How long do you actually have before a viewer scrolls past?

Treat the first two seconds as the decision window and the third second as confirmation. There is no universal cut-off number that applies to every feed and every audience, which is why you should read your own retention curve rather than a quoted average. What is consistent is that the first frame and the first spoken line carry almost all of the load.

Should I say hi and introduce myself at the start?

No. A greeting spends the exact seconds you need for the promise. If a viewer needs to know who you are, your on-screen name and your channel page already say it. Every word before the hook is a word the viewer can leave on.

Can I fix a weak hook by changing the edit?

You can fix a slow first second with a tighter cold open. You cannot fix a hook when the topic itself has no pull, because no wording will make an uninteresting subject land in two seconds. If several clips on the same topic all flatten out early, change the topic, not the script.

Is it better to reupload a clip with a new hook, or leave it?

Test the hook on the next clip instead. Reuploading the same content to chase a better first second usually splits the performance you already earned and gives you no clean comparison. A cleaner method is to film two first seconds for the same clip and post them as separate videos on different days.

Short Video Hooks: Holding Viewers Past the First Two Seconds

Most creators do not have an attention problem, they have a first-two-seconds problem. The clip is fine, the payoff is real, and the topic is searchable — but the viewer never reaches any of it, because the opening gave them no reason to stay. Hooks are not a personality trait. They are a structure you can build, score and test.

What a hook actually is

A hook is not a sentence. It is three things arriving at once, and a weak one normally fails on the part the creator paid least attention to.

The first frame. This is what a viewer sees while the clip is loading and what becomes your cover image in the channel grid. If the frame at zero seconds is your face mid-blink against a blank wall, you have spent the frame on nothing.

The first spoken line. Not an introduction. A statement, a question, or a claim that the rest of the clip has to pay off. If the line could be cut and the clip would still make sense, it is not a hook.

The first action. What moves on screen in that window — a hand doing something, a result appearing, a visual change. Movement is what stops a thumb that is already halfway to the next clip.

Fix the parts in that order. A strong opening line over a dead first frame still loses, because the frame is read before the audio starts.

There is a useful distinction between a hook and a premise. The premise is what the clip is about. The hook is the reason a stranger is still there when the premise arrives. A clip with a strong premise and a slow hook is not a bad idea — it is a good idea with the wrong opening, and that distinction tells you exactly what to change. Rewriting the premise throws away a good idea; rewriting the hook is a ten-minute edit.

Budget the window honestly. Two seconds is not much, so the first spoken line has to be short enough to land inside it — roughly one clause, not a sentence with a setup and a payoff. Long openings are almost always a symptom of a script written for the edit rather than for the feed. Read your first line out loud and time it. If it takes longer than three seconds, it is a paragraph wearing a hook costume.

Five opener structures that earn the second second

These are shapes, not scripts. Each one works because it removes a reason to leave.

1. Result first, process after. Show the finished thing, then rewind. The viewer now knows what they are waiting for, which converts curiosity into patience. This is the most reliable opener for anything with a visible output — a plate of food, a repaired object, a dashboard number.

2. The contradiction. State something the viewer believes and immediately complicate it. “More uploads made my reach worse” works because it cannot be dismissed without hearing the rest.

3. Mid-action entry. Start inside the task, not before it. No setup, no walking to the desk, no clearing of the throat. You are borrowing the momentum of a scene that is already happening.

4. The stake. Name what is at risk or what is being spent. Money, time, or a deadline all work, as long as the stake is real and you do not inflate it.

5. The specific number or constraint. A count, a limit, a budget. Specificity reads as competence, and competence is what makes a viewer trust the next eleven seconds.

Pick the structure that matches the shape of your content rather than the one that feels most like your personality. Visible outputs want result-first. Opinions and myth-corrections want the contradiction. Process content wants mid-action entry. Anything with a price, a deadline or a hard limit wants the stake. Data-led content wants the number. When a format stops working, most creators change their delivery style; the structure is usually the faster and more honest fix.

How to read a retention curve to score a hook

Your platform already tells you whether the hook worked. The retention graph is the scorecard, and it is more useful than view count because it separates the two failure modes.

