What Makes a Good Hook for TikTok and YouTube Shorts

Researchers building a video streaming system needed to know when people swipe. They recorded 15,344 swipes from 133 paid participants and 3,069 more from 25 students, on a real short-video feed.

The swipes did not spread out. 29% landed in the first fifth of a video. 42% landed in the last fifth. Between 60% and 80% of the way through, only 6% of swipes happened at all.

That is what a hook is for, and it is the whole reason the first seconds get treated as a separate job from the rest of the video.

Show one specific moment in the first 3 seconds and leave one question open

The whole answer, before the evidence behind it.

• Show a specific moment that is already happening, rather than announcing what the video will be about.
• Get the point of the video on screen inside 3 seconds, because that is what the highest click-through videos do.
• Leave exactly one question unanswered, so the viewer has a reason to still be there in second four.
• Say the hook out loud and write it on screen at the same time, because the two are counted separately in TikTok's own data.
• Be specific when the videos around you are vague, and vague when the videos around you are specific.

Here is every common opening, and how much evidence sits behind each one.

Fix the first fifth of your video, because 29% of every swipe away happens there

The best measurement of when people leave a short video came from engineers who were not studying hooks at all.

They were building a streaming system and needed to know which parts of a video to download. So they recorded real swipe behaviour on a real feed.

• 133 paid participants produced 15,344 swipes, and 25 student volunteers produced another 3,069.
• 29% of swipes landed inside the first 20% of the video's length.
• 42% of swipes landed inside the last 20%, which is mostly the video simply ending.
• Only 6% of the students' swipes fell between 60% and 80% of the way through a video.
• One video in the sample took 60% of its swipes inside the first 20%, which is what a failing hook looks like in data.
• Two other videos in the same sample took 60% and 80% of their swipes in the final few seconds, which is what a working hook looks like.

The same video, watched by different groups of people, produced almost the same swipe shape both times. So the shape belongs to the video, not to the crowd.

Put your key message in the first 3 seconds, because 63% of the best videos do

TikTok publishes its own creative guidance, drawn from the highest-performing ads on the platform.

• Over 63% of the videos with the highest click-through rate show their key message or product inside the first 3 seconds. Click-through rate here means the share of the times a video was shown where somebody tapped the link.
• TikTok's own instruction is to get straight to the point and keep it short and direct.
• 61% of the best-performing ads on TikTok use half or more of the nine creative tips TikTok publishes.
• The 63% figure is a count of what winning videos happen to do, with no losing group to compare against. A pattern, not proof.

The 3 second figure keeps appearing because it is roughly how long a person gives a video before the decision is made.

Build the hook for 1.7 seconds of attention, because that is what a mobile feed gives

The 3 second rule is generous compared to what the attention research says.

• Meta reports the average person spends 1.7 seconds with a piece of content in a mobile feed.
• Attention on a desktop screen averages 2.5 seconds, so a phone gets about two thirds of what a computer gets.
• Nielsen found that people who saw a video ad for under 2 seconds still produced 38% of all the brand recall those ads generated.
• Meta has not published the method behind the 1.7 second figure, so treat it as a budget rather than as a measurement.

Read your hook out loud with a timer. If the point has not arrived by the time you hit two seconds, the point is too late.

Open a gap the viewer wants closed, because curiosity is felt as a missing piece

The reason a hook works at all was described in 1994 by the economist George Loewenstein, and it is not the reason most people assume.

Curiosity is not a pleasure you offer somebody. It is a discomfort you create.

• Curiosity starts when a person notices a gap, not when they are offered something interesting.
• The gap has to be small enough that the viewer can feel the shape of the missing answer. A completely unknown subject creates no gap at all.
• A viewer who can guess the answer has no gap left, so the hook stops working the moment it is predictable.
• A hook that promises a gap without showing anything gives the viewer nothing to feel deprived of.
• Showing the end result first opens the largest gap available, because the viewer can see the answer but not how it happened.

Use Repeat Recorder to hear your first line back a second after you say it

A hook has to land in about two seconds of speech, and the only way to know whether yours does is to hear it. Repeat Recorder, one of my own apps, records your voice and plays it straight back to you without you touching anything, so you can fix a line while you still remember what you were trying to fix.

• The gap between saying your hook and hearing it back is about one second, which is short enough that you can still hear what you meant to say.
• Automatic mode starts recording when you speak and replays when you stop, so you can run twenty attempts at one line without ever looking at the screen.
• The countdown before recording gives you a breath to get ready, which matters when the thing you are practising is a two second line.

Expect an open loop to be finished, not remembered, because the recall effect did not replicate

The open loop idea comes from Bluma Zeigarnik, who reported in 1927 that people remembered interrupted tasks better than finished ones.

Half of that famous result has now failed.

