So far the feed has chosen from what exists. Now turn it round. The people who make videos can see their numbers. A creator posts ten videos and watches which one takes off. Then they make more like that one. The score does not only pick videos. It teaches people what to make.
15-second video, watched to the end80 in every 100 viewers
3-minute video, watched to the end25 in every 100 viewers
Made-up numbers, real shape.
Take length. If the score rewards watching to the end, a 15-second video gets finished far more often than a 3-minute one. Creators notice: their short videos rise and their long ones sink. Within months, most of what gets made is short. Nobody ordered that. The score did.
Short videos, views400,000 views
Long videos, views40,000 views
A creator's short videos get ten times the views of their long ones. What do they make next?
Long ones, to balance things out.
Not yet. Balance is not what a creator sees. They see one kind of video getting views and the other kind vanishing, and views are what they are paid for.
Short ones. They can see the numbers, and the numbers pay.
Right. Creators watch their numbers the way the app watches yours. What rises gets repeated. That is the score reaching back to the maker.
The same mix. Length does not matter.
Not yet. Length matters to the score if the score counts finishing. The creator's own numbers just showed it mattering ten times over.
Creator
↓→
Views and minutes
↓→
A share of the ad money
Many creators are paid by the app out of the ad money, by the view or by the minute. So the score is their wage. A video that scores well pays rent. One that scores badly does not. That is a strong reason to make what scores well, whatever they would rather make.
Set the video's length and the company's weight on finishing. Minutes watched and likes are added too, on a made-up rule written beside the frames.
From finishing0.8 points
From minutes watched0.2 points
From likes0.8 points
15 seconds · 1A 15-second video at weight 1: 0.8 from finishing, 0.2 from minutes watched, 0.8 from likes. 1.8 points.
From finishing4 points
From minutes watched0.2 points
From likes0.8 points
15 seconds · 5A 15-second video at weight 5: 4.0 from finishing, 0.2 from minutes watched, 0.8 from likes. 5.0 points.
From finishing16 points
From minutes watched0.2 points
From likes0.8 points
15 seconds · 20A 15-second video at weight 20: 16.0 from finishing, 0.2 from minutes watched, 0.8 from likes. 17.0 points.
From finishing0.5 points
From minutes watched0.5 points
From likes0.8 points
1 minute · 1A one-minute video at weight 1: 0.5 from finishing, 0.5 from minutes watched, 0.8 from likes. 1.8 points.
From finishing2.5 points
From minutes watched0.5 points
From likes0.8 points
1 minute · 5A one-minute video at weight 5: 2.5 from finishing, 0.5 from minutes watched, 0.8 from likes. 3.8 points.
From finishing10 points
From minutes watched0.5 points
From likes0.8 points
1 minute · 20A one-minute video at weight 20: 10.0 from finishing, 0.5 from minutes watched, 0.8 from likes. 11.3 points.
From finishing0.2 points
From minutes watched1 point
From likes0.8 points
5 minutes · 1A five-minute video at weight 1: 0.2 from finishing, 1.0 from minutes watched, 0.8 from likes. 2.0 points.
From finishing1 point
From minutes watched1 point
From likes0.8 points
5 minutes · 5A five-minute video at weight 5: 1.0 from finishing, 1.0 from minutes watched, 0.8 from likes. 2.8 points.
From finishing4 points
From minutes watched1 point
From likes0.8 points
5 minutes · 20A five-minute video at weight 20: 4.0 from finishing, 1.0 from minutes watched, 0.8 from likes. 5.8 points.
At weight 1 the five-minute video scores highest. At weight 5 the fifteen-second one does. What does that tell a creator?
Which video is 'best' depends on a weight they cannot see and did not choose.
Right. The same three videos swap places when a number inside the company changes. A creator sees only the result: last month long won, this month short does.
The score is random.
Not yet. Every frame is the same sum. What moved was one weight, and it moved every video's score in a way you can follow with the control.
Short videos are always better.
Not yet. At weight 1 the long one scored highest. Short is better only under one setting of a dial the creator does not hold.
Move both controls.
2022: the app pushes short video
↓→
Photo posts reach fewer people
↓→
Makers switch to video, or leave
Real example. In 2022 a photo app began pushing short videos hard, because a rival video app was winning minutes. Creators who posted photos saw their reach fall. Many switched to video, or left. Later the app said it had gone too far and pulled some of it back. The weight moved, and millions of makers moved with it.
Short video, views400,000 views
Long video, views40,000 views
A creator is paid 2 dollars for every 1,000 views. Their short video got 400,000 views and their long one 40,000. How many dollars more did the short one pay?
dollars
Work out each video's pay, then take one from the other.
Right. Eight hundred against eighty. Seven hundred and twenty dollars more, for a video that took a tenth of the time to make. A pay gap like that is a lesson a creator learns once.
Put in order how a weight inside the company ends up changing what a creator makes.
Tap them in order, first to last.
Weight raised→Short scores higher→Creators see it→Creators make short→Feed fills with short
Right. The step to see is the third. Nobody told the creators anything. They read their own numbers, and the numbers had already been changed by a weight they never saw.
Intro of a hit song, 198620 seconds
Intro of a hit song, 20155 seconds
One study's averages.
A music app's score rewards songs that people do not skip in the first 30 seconds. Over a few years, what happens to songs?
Nothing. Songs are art and do not follow numbers.
Not yet. Songwriters are paid by the stream, and a skipped song is not a paid stream. Songwriters who need the money follow the number, and intros shrank.
Songs get longer, to earn more minutes.
Not yet. The score in the question rewards not being skipped, not minutes. A long song skipped at second ten earns nothing. Length did not grow; intros shrank.
The chorus arrives sooner. Songwriters learn that a slow start gets skipped, and a skip costs the score.
Right. This has been measured: intros of hit songs fell from about 20 seconds in 1986 to about 5 in 2015 as streaming grew. The makers followed the number, the way video creators do.
Lesson complete
Creators make what the score rewards, so the feed shapes what gets made, not only what gets shown.