I cut a 62 minute interview last month where the host had sent me 40 minutes of b-roll. Coffee pouring. Hands on a laptop. A slow push on a plant near the window. It was nicely shot. I used 11 seconds of it.
That ratio is not a failure of the footage. It is the whole point. Most b-roll is filler that got promoted to "production value" because it fills space and looks busy. My rule is simpler and a little unfair: a b-roll shot has to do a job the talking head cannot do, or it does not go in.
Everything below is what that rule looks like in practice, after a few hundred episodes of putting it into other people's shows.
The three jobs b-roll is actually allowed to have
When I am watching an episode for the first time, I am not hunting for places to decorate. I am marking three specific problems. B-roll exists to solve those and nothing else.
- It shows something words are failing at. The guest describes a machine, a room, a chart, a product. Words are doing badly. Cut to the thing.
- It hides a repair. A jump in the audio, a removed tangent, a coughing fit, a section where the host talked over the guest for 20 seconds. B-roll is the bandage over the stitch.
- It resets attention at a real seam. Not every 90 seconds because a template said so. At the actual moment the conversation changes subject.
Notice what is not on that list. "The middle felt slow." "Two heads talking is boring." Those are edit problems, not coverage problems, and covering them with stock drone footage of a city at dusk does not fix a single one. It just moves the boredom behind a prettier window.
If you can delete a b-roll shot and no viewer would ever know it was gone, it was never doing a job. It was doing time.
Slow does not mean cinematic
The most common b-roll mistake I see from hosts who have started shooting their own inserts is length. Someone spends real effort on a shot, so they let it breathe for six seconds. Six seconds is an eternity. The viewer read the shot in under one.
My working numbers, and I do break them when the shot earns it:
- Explanatory insert, the thing being described: 2 to 3 seconds.
- Repair cover over a cut: as long as the repair needs, usually 1 to 2 seconds.
- Section reset at a topic change: 3 to 4 seconds, and only if there is music or a beat under it.
Anything longer than four seconds and you are no longer showing, you are pausing. There is a version of a podcast where pausing is the aesthetic. Almost no interview show is that version.
Length is also where retention dies quietly. You will not see one huge drop. You will see the graph sag every time the episode goes to a pretty shot of nothing, and those sags add up into a shape that looks like a show nobody finishes.
The best b-roll is usually already in the room
Here is the part that annoys people who just bought a gimbal. Most of the inserts I actually use in an episode were never shot as b-roll. They came from the cameras that were already rolling.
A wide from the second camera while the guest laughs. A tight profile held two seconds after the punchline lands. The host's hand going up mid sentence. That footage carries something staged coffee never will, which is that it happened during the conversation you are watching. It is synchronous. It belongs.
This is a big argument for shooting with more than one body. On multi-cam editing I get free inserts out of angles I was going to have anyway, and the cutting rhythm between them does the pacing work that fake b-roll was pretending to do. I wrote more about how that rhythm should actually feel in a piece on when to switch cameras and when to sit still, and the short version is that a well timed angle change beats a decorative insert almost every time.
If you shoot a single camera show, you are not locked out of this. You are just limited to the frame you have, which means your b-roll budget should go to screen recordings, photos, product shots, and the things the conversation genuinely points at.
Screenshots and graphics count, and they are better
Podcasters underrate the least glamorous insert there is. A screenshot. A chart. A photo of the thing.
When a guest says "we went from 12 people to 90 in a year", a number on screen for two seconds is worth more than any lifestyle shot you can buy. It is specific. It is unfakeable. It gives the viewer a reason to trust that the show did its homework.
This is where I spend most of the b-roll effort on shows I cut regularly:
- Product and website screenshots when a company is named.
- The actual chart when a statistic is quoted.
- A photo of the person or place being described.
- A short screen recording when the guest is walking through a tool.
None of this needs a camera. All of it needs someone who watched the episode and heard what was said. That is exactly why our editors sit through the whole thing rather than letting software cut on silence. A silence detector cannot know that the guest just referenced a specific chart at 34 minutes. A person can.
Where b-roll actually hurts you
Three failure modes, all common, all fixable.
It buries the reaction. Comedy and conversation live on faces. If the guest says something ridiculous and you cut away to a stock shot instead of holding on the host's expression, you threw away the funniest frame in the episode. On comedy podcast editing this is close to a cardinal sin. The reaction is the payoff. B-roll over a reaction is a joke told with your back turned.
It fights the graphics. If a name card is up, an insert is not. Pick one thing on screen at a time. I go into this more in the post on lower thirds that stay out of the way, but the principle is the same principle as b-roll: one job per moment.
It papers over problems that need real fixing. Covering a badly lit stretch with inserts does not make the episode look good, it makes it look evasive. Lighting problems in particular tend to be visible in every angle you cut to, which is why they are worth solving at the source. There is a whole list of them in the lighting mistakes that show up in the edit, and none of them are solved by more coverage.
What to actually shoot, if you are going to shoot
If a host asks me what inserts to send, my list is short and boring on purpose. Boring lists get used.
- Two or three establishing shots of the room, from angles the main cameras do not cover.
- Hands doing whatever the show is about, if the show is about a physical thing.
- The guest arriving or sitting down, 5 seconds, one take.
- Any object that gets discussed, shot flat and clean against whatever is nearby.
That is maybe 90 seconds of footage per episode, and I will use 15 to 25 seconds of it. That is a healthy ratio. If you are shooting 40 minutes of inserts for a weekly show, the time would be better spent on your audio, your questions, or literally anything else.
Clips are the one place the rules loosen
Vertical clips are a different animal. Nine by sixteen is a hostile frame for two talking heads, and the attention window is measured in the first second, not the first minute. Cutaways there do real work: they fill the top and bottom of the frame, they give the eye motion during a slow line, they keep a 45 second clip from looking like a static Zoom call.
So when I am building vertical clips out of an episode, I will use inserts at roughly triple the density I would use in the long form cut. Same test though. Each one still has to be doing a job. It is just that "stop this from looking static in the first two seconds" is a genuine job on a phone screen and not much of one on a TV.
The same logic runs through packaging. A thumbnail is a single frame doing the same work an insert does, which is telling the viewer what they are about to get. Our sister studio ClickTheory does that side for the shows we cut, and the good thumbnails are almost never the pretty frame. They are the specific one.
The test I use before every export
Before I send a first cut, I scrub the timeline and stop on each insert. One question: what breaks if this is gone?
If the answer is "nothing", it goes. If the answer is "the cut becomes visible" or "the viewer loses the thing being described", it stays. About a third of the inserts I place during a first pass do not survive that scrub, and the episode is always tighter for it.
That is the entire philosophy. B-roll is not seasoning you sprinkle on to make an edit feel produced. It is a tool with three uses, and using it for anything else costs you attention you cannot get back.
If you want a second set of eyes on how your show is using coverage, send me an episode at samy@pixelsamy.com and I will tell you honestly what I would cut. Most of the time the answer is less than you think, and the episode gets better for it.