• A 2025 meta-analysis of the Zeigarnik and Ovsiankina effects found no memory advantage for unfinished tasks at all.
• The 2025 meta-analysis did find a real tendency for people to go back and finish something they had started.
• An earlier replication by Van Bergen in 1968 also found no difference in recall between finished and interrupted tasks.
• The surviving half of the Zeigarnik effect is the useful half for a video hook, because you are trying to stop somebody leaving rather than trying to be remembered.
• The pull to finish is strongest when the person cares about the task, so an open loop about something the viewer does not care about closes by itself.

Be concrete when rival videos are vague, and vague when they are concrete

How much to give away in a hook has been measured properly, on headlines rather than on videos.

The publisher Upworthy ran real randomised tests on its own readers, and the whole archive was later released to researchers.

• 8,977 randomised headline experiments were analysed, covering 70,546 headlines at an average click-through rate of 1.33% of the times a headline was shown.
• Where the competing headlines were vague, being more concrete raised the click-through rate by a predicted 5.5%.
• Where the competing headlines were already concrete, being more concrete lowered the click-through rate by a predicted 9.9%.
• Half of all 8,977 tests were run in fields where the headlines were already concrete, so being more specific was the losing move more often than the winning one.
• The winning amount of detail depends entirely on the videos beside yours, which means there is no fixed rule to copy.
• Nobody has run the equivalent experiment on video hooks, so this is the closest available evidence rather than direct evidence.

The practical version is to open the feed you are posting into and read the first line of the videos already there.

Rewrite a topic hook into a moment already happening, because a topic invites judgement

A hook that names a topic asks the viewer to decide whether they are interested in that topic.

Deciding is how you get skipped. A moment already in progress gives them nothing to decide about yet.

Use Camera to Clipboard to compare your first frames on a big screen

Choosing between opening frames is much easier when you can see them next to each other at a decent size instead of on the phone that shot them. Camera to Clipboard, another of my own apps, puts a photo from your phone straight onto your computer's clipboard, so a frame you just shot is one paste away from a document you can lay them all out in.

• The photo arrives on your computer's clipboard as soon as you take it, so there is no cable, no email to yourself and no cloud folder to wait on.
• Laying several opening frames side by side is the only way to answer whether yours is more specific or vaguer than the videos it will sit beside.
• Pairing is a one-time QR code scan between the phone app and the desktop app, and it works on Windows and macOS.

Look straight down the lens, because 33% of the best performing TikTok ads do

TikTok calls this breaking the fourth wall, which just means addressing the viewer directly instead of performing beside them.

• 33% of the TikTok ads that people watched furthest through have somebody looking straight into the camera and speaking to the audience.
• TikTok's own advice is to turn looking down the lens into a real exchange by asking the viewer questions and provoking a reaction.
• The 33% figure is a count of what winning ads happen to do, with no losing group to compare against. A pattern, not proof.
• Looking down the lens costs one framing decision and no money, which is why it is worth doing on weak evidence.

Put words on screen in the first second, because 40% of the best TikTok ads do

The written hook and the spoken hook are separate jobs, and they are counted separately in TikTok's own data.

• 40% of the TikTok ads that people watched furthest through use text overlays, effects or stickers.
• A written hook is the only part of your hook that survives when the phone is on silent.
• Meta found that adding subtitles raised the time people spent watching a video ad by an average of 12%.
• The words on screen and the words you speak should not be identical, because reading and hearing the same sentence wastes one of the two channels.
• Keep the on-screen line short enough to read in one look, because a viewer who is still reading at second three has not started watching yet.

Turn the sound on, because 88% of TikTok users say sound is essential to the app

TikTok is one of the few feeds where the sound plays by default, and the platform has surveyed what that is worth.

• 88% of TikTok users told Kantar that sound is essential to the TikTok experience.
• 73% of the same people said they would stop and look at an advert that had audio.
• TikTok reports that music tracks above 120 beats per minute tend to keep people watching further through than slower ones.
• The three sound figures above are answers people gave in a survey rather than behaviour anybody recorded, so read them as a direction rather than a size.
• The sound survey was run for TikTok by Kantar, so it was commissioned by the company the result flatters.

Write the hook so it still works with the sound off, then add sound that makes it better.

Shoot vertical and fill the screen, because 25% more people then reach the 6 second mark

The cheapest number on this page belongs to a decision you make before filming anything.

• TikTok reports that videos shot vertically at 9 by 16 keep 25% more people watching to the 6 second mark than videos that do not fill the screen.
• A horizontal video pasted into a vertical feed leaves black bars, which reads as an advert before a single word is spoken.
• TikTok also advises balancing an authentic look against a polished advertising look, because content that looks like the feed around it gets treated like the feed around it.
• The 25% vertical video figure is TikTok's own, with no published method, so treat it as a strong reason to do something free rather than as a measured effect.

