Last month I sat with a two hour interview where the host and the guest laughed 61 times. I counted, because I was losing my mind. By the third pass I realised the episode was not long because of the talking. It was long because of the laughing. And the first editor who touched it had done the obvious thing: cut the laughs out to save time.
The result was horrible. Technically tight, forty minutes shorter, and completely dead. Two people who clearly liked each other now sounded like they were doing a deposition.
Laughter is not filler. It is the proof that the conversation is real. It is also the single easiest thing to ruin in an edit, because it looks like empty air on the waveform and it looks like wasted seconds on the timeline. I want to write down how I actually handle it, because I have got this wrong more times than I have got it right.
Why silence detection kills comedy first
Every automated tool I have tested treats a laugh as noise or as a gap, depending on how loud it is. A big open laugh registers as speech and survives. A small exhale through the nose, the one that means "yeah, I know exactly what you mean", registers as room tone and gets deleted.
That quiet laugh is the most valuable half second in the whole episode. It is the sound of two people being on the same page. Cut it and you get a version of the conversation where nobody agrees with anybody, they just take turns.
This is the main reason a human watches the whole episode on our side. Not because we are precious about craft, though I am, but because there is no rule you can write that tells a machine the difference between dead air and a beat where somebody is deciding whether to laugh. You have to hear it. Every episode we cut gets watched end to end by a person, and the laughs are one of the first things I look for on that first pass.
The three kinds of laugh, and what each one is worth
I sort them roughly like this while I watch.
- The reaction laugh. Short, immediate, on top of the other person's last word. This is glue. Keep it, always, even if it overlaps the next line. Especially if it overlaps.
- The release laugh. Comes after a tense or vulnerable stretch. It has a job: it tells the listener the danger has passed and they can relax. Keep the whole thing, including the sigh at the end of it.
- The stall laugh. Nobody said anything funny. Somebody is buying time, or covering for a question that landed badly. This one you can trim, and often should.
Most bad edits happen because the editor treats all three as the same object. Either they keep everything and the episode drags, or they cut everything and the episode goes cold. The skill is telling them apart in real time, and that only comes from actually listening to the conversation instead of scrubbing the waveform.
The test I use: if I cut this laugh, does the next line still make sense emotionally? If the answer is no, the laugh was doing structural work and it stays.
Trim the tail, never the front
Here is the practical rule I keep coming back to. The front of a laugh is sacred. The tail is negotiable.
The front carries the reaction, the timing, the surprise. If you clip even four frames off the start, the laugh stops feeling like a response and starts feeling like a sound effect. Listeners cannot articulate what is wrong, but they feel it. Something is off.
The tail is where the seconds hide. A good laugh has a decay: the peak, then the settle, then the "ah" as somebody catches their breath, then usually a false start before the next real sentence. You can take two thirds of that decay and nobody will ever notice, as long as you leave enough of it that the room does not slam shut.
My habit is to cut on the breath, not on the silence. Find the inhale that starts the next thought and land the cut just before it. The laugh keeps its shape, the pace picks up, and the listener never sees the seam. That is the same instinct behind everything I have written about why some conversations feel fast even when they are long, and honestly it is the same muscle.
Multi-cam makes this easier, and also more dangerous
On a two to four camera setup you can do something you cannot do on one camera: you can watch the laugh instead of just hearing it.
Cutting to the listener during a laugh is one of the highest value moves available in multi-cam editing. The person who is not talking is the one carrying the emotion. Their face is the punchline. If you stay on the speaker while the other person cracks up off screen, you have thrown away the best shot in the sequence.
The danger is overcutting. New editors discover this trick and then cut on every single laugh, which turns the episode into a tennis match. My rough ratio: cut to the reaction on the big ones and on the ones where the listener does something with their body. Stay wide on the small ones. If you are cutting more than once every eight seconds through a comedic stretch, you are cutting to show off, not to serve the scene.
On single-camera shows you lose that option, so the audio work has to carry more. That usually means being slightly more generous with laugh tails, because there is no visual payoff to cut away to.
