. Stop [COMMON PRACTICE] — it's [DAMAGING CONSEQUENCE] without you noticing. Maya: "Stop stretching before you lift — it's sapping your strength without you noticing."

7. Listicle — a number sets the size of the commitment up front; "5 things" feels finishable. [NUMBER] [THINGS] that [DESIRED OUTCOME] — number [X] surprised me. Maya: "Five snacks that build muscle while you sleep — number three surprised me."

8. POV / relatable — naming a hyper-specific shared moment triggers instant "that's me" recognition. POV: you [SPECIFIC RELATABLE SITUATION YOUR AUDIENCE LIVES]. Maya: "POV: it's day three of your new workout plan and the couch is negotiating."

9. "Stop scrolling" / pattern interrupt — a direct command or odd first frame breaks the feed's rhythm for one beat. Stop scrolling if you [SPECIFIC AUDIENCE + SPECIFIC DESIRE]. Maya: "Stop scrolling if you want visible arms by September without a gym."

10. Identity call — people respond to labels they claim; ignoring the video feels like ignoring themselves. If you're a [IDENTITY], this [THING] will change how you [ACTIVITY]. Maya: "If you're a busy mom, this 12-minute routine will change how you train."

11. Results-first — show the payoff before asking for attention; proof up front removes the "will this be worth it?" doubt. Top-performing type in OpusClip's hook testing. This is [THE RESULT, SHOWN OR STATED] — here's exactly how it happened in [TIMEFRAME]. Maya: "This is me holding a 3-minute plank — here's exactly how I got there in six weeks."

How to write hooks with AI, step by step

1. Pick the idea first. One video = one idea, one plain sentence. Hooks amplify an idea; they can't rescue a missing one.
2. Pick 3 hook types — deliberately different ones (say, contrarian + curiosity gap + results-first). Choosing before prompting keeps the AI off generic questions.
3. Load your brand context with Prompt 1 (or a saved Project), then generate 10 options. The first three are usually obvious; gold tends to appear around option six.
4. Cut with judgment. Kill anything you wouldn't say to a friend, anything over ~14 words, anything that overpromises. You're the editor; the AI is the intern.
5. Rewrite the survivors with Prompt 2 for A/B variants of your best 2-3.
6. Say each finalist out loud and time it — over ~3 seconds or you stumble, shorten it — then pair it with a visual (movement, the result, bold text). Verbal + visual, always.

Prompts (copy and use)

Prompt 1 — Brand-Context Hook Generator

Purpose: turn one video idea into 10 hook options across the types you choose, in your voice for your audience.

When to use it: every time you script a video, after you've picked the idea and 3 hook types.

Editable variables: everything in [BRACKETS] — fill in all seven before sending.

You are my short-form hook writer. First, my brand context:

- Niche: [YOUR NICHE, e.g. "home fitness for busy beginners"]
- Audience: [WHO THEY ARE + THEIR #1 STRUGGLE, e.g. "women 25-45, no time, intimidated by gyms"]
- Voice: [3 ADJECTIVES, e.g. "warm, no-nonsense, lightly funny"]
- Never say: [WORDS/CLAIMS TO AVOID, e.g. "guaranteed, shredded, secret hack"]

My video idea (one sentence): [YOUR IDEA]

Write 10 hooks for this idea, using only these hook types: [TYPE 1], [TYPE 2], [TYPE 3].
Rules:
- Maximum 14 words each.
- Spoken language, not ad copy. I must be able to say it naturally on camera.
- Honest: promise only what the idea above actually delivers.
- Label each hook with its type.
- After the list, mark the 2 you'd bet on and say why in one line each.

Warnings: the model will happily overpromise — keep the "honest" rule and your Never say list. Never publish an AI hook you haven't said out loud; if it sounds like a billboard, rewrite it.

Prompt 2 — Hook A/B Rewriter

Purpose: take your best hook and produce disciplined variants so you can test versions instead of guessing.

When to use it: after you've cut Prompt 1's output down to 2-3 survivors, or to re-hook an old video that underperformed.

Editable variables: the two [BRACKETS] fields.

Here is a hook I like for a short-form video:

"[YOUR HOOK]"

The video delivers: [WHAT THE VIDEO ACTUALLY SHOWS/PROVES, one sentence]

Rewrite it 6 ways:
1. Shorter (under 8 words).
2. As a direct question.
3. As a results-first version.
4. As a negativity/warning version.
5. With a number in it.
6. Wildcard — surprise me, any type.

Keep every version under 14 words, sayable out loud, and honest to what the
video delivers. Then tell me which version you'd test first against the
original, and what the on-screen visual should be for that version.

Warnings: don't test six variants at once — pick original vs. one challenger per video, or you'll learn nothing. Keep the delivery line truthful: the rewriter can only stay honest if you tell it what the video really proves.

Worked example: one idea, three doors in

Maya's idea: "12-minute strength beats 45-minute cardio for busy people."

• Bold claim: "Your 45-minute cardio habit is why you're still exhausted." — attacks a held belief; instant friction.
• Results-first: "I dropped cardio, lifted 12 minutes a day — these are my six-week results." — proof up front.
• Identity call: "Busy moms: 12 minutes beats your 45-minute cardio. Here's the swap." — names the viewer, promises a concrete takeaway.

Same idea, three different doors in. The idea stays honest and constant; the entry point rotates.

Common mistakes and how to fix them

Exercise: three doors, one idea

1. Take one real video idea from your niche — one sentence.
2. Pick 3 different hook types from the taxonomy.
3. Use Prompt 1 to generate options, then cut and rewrite until you have one finished hook per type — your words, said out loud, timed.
4. Score each hook with the rubric (self-score, 1-5 per row):

Expected result: three hooks of three different types for the same idea, each 14 words or fewer, each scoring 20+ out of 25. If one scores under 15, run it through Prompt 2 and re-score.

Checklist before you move on

• I can name the 11 hook types without looking.
• I know the 3-second / 10-14-word benchmarks — and that they're benchmarks to test, not laws.
• I have Prompt 1 saved with my brand context filled in (ideally as a Project).
• I wrote 3 finished hooks of 3 different types for one real idea.
• Every hook survived being said out loud with a timer.
• Each hook has a planned visual for the first frame.
• Nothing I wrote promises more than the video will deliver.

Toolkit for this lesson

(Free tier on both — Verified: Jul 2026.) Reuse Lesson 1.1's Brand Master Prompt as the context block in Prompt 1.

Next step

You can now stop the scroll. Next, in Lesson 1.3 — Short-form scripts: the 4-beat structure, you'll learn what to do with the attention you just earned: Hook → Build → Payoff → CTA, so the viewer who stayed for second 3 stays to the end — and comes back.