Hooks, Captions and the First Three Seconds

The short answer

In the first three seconds a viewer decides whether to keep watching, and most of that decision is made from the first frame and the caption before you've said a word. The practical implication: your opening visual and your on-screen text do more work than your script, and most creators spend their effort the other way round.

Roughly half of the people who see a short video leave in the first three seconds. That's normal. The job is moving that number a few points, not eliminating it.

What actually happens in three seconds

Break it down and it's not one decision, it's three:

Frame one, before playback registers. The viewer sees a static image and reads the caption overlay. This is closer to how someone scans a photo than how they watch a video. If the frame is a blurry mid-blink shot of you with your mouth open, you've lost people who never heard a word.

Seconds 0–1. Motion and sound start. The viewer is establishing what kind of thing this is — a talking head, a demo, an edit. Whether they can be bothered is largely a question of energy and clarity, not content.

Seconds 1–3. Now they're processing meaning. This is where a hook either gives them a reason to stay or doesn't. If your first sentence is still a preamble at this point, you're relying on goodwill.

Most weak openings fail at the first stage, not the third. People obsess over the hook line and publish it under a bad first frame.

The first frame

Some platforms let you pick a cover; some just take frame one. Either way, treat it as a deliberate choice rather than an accident of where the edit started.

What works: face visible and in focus, eyes open, an expression that isn't neutral, subject filling a decent portion of the frame, clean background. What doesn't: mid-word mouth shapes, motion blur, wide shots where you're small in frame, dark or backlit openings.

A cheap fix that reliably helps: start on a held beat. Half a second of you looking at camera, still, before speaking. Clean first frame, and it reads as composed rather than as if the recording caught you mid-sentence.

Hook types that still work

Some categories have survived heavy use because they're structurally sound rather than novel:

The specific claim. "This one setting is why your audio sounds thin." Concrete, immediately checkable, promises a payoff.
The correction. "You've been told to post daily. That's wrong for most people." Works because disagreement holds attention.
The named audience. "If you shoot on a phone, this matters." Self-selects viewers and tells the rest they can go.
The cost. "This mistake cost me about £2,000." Numbers and stakes.
The demonstration. Just start doing the thing. No verbal hook at all. Massively underused and often the strongest option for anything visual.

The exhausted ones

These have been used to death and now signal "generic content" to anyone who spends time on these platforms:

"Stop scrolling."
"Nobody is talking about this." (People are always talking about it.)
"Here's a hack that changed my life."
"POV:" on anything that isn't actually a point of view.
"Wait for it." Especially when the payoff doesn't justify the wait, which is most of the time.
Fake-mistake openings — dropping something, pretending the camera wasn't rolling. Audiences learned this pattern.

The common problem is that they're about attention-grabbing rather than about the content. Test it this way: could your opening line be swapped onto a completely different video unchanged? If yes, it's not a hook, it's noise.

Caption as hook

A large share of feed viewing happens with sound off. Which means for those viewers, your on-screen text is the hook and your spoken line is irrelevant.

So write a text hook that stands independently. Not a transcription of your first sentence — a headline for the video. Four to eight words, high contrast, positioned in the upper-middle of the frame where platform UI won't cover it.

Practical points:

It must be legible at thumbnail size. If you can't read it on a phone at arm's length, it's too small or too thin.
Keep it up for at least two seconds. Text that flashes past can't be read.
Don't duplicate your speech. The text should add a layer, not echo.
Watch the safe zones. Bottom quarter and right edge belong to the platform.

Separately, burned-in captions for the whole video are worth the effort. Sound-off retention with captions is meaningfully better than without, and it's the least creative work in the entire process — which makes it the easiest thing to systematise or delegate.

Testing without fooling yourself

Most "hook testing" is people looking at two videos and drawing conclusions. That's not testing, that's noise-reading.

What's actually workable:

Use retention curves, not view counts. Views are driven by distribution. The 3-second retention percentage is the number that reflects your hook.
Change one thing. New hook style, same topic, same format. If you change the hook and the editing style, you've learned nothing.
Wait for volume. A minimum of five to eight posts per variation. Individual videos vary enormously for reasons that have nothing to do with the hook.
Look for 5%+ differences. Anything smaller is within normal variance.
Repost your best performer with a different opening after 60 days. Same content, different hook, different audience slice. The cleanest comparison you'll get.

Keep a simple record: date, topic, hook type, 3-second retention, average watch time. After thirty entries you'll see patterns you'd never spot by feel. Usually they're specific to you and contradict at least one piece of general advice.

The caveat worth stating

A strong hook on weak content produces a spike in views and a collapse in retention, and platforms notice that pattern. It also trains your audience that your videos don't deliver, which is a slow, expensive problem.

Hooks make good content get watched. They don't make bad content good. If retention falls off a cliff at eight seconds, the problem isn't your first three.

Where this fits

Hooks, captions and cover frames are exactly the kind of detail that gets skipped when you're editing your twelfth clip of the day, and they're also where a lot of the performance sits. CORE handles repurposing and channel management, including the unglamorous consistency work — captioning, cover selection, testing across a run of posts. If your content is good but the openings are rushed, that gap is usually worth closing.