The AI tools I use to write hooks that stop the scroll

I’ve published videos with good content and terrible openings, and they died. I’ve published thin content with a great first line and watched it travel further than it deserved. The hook is not the most important part of your work, but it is the part that decides whether the rest gets seen, which is a hard thing to accept and an expensive thing to ignore.

Writing twenty hooks by hand is miserable. Writing twenty hooks with AI takes ninety seconds, and it’s changed how I open everything.

Why AI is good at this specific job

AI writing tools are mediocre at producing your best hook and excellent at producing volume. That distinction is the whole method.

Left alone, a model writes a competent, slightly generic opener. It’s fine. It’s never the best one. But ask for twenty variations across different angles and something useful happens: you stop trying to invent a hook and start recognising one. Recognition is a much easier cognitive task than generation, and humans are far better at it.

The good hook is usually option fourteen, and it’s usually one you’d never have arrived at cold.

The prompt structure that actually works

Three things separate a useful hook session from generic slop.

Give it your winners. Paste your five best-performing openings from real data. This single step does more than everything else combined, because it anchors the model to your voice and your audience rather than the average of the internet.

Give it the constraint. “This is a 45-second vertical video about X. The hook is spoken, must work in the first two seconds, and must be literally true.” Constraints produce better output than freedom, every time.

Ask for angles, not variations. Twenty rewrites of the same idea are useless. Ask for twenty different approaches: the contrarian take, the specific number, the mistake, the before/after, the question, the confession. Now you’re exploring rather than paraphrasing.

I run most of this in Jasper when I’m working on marketing copy, because the brand voice feature means I don’t have to re-paste my examples every session. For one-off video hooks a general chat model is honestly fine. My Jasper review covers whether the subscription makes sense, and Jasper vs Copy.ai if you’re deciding between the dedicated writers.

What makes a hook actually work

Patterns I’ve watched succeed repeatedly, across formats:

• Specific beats clever. “I lost £4,000 on this” beats “my biggest business mistake”. Numbers and concrete nouns do the work.
• Slight discomfort. The best hooks admit something. Competence is boring, and nobody stops scrolling for a person who has it all figured out.
• Start mid-action. Cut the throat-clearing. “So today I want to talk about” is where attention goes to die.
• Tension you’ll actually resolve. Open a loop, close it in the video. Open a loop and don’t close it and you’ll never see that viewer again.
• The unexpected qualifier. “This works, but only if you’re under 10k followers” is more interesting than “this works”.

What consistently fails: rhetorical questions everyone answers “no” to, “in this video I’ll show you”, and anything that sounds like it was written for an algorithm rather than a person.

Rewriting per platform

Once you have the angle, the format work is pure tedium, which makes it ideal to hand off.

• Short-form video. Spoken, two seconds, plus on-screen text that says something different from the audio rather than duplicating it.
• Newsletter subject. Under 40 characters or it truncates on mobile. The preview text is a second hook, and almost nobody uses it properly.
• LinkedIn / X. Everything before the “see more” fold. Write it, then count the characters, then cut it again.
• YouTube title and thumbnail. These are one unit and must not repeat each other. Title says what, thumbnail shows the tension.

I’ll take one strong angle and ask for all four formats in a single pass. The AI tools I use to grow on LinkedIn goes deeper on that platform specifically, and the best AI tools for writing YouTube scripts covers the long-form side where the hook has to survive a whole intro.

Test, don’t admire

Hooks are the one part of content where you get fast, clean feedback. Retention graphs and open rates tell you within a day whether the opening worked.

Keep a running document of every hook you’ve used and how it performed. Two things happen. You build the example bank that makes your prompts better, and you start noticing your own patterns, which are usually narrower than you think. Mine, embarrassingly, was that I opened almost everything with a confession about a mistake. It worked, and then it stopped working, because I’d trained my own audience to expect it.

The archive is also where you find out that your instinct is wrong roughly a third of the time. The hook I’m most pleased with is frequently not the one that performs.

My take

Don’t use AI to write your hooks. Use it to write twenty candidates so you can pick one, then rewrite that one in your own words. The rewrite matters, because the model’s phrasing is always half a degree off, and half a degree is the difference between sounding like you and sounding like content.

The bigger unlock isn’t the tool at all, it’s the habit of treating the opening as a separate piece of work rather than the first thing you happen to say. I now write the hook last, after the piece exists and I know what it actually delivers. That way the hook is a promise I’ve already kept rather than one I’m hoping to.

If you want the tool that makes the volume part painless, Jasper with a properly configured brand voice is the one I keep paying for. But a free chat model with five of your own winners pasted in will get you most of the way there, and the examples matter far more than the software. For the wider writing stack, best AI writing tools for creators who hate the blank page is the fuller rundown.