Testing short video feeds with UK viewers

The short version

• Every opening helped. Each of the 30 raised the chance of keeping watching by 8 to 20 points, and the five topics averaged within five points of each other.
• Money pushed people away most. Four of the five openings most often scrolled past were about money.
• There are three types of viewer. 16 want the future and wealth, 25 want the body and tech, 19 want relationships and family. AI clips are worth 26 points to the first type and 7 to the third.
• No single topic holds all three. Health clips lose the future-and-wealth viewers, AI clips lose the relationships-and-family viewers, relationship clips lose the body-and-tech viewers.
• Health and AI together do. With both in the feed, every type gets a topic worth 19 points or more.

The brief

A short video has about seven seconds to earn the next seven. The thumb is already moving, and the clip has to give it a reason to stop.

The Diary of a CEO, a UK interview podcast, publishes hundreds of these clips, and they fall into five topics: health, money, relationships, psychology, and AI and the future. Anyone programming a feed of them has the same question: which topics keep people watching, and whether the same topics work for everyone.

We took 30 of those Shorts, six per topic, and kept the first seven seconds of each: the opening, before a title or a guest's name has done any work. We showed nothing else. No titles, no thumbnails, no view counts.

Then we asked one question of everyone who took part. How likely are you to keep watching this feed? 1 means scroll past straight away, 5 means definitely keep watching.

How it ran

60 adults in the UK, aged 18 to 54, took part, recruited from a research panel. Each saw 45 feeds, one after another, and rated every one. A feed strung together two to four openings, never two on the same topic, the way a phone feed moves from one clip to the next. That made 2,700 feeds.

The feeds were assembled by design, so each opening sat next to every other opening about equally often and no two people saw the same set. That is what lets us work out, afterwards, how much each opening raised or lowered the chance that a viewer answered 4 or 5. We call that the opening's points: 10 points means 10 more viewers in 100 would keep watching. An answer of 1 or 2 counts as scrolling past.

What everyone said

Viewers kept watching 56 of every 100 feeds, and they were sure of it. 32 in 100 feeds got a 5, definitely keep watching, and 21 got a 1, scroll past straight away. Only 14 in 100 got the middle answer.

Every opening helped. The weakest, "Financial mistake", added 8 points to the chance of keeping watching, and the strongest, "AI is manipulating everyone", added 20. Averaged by topic, the five topics sit close together: AI and the future 17, health 16, relationships 16, money 12, psychology 12. Nothing there separates them enough to programme a feed on.

Money is the exception, and not in a good way. Money openings also pushed people to scroll past: four of the five openings that most often earned a 1 or 2 were about money. "Never see money the same" added 10 points to keep watching and 15 to scroll past at the same time, which means it split the room. The next section shows how.

Three types of viewer

Instead of averaging the 60 viewers, we grouped them by how they answered. People whose scores rise and fall together belong together. Three groups come out, and we named each one after what it kept watching.

16 want the future and wealth. AI openings are worth 26 points to them and money openings 21. Health openings are worth 5.

25 want the body and tech. Health openings are worth 20 points to them and AI openings 18. Relationship and money openings are worth 8.

19 want relationships and family. Relationship openings are worth 26 points to them and health openings 19. AI openings are worth 7.

Howard Moskowitz called groups like these mind-sets: people who watch the same clip and want different things from it. The average blends the three into a score none of them gave.

Where they disagree

Put the five topics next to the three types and the scores point in opposite directions. AI clips are the top choice of the future-and-wealth viewers at 26 and the bottom choice of the relationships-and-family viewers at 7. Relationship clips are worth 26 to the relationships-and-family viewers and 8 to the body-and-tech viewers. Health clips are worth 20 and 19 to two types and 5 to the third.

Within one type the topic matters a lot. Each type's favourite topic beats its least favourite by 12 to 21 points. Across all 60 viewers the gap between the best and worst topic shrinks to five, which is why the average looked flat.

One feed for all three

Each popular topic loses one type. A feed of health clips loses the future-and-wealth viewers, who give it 5. A feed of AI clips loses the relationships-and-family viewers, who give it 7. A feed of relationship clips loses the body-and-tech viewers, who give it 8.

Mix health and AI and nobody is lost. The future-and-wealth viewers get AI clips at 26, the body-and-tech viewers get health clips at 20, the relationships-and-family viewers get health clips at 19. One feed, and every type has a topic worth 19 or more.

In our manifesto this is the second of three answers to a split crowd: give each group its part on one object. Here the object is the feed and the parts are the clips in it.

Who they are

The three types are not ages or apps, though they lean. The body-and-tech viewers are the youngest, median age 29, and most of them watch short videos on TikTok. The future-and-wealth viewers are the oldest, median 32, and watch on Instagram more than anywhere else. Seven in ten relationships-and-family viewers use TikTok. These profiles are estimates, for a reason given in the notes.

What to test next

Two things would sharpen this. Ten openings per topic instead of six, so the differences between openings on the same topic stop being noise. And each clip's title shown with its opening, to learn whether a title changes the answer.