B-roll gets recommended as a retention trick — cut away every few seconds and people stay. The mechanism is real but the prescription is wrong, because it optimises for change and ignores meaning. A cutaway that shows something the words can’t is worth several seconds of attention. A cutaway that shows a stock shot of a city skyline is a subtraction dressed as an addition.
What B-roll is actually doing
Two separate jobs, which is why generic advice about it conflicts with itself:
- Pacing. A visual change resets attention. Uninterrupted sameness — one framing, one subject, no movement — reads to a scrolling viewer as a video that has already finished. This is real, and it’s what the “cut every three seconds” advice is grasping at.
- Illustration. Showing the thing being described. This is the job B-roll is uniquely good at, and the one that justifies taking your face off screen.
The important asymmetry: pacing has cheaper solutions. A punch-in, a caption treatment change, a reframe, a gesture — all reset attention without spending your strongest asset. Faces are found by the visual system before anything else in a frame; cutting away from one has a cost, and it should buy something.
When a cutaway earns its place
| You’re saying | Cut away? | Why |
|---|---|---|
| “It looked like this” | Yes | The image is the content; words are the caption |
| A number, a comparison, a proportion | Yes | Quantities are grasped visually far faster than verbally |
| A place, a product, a person | Yes | Specificity that words spend a sentence establishing |
| An argument or an opinion | No | There’s nothing to see; your delivery is the content |
| An emotional beat | No | The face is doing the work; cutting away throws it away |
| Anything, but the clip feels static | Not with B-roll | Use a punch-in or a caption change — cheaper, keeps the face |
The generic-footage failure
Everyone recognises this instantly even if they can’t name it: the laptop on a desk, the skyline timelapse, the anonymous hands typing. The problem isn’t that it’s stock — it’s that it’s unspecific. It illustrates the category of thing you mentioned rather than the thing itself, and viewers read that as filler because it is.
The same test applies to generated footage, and it’s the only test that matters: does this shot show something no other shot could? Generation is genuinely good at the cases where the answer is yes — an abstraction with no photograph, a scale nobody can film, a scenario that would be a shoot. It produces the identical skyline-timelapse failure when used to fill time.
Making cutaways feel edited rather than inserted
- Cut on the word. The shot should land on the syllable it illustrates, not a beat after. Late cutaways feel like an afterthought because they are one.
- One to two seconds. Long enough to read, short enough that the viewer hasn’t started wondering where you went.
- Keep the captions running. The words continue over the cutaway — that continuity is what makes it read as one video rather than two.
- Match the grade. A cutaway with completely different colour and contrast announces itself as imported material.
- Return on a strong frame. Coming back to yourself mid-blink undoes the whole effect.
Cheaper ways to get the same pacing
- The punch-in. A single cut to a tighter crop of the same footage. Free, instant, and it keeps the face. The most under-used edit in short-form.
- A caption treatment change. Shifting the emphasis style at a section boundary marks a change of subject without touching the picture.
- A title card. Two words on screen for a second is a hard visual beat, and it doubles as structure — the same device that carries your opening hook.
- Moving the subject. Standing up, turning, picking something up. Change in the frame, no edit required.
Quick answers
Does B-roll improve retention?
A visual change every few seconds does — uninterrupted sameness reads as a video that has ended. But B-roll is only one way to produce that change, and the wrong B-roll is worse than none: an unrelated stock shot pulls attention away from what's being said rather than reinforcing it.
When should I cut away from myself?
When the thing you're describing is easier to see than to hear — a place, an object, a number, a before-and-after. If the cutaway doesn't show something the words can't, you're replacing a face with wallpaper, and faces hold attention better than wallpaper.
Does AI-generated B-roll look bad?
It looks bad when it's generic, which is the same failure mode as generic stock footage. It works when it shows something specific that would otherwise be impossible or expensive to film — an abstraction, a scale, a scenario. Judge each shot by whether it's doing work no other shot could.
How long should a cutaway be?
Long enough to be understood and no longer — usually one to two seconds in short-form. A cutaway that outstays the point it illustrates becomes the point, and the viewer starts wondering when you're coming back.
Find your cutaway moments in the transcript
Reading your own words is the fastest way to spot where a shot would help — and where a punch-in would do the same job for free.
Open Moonshot free