B-roll in an ad has one job: support the claim the voiceover is making at that exact moment. A cutaway that illustrates the point buys you attention. A cutaway that exists because the edit felt static costs you the viewer, and B-roll is where ads get padded more than anywhere else. Keep cutaways to one to three seconds, tie every one to a specific line of script, and never ask B-roll to make the argument on its own.
What does a cutaway have to accomplish?
Every cutaway in an ad should be doing one of a small number of concrete jobs. If you cannot name which, cut it.
| Job | What it looks like | Typical length |
|---|---|---|
| Show what the voice is asserting | Voice says the strap tightens, we see the strap tighten | 1.5 to 2.5 seconds |
| Give a claim a beat to land | A texture or detail shot right after a number is stated | 1 to 1.5 seconds |
| Hide a cut in the main footage | A hands shot bridging two takes of the same speaker | 1 to 2 seconds |
| Change the visual register | Wide environment shot between two close shots | 1.5 to 2 seconds |
| Show scale or context | Product against a hand, a door, a room | 2 to 3 seconds |
| Buy time for a caption to be read | Static-ish shot while a long line of text sits on screen | 2 to 3 seconds |
Notice what is absent from that table: atmosphere, mood, brand feeling. Those are legitimate goals in a brand film and they are not goals in a 20 second performance ad, where every second is competing with a skip button or a thumb. An ad has room for a claim, evidence, an objection handled and an ask. There is no unassigned time for mood.
The practical test we apply in review: point at each cutaway and say the line of voiceover it is serving. If the answer is "it just looked good there", it goes. Our role definitions, and how B-roll differs from an evidence beat, are set out in the clip types documentation.
Why B-roll must never carry the claim
This is the structural rule, and breaking it is the most common reason a well-shot ad underperforms.
A cutaway is watched with about half the viewer's attention, because the voice is where the argument lives and the picture is confirmation. That is fine, it is how video works. The problem starts when the only place a piece of information appears is in a cutaway: a price rendered as a graphic over a texture shot, a comparison shown as a chart in a two second insert, a differentiator conveyed only by what the shot depicts.
Two things go wrong. Muted viewers get the information and interested viewers with sound on often do not consciously register it, because they are listening rather than reading. And when you rearrange the ad, which is the whole point of building variations, the argument breaks in ways nobody notices until performance drops.
So the discipline is: the claim lives in the voiceover and the captions. B-roll confirms it. If a fact only exists in a cutaway, either say it out loud or accept that most viewers will not receive it.
There is one exception worth conceding. Where the visual is the proof, a before and after, a stain lifting, a bend test, the shot does carry weight, and it deserves more than three seconds and a proper place in the structure. At that point it is not B-roll, it is your evidence beat, and it should be scripted as one.
How long should a cutaway be?
One to three seconds, with the majority landing around two. Longer than three seconds and a cutaway starts to feel like the ad has changed subject; shorter than one and it registers as a flicker rather than as information.
Pacing matters as much as individual length. On vertical placements a visual beat every two to three seconds is roughly what holds a scroll, so cutaways are the mechanism that produces that rhythm. On YouTube in-stream, where the median hook rate is 22 percent and the skip button lands at five seconds, the first five seconds should contain product and claim rather than a cutaway. Save the B-roll for after the skip, when you are talking to people who chose to stay. Both figures come from the platform medians in our 2026 fatigue benchmark.
A rough budget for a 20 second ad: two, maybe three cutaways, totalling four to six seconds. If you are at five cutaways in 20 seconds, the ad is a montage and the argument has nowhere to sit.
Generated versus filmed B-roll, with costs
B-roll is the one place where generated footage is straightforwardly good value, because the shots are short, abstract and usually contain no people. Under three seconds is where generated clips stop being identifiable in a feed, and that happens to be exactly the length a cutaway wants to be.