What laughter does to your retention graph
This is where I will get argumentative.
Most people look at a retention graph and read the dips as "boring bit, cut it next time". Sometimes true. But I have watched enough graphs now to notice something else: the flat, high, stable stretches are almost never the smartest parts of an episode. They are the warm parts. They are where two people are enjoying each other.
Laughter is a retention signal in disguise. It tells the listener that sticking around is pleasant. When I go through the process of reading a podcast retention graph with a host, we usually end up marking the laugh clusters and finding that those sections hold better than the tightly edited information dumps around them.
So when somebody asks me to make an episode shorter, the laughs are the last place I go looking. I would rather cut a whole tangent than sand down twenty laughs across the hour. One removes a section. The other removes the personality.
Clips are where most people destroy the laugh
Vertical clips are the worst offender, and I include my own past work in that.
The temptation is obvious. You have got sixty seconds, the laugh takes four of them, so you cut the laugh and fit more words in. Then you wonder why the clip gets views but nothing sticks. I have written a whole post about why clips pull views but almost no subscribers, and the missing laugh is a bigger part of it than most people think.
A clip with the laugh left in tells a stranger two things at once: this person is interesting, and this room is fun to be in. A clip with the laugh stripped out only tells them the first one. The first one gets a view. The second one gets a subscriber.
When we build clips I look for moments where the laugh is inside the clip, not hanging off the end where it will get chopped by the platform. If the funny moment lands at second 55 of a 60 second cut, the clip is built wrong. Start later.
When to actually cut a laugh
I do not want to sound like laughter is untouchable. It is not. Here is when it goes.
- Nervous laughter after a bad question. It makes the guest look uncomfortable and it makes the host look like they know it. Trim it, or cut the question.
- Courtesy laughs at your own joke. If the host laughs and the guest does not, that gap is loud. Shorten it hard.
- Repeated laughs at the same beat. The third time the same running joke gets a laugh, it stops being funny and starts being an in-joke the audience is outside of.
- Laughs over important information. If somebody says a number or a name under a laugh and it is unintelligible, the information wins.
That last one matters more than you think if the episode is going to a sponsor or a brand partner. There is a real difference between an episode that sounds warm and an episode that sounds unserious, and that line lives almost entirely in how you handle the third and fourth laugh of a cluster. Anyone thinking about the business side should read what I put together on what sponsors actually look at before they pay, because tone consistency shows up there more than production polish does.
How I brief this on a real episode
When a host sends me a first episode, I ask one question about humour: is this a funny show, or a serious show that occasionally laughs? The answer changes the whole approach.
Funny show, I protect every reaction laugh and cut tangents instead. Serious show, I keep the release laughs and tighten everything else, because those two or three laughs across the hour are doing heavy lifting exactly because they are rare.
We turn a first cut around in 48 hours and revisions come back the next working day, so this is a thing hosts can actually iterate on. Send me a note that says "you cut too many of my laughs in the second half" and that is a fixable, specific piece of direction. It is one of the most common notes I get, and it is always right. Editors under time pressure default to tightening, and tightening eats laughter first.
If you are building the video side of the show at the same time, the same principle carries across. Warmth reads on camera even more than it reads in audio, which is part of why I keep telling people that growing a podcast with video is less about gear than about not editing the humanity out of the room.
The version I keep coming back to
The episode I mentioned at the top, the one with 61 laughs. We recut it, kept 54 of them, trimmed the tails, cut two long tangents instead. It came out eleven minutes shorter than the bad version and it felt twice as fast.
That is the whole thing in one sentence. Cutting laughter makes an episode shorter. Keeping laughter makes it feel shorter. Those are not the same goal, and most editors are optimising for the wrong one.
If you are sitting on an episode that feels flat and you cannot work out why, go back and check what happened to the laughs. Nine times out of ten somebody tidied them away. I am at samy@pixelsamy.com if you want a second pair of ears on one.