AI video ads: 12 hooks to test for one offer, and how to read the results

The hooks worth testing in AI video ads fall into about twelve types, and the fair way to compare them is to keep the offer and the body of the ad identical and change only the first two seconds. Then read three numbers in order: how many people stopped, how many stayed, and what each result cost. This article lists the twelve types with an example line for each, and explains the test.

One note before the list. Nothing below claims that one hook beats another. Which one wins depends on your product and your audience, and anyone who names a winner without seeing your numbers is guessing. That is the reason to test.

What counts as a hook in AI video ads?

The hook is the part of the ad a viewer sees before deciding to scroll on, which in a vertical feed is two or three seconds. It is a picture, a first line, text on screen and a sound, working together. Changing any one of them makes a different hook.

For the examples we use an imaginary everyday product: an insulated travel mug that keeps coffee hot and does not leak. The lines are illustrations, not ads that ran.

12 hook types to test, with example lines

1. The problem, said out loud. Name the annoyance before the product. Example: “Your coffee is cold before you reach the office.” Suits products that fix a familiar daily irritation.
2. The result first. Open on the end state. Example: “Poured at eight. Still hot at one.” Suits products with one clear, provable benefit. The claim has to be true and you need to be able to back it up.
3. The direct question. Ask something the viewer answers in their head. Example: “Why does every reusable cup leak in your bag?” Suits audiences who already know the category.
4. The silent demonstration. No words, only proof. Example: the full mug turned upside down over an open laptop. Suits physical products and viewers watching with the sound off.
5. The impossible image. A picture that could not be filmed cheaply. Example: the mug steaming quietly in the middle of a snowstorm. Suits crowded categories, and it is where generated video earns its place.
6. The mistake. Tell viewers what to stop doing. Example: “Stop carrying your coffee in a cup with a push-on lid.” Suits markets with a common bad habit. Use it sparingly because it tires quickly.
7. The number. Promise a short list. Example: “Three things your travel mug gets wrong.” Suits considered purchases and viewers who like to compare.
8. The call-out. Address one group by name. Example: “If your commute is longer than forty minutes, this is for you.” Suits products with a clear core buyer. It lowers reach and tends to improve relevance.
9. The side by side. Two options in one frame. Example: a paper cup and the mug after an hour outdoors, with a thermometer in each. Suits products that replace something the viewer already owns.
10. The story that starts in the middle. Open on the moment things went wrong. Example: “Monday, 8.40, a flat white all over my keyboard.” Suits presenter-led ads. Keep it as a presenter telling a story, never a generated person posing as a customer. Our guide to AI spokesperson videos and where they can be used explains where that line sits.
11. The offer first. Lead with the deal. Example: “Two mugs for the price of one, until Sunday.” Suits retargeting and sale periods, when the viewer already knows the product.
12. The reply to a real question. Open on a genuine question from your comments or inbox. Example: “Does it fit a car cup holder? Let’s check.” Suits products with repeated objections. The question must be real.

How do you run a fair hook test on Meta and TikTok?

A test is fair when the hook is the only difference. In practice that means five rules.

1. One variable. Same offer, same body, same end card, same caption, same landing page. Only the first two or three seconds change.
2. Use the platform’s own test tool. Meta has A/B testing in Ads Manager and TikTok has Split Test. Both divide the audience into separate groups so the versions do not compete for the same people. If you place twelve ads in one ordinary ad set instead, the delivery system quickly favours two or three of them and the rest never get a real chance.
3. Test in rounds. Splitting one budget twelve ways leaves each version with too little data. TikTok’s Split Test compares two versions at a time, so treat the set as a knockout: a few hooks per round, then the round winners against each other.
4. Give it time. Meta’s best practices for A/B tests recommend at least seven days, and longer if your customers usually take more than a week to buy. TikTok’s split test best practices also recommend a minimum of seven days, with a maximum of thirty.
5. Give it budget. We cannot give you a figure because it depends on what a result costs in your account. TikTok shows an estimated testing power while you set the budget and recommends a power value of at least 80%. On either platform, a version with a handful of conversions has not proved anything yet.

Do not stop early because one version is ahead on day two. Early delivery is uneven, and weekdays and weekends behave differently.

Which metrics tell you a hook is working?

Read them in this order. Hook rate and hold rate are terms advertisers use, not metrics the platforms report, so you calculate them yourself from the official ones.

What you want to know	On Meta	On TikTok
Did people stop? Hook rate, also called thumb-stop rate	3-second video plays divided by impressions	2-second video views divided by impressions
Did they stay? Hold rate	ThruPlays divided by 3-second video plays	6-second video views, or video views at 75%, divided by 2-second video views
Did it pay?	Cost per result for your chosen objective	Cost per result for your chosen objective

The definitions matter. Meta counts a 3-second video play when the video plays for at least three seconds, or for nearly its whole length if it is shorter, and a ThruPlay when it plays to completion or for at least 15 seconds. TikTok’s video play metrics count a 2-second video view when the video plays for at least two seconds, and a 6-second video view when it plays for at least six seconds, plays in full if shorter, or receives an engagement within the first six seconds. Both exclude replays. Because the two platforms measure different moments, compare hooks within one platform and never across them.

Teams also use different formulas for hold rate. Pick one, write it down and keep it for every test.

How to read the combinations

• High hook rate, low hold rate. The opening works and the body loses people. Fix the next five seconds, not the hook.
• High hook rate, high hold rate, poor cost per result. The opening attracts people who were never going to buy, or it promises something the offer does not deliver. Curiosity hooks do this often.
• Modest hook rate, good cost per result. Usually a call-out or an offer-first hook. Fewer people stop, and the ones who do are buyers. Keep it.
• Low on everything. Drop it and move on.

Cost per result decides the winner. Hook rate only explains why.

How does AI production make twelve variants affordable?

With filmed ads, every new opening needs a person, a place and time on set, so most brands test two hooks and call it a test. With generated production the body of the ad is made once, and each hook is a few extra shots, a new first line and a re-edit. The presenter, the product and the light stay the same across all twelve, which is exactly what a fair test needs.

At Zero Shot, an AI video studio in London, we plan AI ads for Meta, TikTok and YouTube as sets from the start: one offer, one body, many openings, with file names that tell your media buyer which hook is which. If a presenter delivers the lines, our article on AI UGC ads and when they backfire covers the rules on testimonials, and the guide to how AI video is made shows where the time goes.

If you have an offer and want a set of hooks to test against it, send us a short brief with the product, the offer and your best ad so far. We will reply with a fixed price.