Open on a face or a curved shape, because curves reach awareness before hard edges do

Faces are the standard advice for a first frame, and the research behind that advice says something slightly different.

• In experiments where images were hidden from conscious awareness, faces broke through into awareness faster than objects did.
• The researchers concluded the advantage probably came from curvature rather than from the images being faces. Curved shapes, not faces specifically.
• A curvature explanation makes a rounded shape, a soft silhouette or a face all equally good candidates for the first frame.
• The face and curvature study measured how fast an image reached awareness in a lab, not how many people watched a video.
• A face is still the easiest curved thing to put in a first frame, which is probably why the advice took the shape it did.

Use Repeat Recorder to drill your hook hands free while you walk

Writing ten versions of a hook takes an afternoon, and picking the one that sounds right takes saying all ten out loud many times over. Repeat Recorder, one of my own apps, runs that loop without a screen, so the practice can happen while you are walking or driving instead of sitting at a desk.

• You never touch the screen in automatic mode, so a hook can be drilled while your hands are busy with something else.
• Saving the best take to a named folder means the version you liked on Tuesday is still one tap away on Friday.
• Playing a take back on a loop that runs until you stop it is how a two second line stops sounding like a line you are reading.

Track viewed versus swiped away, because that number is the closest thing to a hook score

YouTube added a Shorts metric that measures exactly the decision a hook is responsible for.

• Viewed versus swiped away shows the share of times your Short appeared in the Shorts feed where somebody watched rather than swiping past.
• Almost nothing except your opening has happened when a viewer chooses to watch or swipe, which is what makes it a hook score rather than a video score.
• You find it in YouTube Studio under Analytics, in the numbers for an individual Short.
• YouTube has published no target for it, so every benchmark you see quoted is a creator rule of thumb rather than an official number.
• Comparing your own Shorts against each other is the only honest use of the metric, because the number moves with who the video was shown to.
• TikTok has no equivalent published metric, so the nearest substitute there is the graph in your analytics showing how many people are still watching at each second.

Change only the first 3 seconds when you test, because everything else moves the result too

Most of the numbers on this page are counts of what winning videos happen to do. Only one came from a randomised experiment, and only one more came from recorded behaviour.

Here is what each source can and cannot tell you.

• Post the same video twice with only the first 3 seconds different, because any other change makes the comparison meaningless.
• Judge the test on viewed versus swiped away rather than on views, because views depend on how far the platform pushed the video.
• One pair of videos proves nothing, so run the same swap across several videos before believing the result.

Count the people who swiped, because under 3 seconds still lifted ad recall by 47%

A swipe reads as a total loss, and the brand research says it is not.

• Nielsen analysed 173 brand-effect studies of video ads on Facebook for the platform itself.
• People who watched for less than three seconds still showed a 47% lift in ad recall.
• The same group showed a 32% lift in brand awareness and a 44% lift in purchase intent.
• The lift began before one second of watching, so it was produced by the first frame rather than by the video.
• The Nielsen numbers are brand measures rather than views, so a strong first frame builds recognition even where it does not win the watch.
• The 173 Nielsen studies were run for Facebook, which is the company the result flatters, so weigh them accordingly.

The first frame is doing work on everybody who scrolls past, not only on the people who stay.

Conclusion: what makes a good hook for TikTok and YouTube Shorts, in one list

Every rule on this page, in the order they appear.

• Show one specific moment inside the first 3 seconds and leave one question open.
• Fix the first fifth of the video. 29% of all swipes away happen there.
• Get the point of the video on screen inside 3 seconds. 63% of the highest click-through videos do.
• Write the hook to land in 1.7 seconds. That is the average attention a mobile feed gives.
• Open a gap the viewer wants closed, because curiosity is felt as a missing piece.
• Expect an open loop to be finished, not remembered. The memory half of that idea failed.
• Be concrete when rival videos are vague, and vague when they are concrete.
• Rewrite a topic hook into a moment already happening. A topic invites a decision, and deciding is how you get skipped.
• Look straight down the lens. 33% of the best performing TikTok ads do.
• Put words on screen in the first second. 40% of the best performing TikTok ads do.
• Make the hook work with the sound off, then add sound. 88% of TikTok users say sound is essential.
• Shoot vertical and fill the screen. 25% more people then reach the 6 second mark, and it costs nothing.
• Open on a face or another curved shape. Curves reach awareness before hard edges do.
• Track viewed versus swiped away in YouTube Studio. It is the closest thing to a score for your hook.
• Change only the first 3 seconds when you test. Anything else makes the comparison meaningless.
• Do not open with a follow request or a slow build. Neither has ever been measured to work.