How do I write hooks that keep people watching, and hold retention mid-video?

Short answer

A hook is the first 30 seconds of your video. Its only job is to make the viewer stay. Retention means how long people keep watching after that. You win both the same way: make a promise, keep it, then make the next promise.

In practice that means: prove the title is true in the first 10 seconds, say why it matters in the next 10, give a one-line roadmap, and then tease the next part every 2 to 3 minutes for the rest of the video. Put your best moment two thirds of the way through, not at the end.

Why hooks and retention decide how far a video spreads

YouTube shows your video to more people when viewers keep watching. If most people leave in the first 30 seconds, YouTube stops recommending the video, no matter how good the rest of it is. The algorithm never sees the good part, because almost nobody reached it.

This is why a great idea with a weak opening dies quietly, and why a plain idea with a strong opening can travel. In YouTube Studio the two numbers to watch are average view duration (how much of the video people watched) and the retention graph (where exactly they left). A sharp drop in the first 30 seconds is a hook problem. A slow slide through the middle is a retention problem. The fixes are different, and both are below.

What to do: the promise chain, step by step

1. In the first 10 seconds, prove the title is true. Show one real example straight away. Do not start with "welcome to my channel" or your name. If the title says 23 facts that sound wrong, the first thing the viewer hears is one fact that sounds wrong.
2. In the next 10 seconds, say why it matters to the viewer. What will they get wrong, lose, or miss if they do not know this? Make it about them, not about you.
3. Only now say what the video will cover. Keep it to one line. Repeat the number from the title if there is one. That is the roadmap, and it sets up every promise that follows.
4. Every 2 to 3 minutes, tease the next part before you show it. For example: "The next one is the one I still cannot believe." This gives people a reason to stay through the transition, which is where most mid-video drop-off happens.
5. On faceless videos, change what is on screen every 5 to 10 seconds. A still image for 30 seconds makes people leave, even when the narration is good. Motion does not have to be fancy. A new map, a zoom, a caption, a cut is enough.
6. Put your best moment about two thirds of the way through, not at the very end. Most people never reach the end. Saving the best for last means most viewers never see it. Two thirds of the way through is late enough to reward people who stayed and early enough that most of them are still there.

How to learn the pattern from videos that already work

You do not have to guess at any of this. Open the transcripts of the three best-performing videos in your niche. Mark every line that teases what is coming next, and note the timestamp. You will see the same pattern in all three: an early proof, a reason it matters, a short roadmap, and then a new tease every few minutes.

In OutlierKit, the outlier finder shows you which videos beat their channel's normal views, and the transcript tool gives you the words with timestamps. The prompt at the end of this page does the whole exercise inside Claude, for three videos at once. If you want the broader script structure around the hook, the script writing guide covers the full video, and the scriptwriting methods comparison shows how named methods handle the same opening.

What the example shows: 23 geography facts, 22.9x

This geography video starts with one surprising fact: Reno is further west than Los Angeles. That is the proof, delivered in the first seconds. Then it says the creator's own map of the world was wrong, which is the "why it matters" beat, because yours probably is too. Then it lists what is coming: 23 facts. That is the one-line roadmap.

After that, a new surprise arrives every 20 to 30 seconds. There is no fancy editing and no face on camera for most of it. People stay because they always want the next fact. That is what got a 19,000-subscriber channel 496,000 views, which is 22.9 times what the channel normally gets. The multiplier is the point: a channel that size cannot get half a million views from subscribers. Strangers clicked, and strangers stayed.

The same promise on three more small channels

One video could be luck. So on 7 September 2026 I pulled the other facts videos in the OutlierKit index that make the same promise in the title: a number, and a claim that the facts sound wrong. Three more small channels had done the same thing.

Maply, with 5,720 subscribers, got 113,000 views on 15 Australia facts "you won't believe are true", 86.9 times its normal, and 38,000 views on 25 US facts "your teacher got wrong", 22.3 times. Borderline Wonders, with 15,400 subscribers, got 347,000 views on 25 Mexico facts that "sound fake", 26.7 times. Four channels between 5,720 and 19,000 subscribers, one title shape, every one beating its normal by 22 times or more. I have not marked up these three transcripts the way I did the first; the prompt at the end of this page does that for you, on your own niche.

The four openings that lose viewers

• The greeting. "Hey guys, welcome back." Nobody clicked for a greeting. Cut it.
• The backstory. Explaining how you came to make the video before making it. Prove the title first; the story can come at minute three.
• The full table of contents. Listing every section in detail is not a roadmap, it is a spoiler. One line is enough.
• The saved-up finale. Holding the best moment for the last 30 seconds, where fewer viewers remain than at any other point in the video.

Words used on this page

Hook

The opening of a video, usually the first 30 seconds.

Retention

The share of the video people actually watch. 50% retention means the average viewer watched half.

Loop

A question or tease you open and close later. Open loops keep people watching.

Outlier multiplier

A video's views divided by what its channel usually gets. 22.9x means the video did 22.9 times the channel's normal. The outlier score guide explains how it is calculated.