The costs, in credits, since that is what gets billed. A standard variation is 1 credit. Generated video is 2 credits per second, so a two second cutaway is 4 credits and a three second one is 6. Uploading footage, editing and re-exporting cost nothing.
| B-roll need | Generate or film | Reasoning |
|---|---|---|
| Abstract texture, material, light | Generate | 4 credits, reusable, no tells at 2 seconds |
| Environment or location you cannot access | Generate | Often the only realistic route |
| Hands performing an action | Film it, even on a phone | Hands are the first thing a viewer audits |
| Your actual product working | Film it | Generation approximates, and buyers notice |
| A face or a person speaking | Film it | We do not offer avatars, and they read as synthetic |
| Several products in one frame | Film it | Generated objects merge and labels smear |
| A before and after | Film it | This is evidence, not B-roll |
The arithmetic that decides it: 4 to 6 credits is expensive for a clip used once and cheap for a clip used across twelve variations. Generated B-roll is worth buying when it enters your library and stays there, which is why the rule is generate short, generate few, reuse constantly. If you generate a fresh three second insert for every ad, you are paying six times the cost of a variation for footage you will never see again.
Filmed B-roll remains cheaper than people assume. Hands, details and textures need no location, no talent and no release, and a phone in daylight clears the quality floor. An hour of deliberate detail shooting produces more usable cutaways than a generation budget will.
The padding test
When an ad feels slow, the instinct is to add cutaways, and that instinct is backwards. Padding is usually a symptom of a script with too much setup in it, and adding pictures over the setup makes the ad longer rather than tighter.
Three questions we run on a draft edit:
Can this cutaway be named against a line of script? If not, remove it and see whether anything is missing.
Does the ad still make its argument with every cutaway removed? It should. If removing B-roll destroys the argument, the argument was in the wrong place.
Is any cutaway over three seconds? If so, either it is evidence and should be scripted as such, or it is dead air with a picture on it.
An ad that passes those three usually runs shorter than the one you started with, which is a feature. The version of a 30 second cut that survives this process is often 22 seconds and performs better, because nothing in it is waiting.
How our product handles B-roll
Genyad is ours, so read this as disclosure. Footage you upload gets transcribed and tagged into a clip library, and each variation is built as a fresh script, shot selection, voiceover and caption set from that library rather than a re-cut of one timeline, so a two second hands shot gets reused across a set instead of being cut in by hand each time. Where a cutaway is missing entirely, generated B-roll fills the gap at 2 credits per second and the clip then joins the library, and the wider trade-offs of generative video apply: short, abstract and non-human works, hands and faces do not.
What it does not do: it will not tell you whether an ad is padded, because it has no predicted performance score and we do not think that number can be produced honestly before an ad has spend behind it. It also has no AI avatars or synthetic presenters, no static banner formats, no product URL import, no product feed rendering, and no direct publishing to Meta, TikTok or Google Ads.
Frequently asked questions
What is B-roll in a video ad?
It is the supporting footage cut in around your main shots: hands, details, textures, environments, product close-ups. In an ad its only job is to confirm the claim the voiceover is making at that moment, which is a narrower job than B-roll has in a brand film. Anything included for mood rather than for a specific line of script is padding.
How long should B-roll clips be in an ad?
One to three seconds, mostly around two, with two or three cutaways in a 20 second ad. Under a second it reads as a flicker, and over three seconds it feels like the ad changed subject. Keep cutaways out of the first five seconds on YouTube in-stream, where product and claim have to land before the skip button appears.
Is AI-generated B-roll good enough to use?
For short, abstract, non-human shots, yes. Under three seconds is where generated footage stops being identifiable in a feed, which is exactly the length a cutaway wants to be, and it costs 2 credits per second against 1 credit for a whole variation. Avoid it for hands, faces, your real product working, or frames with several products in them.
Can B-roll carry the message of an ad?
It should not. Information that exists only in a cutaway gets missed by viewers who are listening rather than reading, and it breaks when you rearrange the ad into new variations, so the claim belongs in the voiceover and the captions. The exception is a genuine visual proof such as a before and after, which is an evidence beat and deserves to be scripted as one.