The cliff in the first three seconds is a hook problem. Viewers arrived and left before the content began. That is the only shape that a rewrite of the opening line can fix.

A gradual slide through the middle is a pacing problem, not a hook problem. Rewriting the first line will not help. You need to cut, reorder, or add a new beat where the slide begins.

A bump at the end is a loop signal: viewers rewatched. When that happens, the last second is doing work, and you should copy whatever you did there into the next clip deliberately rather than by accident.

Two notes on reading the graph properly. First, compare clips on the same topic rather than across your whole channel, because the topic changes the curve shape more than the quality of the hook does. Second, a clip that holds a small, well-matched audience is a better hook test than one that briefly holds a large, mismatched one — a lower number with a flatter curve can be the stronger result.

Score the hook itself with arithmetic you can reproduce from your own dashboard:

Example: a clip is shown 40,000 times and 12,000 viewers are still watching at the three-second mark. That is a 30% three-second hold. Raise it to 45% and, at the same impression volume, 6,000 more viewers see the rest of the clip.

Those numbers are an illustration of the calculation, not a published platform figure — the point is that you can compute your own hold rate from two counts you already have. Track it clip by clip on one topic. A number that moves when only the opening changed is the only clean evidence you will get.

Testing a hook without reposting the clip

The trap here is impatience. Deleting and reuploading to chase a better opening usually destroys the data you needed and gives you nothing clean to compare against.

Better options, in order of how much they tell you:

Film two openings for the same clip. Same body, two different first two seconds. Post them on different days, so feed conditions are at least roughly comparable. This is the only method that isolates the hook, because everything after the opening is identical.

Change the cover frame on your channel grid. On platforms that let you pick the still, the cover is a separate surface from the feed and it affects clicks from your profile and channel page. A better cover tells you whether the hook problem was visual or verbal.

Rewrite the on-screen text in the first second. Many viewers encounter the clip before the audio lands, so the first line of text is doing part of the hook’s job. If that text was your title repeated, it was wasted.

Post the clip again as a different format. A three-minute version and a forty-second cut of the same idea are not the same clip. That is a format test, and it is worth doing, but do not confuse its result with a hook result.

What you should not do is judge a hook from a single clip. Short-form distribution varies between uploads for reasons that have nothing to do with your opening line, and one bad posting tells you almost nothing.

When the hook is not the problem

If both openings of a topic flatten in the same place, stop rewriting sentences. The topic is the constraint.

Three diagnostic questions:

Would a stranger click this if a friend sent it to them? If not, the idea is not shareable, and no opening line changes that.
Does the clip answer a question someone is already asking? Hooks accelerate demand. They do not create it.
Is the promise in the first two seconds actually delivered? A hook that overpromises buys attention and then loses the audience at the payoff — which shows up as a mid-clip slide, not as an early cliff.

One production note worth knowing if you are planning longer clips: Shorts can now run up to three minutes, and square or vertical videos uploaded after 15 October 2024 at up to that length are classified as Shorts rather than long-form uploads. That extra time is not an invitation to add an intro. It is room for a second hook at around the twenty-second mark, which is where the mid-clip slide usually starts.

A twenty-minute hook pass for your next batch

Before you film anything, run the same short pass on every clip in the batch.

Write the first spoken line as a sentence you would say to someone who did not ask. If it needs context, it is not a hook.
Decide what is on screen at zero seconds. If the answer is nothing in particular, film a new opening shot.
Cut every greeting, every “so”, and every restatement of your title.
Place your second hook at the point where previous clips started to slide.
Write the last line so it can be rewatched without feeling like a repeat.
After posting, compute your three-second hold rate and write it next to the topic, not next to the clip. Topics are what you are actually testing.

None of this requires better gear or a bigger audience. It requires you to spend the first two seconds on the viewer instead of on yourself.

Run the pass on a whole batch rather than one clip, because comparability is the point. If you change three things between two clips, the result teaches you nothing. If every clip in the batch gets the same treatment and only the topic and the opening line differ, the hold rates you collect start to mean something you can act on.

If you want to know what to do with the audience once the hook works, see turning viewers into an email list you own — and for the arithmetic of whether the views are worth chasing at all, read how platforms actually pay per 1,000 views.