Pattern interrupt

The hook buys the first second. Pattern interrupts buy every second after that. Attention on a feed decays on a predictable curve, and a change in the signal resets it. That is the entire mechanism.

Where they go

Most short videos need one every two to four seconds. On a faceless video with a single gameplay background, the interrupts cannot come from the footage, so they come from everything layered on top:

• A caption that changes style, colour or size on a key word.
• A zoom or a shake on the punchline.
• A sound effect: a whoosh, a ding, a record scratch.
• A new image, meme or panel appearing.
• A cut in the voice track, such as a deliberate half-second pause before a reveal.

The limit

Interrupts that arrive on a fixed metronome stop being interrupts. If a caption pops every two seconds exactly, the pop becomes the pattern and the brain filters it. Vary the spacing, and save the biggest change for the moment the story turns.

The second failure is interrupts that carry no information. A zoom on a filler word trains viewers to ignore zooms. Put the change on the word that matters.

How to check

Watch the retention curve for a drop that is not at the start. A cliff at four or five seconds almost always means the hook worked and nothing happened afterwards. Compare against retention rate and average view duration.

MakeViral times word-by-word captions to the voiceover audio rather than estimating, so the caption change lands on the spoken word. See the AI Shorts generator.