Video Ad Hooks: Second Three Stops the Scroll, Seconds Four to Ten Make the Sale

Meta and Nielsen found that up to 47% of the value in a video campaign was delivered in the first three seconds, and up to 74% in the first ten.

Nearly every guide to video ad hooks quotes the first half of that sentence.

We think the second half is the more useful one.

Up to 27 points of value sit between second three and second ten, and that is exactly where most hooks fall apart.

They stop the scroll with a promise, and the next seven seconds don't keep it.

A hook is a promise, and second four is where it gets checked

A video ad hook is the opening one to three seconds of an ad, and its only job is to earn the next few seconds.

The metric everyone grades it on is hook rate: 3-second video plays divided by impressions.

That number rewards anything surprising.

A fake notification sound, a stranger shouting, a bold claim the product can't back up.

All of them stop thumbs, and all of them lose viewers the moment the ad turns out to be about something else.

So grade the hook on two numbers together.

Hold rate is ThruPlays divided by 3-second plays, where Meta counts a ThruPlay as a view to completion or at least 15 seconds.

Multiply the two and you get the share of everyone who saw the ad that actually watched it.

Here is the math with made-up numbers.

Hook A gets a 40% hook rate and a 15% hold rate, so 6% of impressions watch the ad through.

Hook B gets 28% and 30%, so 8.4% do.

Hook B looks worse on the dashboard most people check first, and it is the better ad by about 40%.

Our rule: never kill a hook on hook rate alone, and never crown one until you have seen what happens after second three.

Our guide to testing AI video ads covers how much spend each variant needs before either number means anything.

Pick the hook by what the viewer already knows

Most hook lists hand you 25 formulas and no way to choose between them.

The oldest useful filter is still Eugene Schwartz's five stages of awareness from Breakthrough Advertising, published in 1966.

The right opening depends on how much the viewer already knows about the problem and about you.

• If they don't know they have the problem, open on a scene they recognize, not on your product. A founder rewriting the same onboarding email for the fourth time says more than any claim.
• If they know the problem, name it in their words in the first line. "Your demo video is eleven minutes long" beats "Struggling with video?"
• If they are comparing solutions, open on the difference. Show the one thing the alternatives can't do, on screen, in the first frame.
• If they already know you, open on the offer or the new thing. Retargeting audiences don't need to be hooked twice.

Here is what each one sounds like for one product.

Take Datavue, a made-up tool that saves founders from copying Stripe numbers into slides at 11pm every month.

The hooks below are illustrations we wrote for this post, not tested ads.

• Unaware: open on a founder at 11pm copying numbers from a spreadsheet into a slide, with the caption "Every month. Same three hours."
• Problem-aware: "Your investor update takes three hours. It should take three minutes." over the product generating one.
• Solution-aware: a split screen, a template on the left still waiting for numbers, Datavue on the right already filled in from the live database. Caption: "Templates don't pull the numbers."
• Product-aware: "Datavue now writes the commentary too. Free for your next update."

Most early-stage startups are selling to people in the second group, who know the pain but not the fix.

That is why the plain problem-statement hook is usually the right first test for them, and the clever pattern interrupt usually isn't.

On a muted feed, the first frame is the hook

Feed video on Meta autoplays without sound, so a spoken hook that opens on a wide shot of an office is no hook at all.

The words need to be on screen in the first frame, and the frame itself needs to show something specific.

Google's guidance for YouTube ads points the same way.

Its ABCDs framework says to "jump in" and to introduce the brand or product "from the start", and Google cites Google/Kantar research from April 2021 finding that ads following the framework saw as much as a 30% lift in short-term sales likelihood.

That cuts against a popular piece of TikTok advice, which is to hide that you are an ad for as long as possible.

We side with Google, with one adjustment.

Show the product early, not the logo.

A viewer who sees the product working in second one and decides to stay has qualified themselves, and that is the person whose watch time you want to pay for.

A good hook should lose some people on purpose. The ones who leave at second two because they saw what you sell were never going to buy it.

YouTube adds its own clock: on skippable in-stream ads the skip button appears after five seconds, so the hook there has to carry the viewer past second five, not just second three.

Writing ten hooks with AI, and why the first nine sound the same

Ask any chatbot for ten video ad hooks and you will get "Tired of...?", "Stop scrolling if..." and "What if I told you..." in slightly different clothes.

The feed is full of them already.

The fix is to give the model your customers' language instead of your product description.

Paste in support tickets, sales call notes or review snippets and ask for hooks that reuse their exact phrases.

A prompt that does this: "Here are 30 sentences our customers wrote about the problem before they found us. Write 10 opening lines for a 20-second vertical video ad, each under 12 words. Reuse their phrasing. No questions, no 'tired of', no 'stop scrolling'. For each, describe the first frame in one sentence."

Asking for the first frame alongside the line matters, because on a muted feed the frame is doing half the work.

The hooks that come back sound like a person, because a person wrote the raw material.

AI video generators also make the cleanest hook test possible.

Generate or cut five different opening shots, keep the body of the ad identical from the same frame onward, and only the first three seconds change.

If one variant wins, you know the hook did it.

Our walkthroughs on making a video ad with AI and making a TikTok ad with AI show how to set up the cut so the swap is clean, and the AI video ads hub covers the tools and platform cuts around it.

Hooks we would stop using

"Stop scrolling!" tells the viewer you know they want to leave.

A fake phone notification or a fake app interface in the first frame wins the stop and loses trust, and Meta's advertising standards bar images that show functionality that doesn't exist, like fake buttons and notifications.

"POV:" openings only work when the shot actually is someone's point of view.

AI avatars playing surprised customers are the one we would drop first.

The FTC's 2024 rule on fake reviews and testimonials covers testimonials attributed to people who don't exist, including AI-generated ones, and an avatar reacting to your product as if it were a real buyer sits close to that line.

An avatar presenting the product as the brand is a different thing, and fine.

The promise has to survive the click too

Everything above stops at the end of the video, and the hook's promise doesn't.

If the opening says "a launch video in 48 hours" and the landing page leads with a pricing grid, the viewer who clicked is checking the same promise a third time and finding it missing.

So the hook with the best hook-rate-times-hold-rate number can still be the wrong one if your page can't back it up.

Before you scale a winning hook, read your landing page's first screen out loud right after watching the ad's first three seconds.

If they sound like two different companies, the test isn't finished.

At Flowjam we make launch and product videos, and if you want hook variants cut for an ad, send the landing page with the brief, because a hook the page can't back up isn't worth testing.

Frequently asked questions