50 ad hooks in an hour: an AI workflow for testing creative faster
How to brief AI on your offer, write dozens of hooks by angle, and pick the few that are actually worth paying to test.
Most ads don’t lose in the middle. They lose in the first two seconds, before anyone hears the offer. That opening line or shot is the hook, and it decides whether the rest of your ad gets watched at all.
The problem is that good hooks come from volume. You need a lot of options to find the few that stop the scroll, and writing 50 by hand takes most teams a week. With the right brief, AI can get you there in an hour. The trick isn’t asking for “50 hooks.” It’s giving the model the same inputs a great copywriter would ask for, then making it work angle by angle.
Here’s the exact workflow I use.
Step 1: Write the brief the AI actually needs
Generic input gives you generic hooks. Before you open any AI tool, write down six things about the campaign. This takes ten minutes and does most of the heavy lifting.
- The offer in one sentence, including the price or the main deal.
- The buyer: who they are, what they do all day, and what they’ve already tried.
- Their top three pains, written in the words customers use in reviews, DMs and sales calls.
- The outcome they want, as specific as you can make it.
- Your proof: numbers, results, testimonials, years in business.
- The objections that stop people from buying.
Then paste it into a prompt like this one. Keep it saved; you’ll reuse it for every campaign.
You are a direct-response copywriter who writes hooks for short-form video ads. Offer: [one sentence] Buyer: [who they are, what they've tried] Pains, in their words: [1] [2] [3] Outcome they want: [specific result] Proof: [numbers, results, testimonials] Objections: [what stops them buying] Rules: - A hook is the first line spoken or shown, under 12 words. - Use the buyer's language, not marketing language. - No hype words: "game-changer", "revolutionary", "secret". - Every hook must make sense without seeing the rest of the ad. Reply "ready" and wait for the angle.
Step 2: Generate by angle, not by volume
If you ask for 50 hooks in one go, you’ll get 50 versions of the same idea. Instead, ask for five hooks at a time, each from a different angle. Ten angles times five hooks gives you 50 hooks that are genuinely different from each other.
- Pain: name the problem the way they’d say it.
- Outcome: lead with the result they want.
- Curiosity: open a loop the ad closes.
- Contrarian: challenge something they believe.
- Proof: open with a number or a result.
- Mistake: call out what they’re doing wrong.
- Story: drop them into a moment.
- Comparison: old way versus new way.
- Objection: answer the doubt before they raise it.
- How-to: promise a clear, quick win.
Send each angle as its own message: “Angle: contrarian. Give me five hooks.” If one batch comes back weak, tell the model why (“too vague”, “sounds like an ad”) and ask again. Two or three rounds of feedback usually fixes the tone for the rest of the session.
Ten angles times five hooks gives you 50 ideas that are actually different, not one idea written 50 ways.
Step 3: Score them and cut hard
Now you have 50 hooks and a budget for maybe five. Put them in a sheet and score each one from 1 to 3 on five questions:
- Specific: does it name a real situation, number or person?
- Believable: would your buyer nod, or roll their eyes?
- On-offer: does it lead naturally into what you sell?
- Fast: can it land in under two seconds?
- Visual: can you picture the first shot?
Anything under 11 out of 15 goes. You can ask the AI to score them first to save time, but make the final call yourself. The model is good at spotting vague lines. It’s not good at knowing what your market is tired of hearing.
Step 4: Test cheap, read the right numbers
Take your top three to five hooks and put each one on the same ad body. Only the first two to three seconds change. That way, any difference in results comes from the hook and nothing else.
Give each version a small, equal budget and let it run until it has enough impressions to compare. Then judge the hooks on early attention, not sales:
- Thumb-stop rate: 3-second video views divided by impressions.
- Hold rate: how many of those viewers stay to the halfway point.
- Click-through rate, as a tie-breaker.
Sales depend on the offer, the page and the price, so they’re a noisy way to judge a two-second line. Attention is what the hook controls, so that’s what you measure it on.
Test hooks on the same ad body and judge them on thumb-stop and hold rate. Save sales numbers for testing offers and landing pages.
Step 5: Feed the winners back in
This is the step most teams skip, and it’s where the workflow starts to compound. When a hook wins, paste it back into the same chat with its numbers and ask: “This one won. What made it work? Write ten more in the same style.” Do the same with the losers so the model learns what to avoid.
After a few rounds, you stop guessing. You end up with a short list of angles that work for your buyer, and every new campaign starts from there instead of from a blank page.
Mistakes that waste the hour
- Skipping the brief. Ten minutes of input saves an hour of rewriting.
- Testing too many hooks at once. Five well-funded versions beat 20 starved ones.
- Changing the body between versions. Then you don’t know what actually worked.
- Letting AI pick the winner. The market picks the winner. AI just helps you get options in front of it faster.
Run this once and you’ll have a hook bank for the next month. Run it every campaign and you’ll have a clear picture of what your audience responds to, which is worth more than any single winning ad.
