Last month I sat with a 74 minute episode where the best moment in the whole recording was technically off topic. The host asked a guest about hiring, the guest answered for maybe ninety seconds, then drifted into a story about a warehouse job he had at nineteen and a manager who fired him for reading during a shift. Four minutes. Nothing to do with hiring. Everything to do with why he hires the way he does.
I kept all four minutes. I cut eleven minutes elsewhere in that same episode without blinking.
That is the whole job, really. Editors love to talk about pacing as if it were a dial you turn, but pacing is just a long series of decisions about which detours earn their runtime. A tangent is not a defect. A tangent that goes nowhere is. Those are different problems and most people treat them as one.
The test I actually use
When I hit a detour in a timeline, I ask one question before anything else: does the listener know more about a person when this ends than when it started?
Not more about the topic. More about a person. Topic information is cheap and the internet is full of it. Character information is the reason anybody chooses one show over the forty others covering the same subject. The warehouse story taught me who that guest was. It made his hiring philosophy land harder three minutes later than any direct answer would have.
If the detour ends and I know nothing new about anyone, it is filler wearing a story costume. Out it goes.
There is a second question I ask right after, and it is less romantic: is the host still driving? A tangent with a host steering it, asking follow ups, reacting, is a scene. A tangent where the host has visibly checked out and is waiting for a gap to reclaim the floor is dead air with words in it. You can hear the difference. The room tone even changes, because nobody is leaning in.
Why software gets this exactly backwards
Silence detection has no concept of relevance. It has a concept of amplitude. So the tools that promise to cut your episode automatically will happily trim the two second beat before a guest says something vulnerable, because that beat looked like silence, and they will leave a nine minute circular ramble fully intact because it was loud and continuous the whole way through.
That inversion is why auto-cut episodes feel subtly wrong even when nobody can name what is off. The machine removed the pauses that carried meaning and kept the words that carried none.
Our editors watch the whole episode. All of it, start to finish, before a single cut is made. That is slower and it costs more and I have never once regretted it, because the tangent question cannot be answered by a waveform. It can only be answered by someone who was paying attention when the guest's voice changed.
The pause before an honest answer is content. The four minutes of throat clearing before a rehearsed one is not. No algorithm can tell those apart, because they look identical on a waveform.
The four kinds of tangent
After a few hundred episodes you start to see the same shapes recurring. I sort them roughly like this.
- The origin tangent. Someone explains why they believe the thing they believe. Almost always keep. This is the highest value material in most interview shows and hosts routinely cut it because it feels like a digression from the "real" question.
- The example tangent. A concrete story that illustrates an abstract point. Keep if the point was abstract. Cut if the point was already concrete, because now you have two examples of the same thing and the second one is just a repeat with different nouns.
- The status tangent. Someone lists credentials, past clients, or numbers to establish standing. Cut ruthlessly. Nobody has ever finished an episode because the guest mentioned a fourth accolade.
- The logistics tangent. Mic checks, scheduling talk, "should I start again", references to something that got cut. Always out. This is the only category where I do not think about it at all.
Most of the arguing between hosts and editors happens in the second category. My rule there is one example per idea, and the strongest one wins even if it came second. Reordering is allowed. A tangent does not have to stay where it landed.
Length is not the variable people think it is
Hosts ask me constantly whether an episode is too long. The honest answer is that length is almost never the problem. Density is. A tight 90 minute episode outperforms a bloated 40 minute one every time, and the drop off in analytics tracks moments, not minutes.
I watch where the flat spots are. If retention falls off a cliff at 22 minutes, the fix is at 20 minutes, and it is usually a tangent that turned inward and never came back out. Trim the flat spot and the whole back half of the graph lifts. Trimming evenly across the episode does nothing, which is why "cut it down by fifteen percent" is the least useful note a host can give.
The same principle governs how long a cold open should run. It is not about seconds. It is about whether the thing you put up front actually creates a question the listener wants answered.
When to keep a tangent that is objectively bad
Here is where I break my own rules, and I break them on purpose.
Sometimes a detour is clumsy, poorly told, and structurally useless, and I keep it anyway because it is the only moment in the episode where the host and guest sound like two humans who like each other. Podcasts are a parasocial medium. People subscribe to a relationship, not an information feed. If I sand off every rough edge, I get a very clean episode that nobody feels attached to.
So my honest position is this: I would rather ship an episode with one saggy human moment than a perfectly efficient one with no pulse. I have cut shows both ways and the polished, personality free version always gets better compliments and worse retention.
Wandering also affects how episodes travel. Shows that keep character moments produce far better vertical cuts, because a clip needs a person in it to work. That connects directly to how podcasts actually find new listeners, which is almost never through search and almost always through a forty second piece of somebody being interesting. When we build vertical clips for a show, the tangents are where we shop first. The on topic answers are usually too dependent on context to stand alone.
What I cut without mercy
To balance the sentimentality above, here is the list that gets no second look from me.
- Any story the guest has clearly told on three other podcasts. You can hear the grooves in it. It is a recording of a recording.
- The second half of any answer where the guest already made their point and is now looking for a landing.
- Cross talk that resolves into nothing. Two people restarting over each other for eight seconds and then agreeing.
- Anything that requires the listener to have heard a previous episode to understand, unless it takes under five seconds to reframe.
- Meta commentary about the podcast itself. Hosts love this. Listeners do not.
That last one is worth expanding. Talking about your own show inside your own show is the podcast equivalent of a band tuning on stage. It signals to the listener that the real thing has not started yet.
One more that is technical rather than editorial. If a tangent has a problem underneath it, a chair creak, a phone buzzing, a room that suddenly sounds different because someone leaned back, that is not a reason to cut it. That is a reason to fix it. Most of the audio issues that make people want to remove a good section are solvable, which is why we handle audio cleanup before we make final cut decisions, not after. Deciding on a section while it sounds bad means you are judging the wrong thing.
The habit that saves the most time
If you host, do this one thing and your editor will love you: when a tangent ends and you know it was gold, say so out loud. Just say "that stays in". When you know it was a dead end, say "we can lose that". Two seconds of your time.
We watch the whole episode anyway, so we will find the good stuff regardless. But your instinct in the moment is better than anyone's instinct in an edit bay three days later, because you were in the room. You felt the temperature change. That information is free and almost nobody records it.
The mirror habit is worth building too. If you catch yourself two minutes into something and cannot remember what question started it, stop and ask your guest a fresh one. Do not try to steer the tangent home. A clean restart cuts beautifully. A slow limp back to the original topic does not, and it forces the editor to choose between an awkward jump and three minutes of nothing.
What this looks like in practice
Our first cut lands in 48 hours, and it always includes the tangent calls we made, not just silence removal. If you disagree with a call, revisions come back the next working day, and on Studio those revisions are unlimited, so nobody has to ration their notes. That matters more than it sounds. Hosts who are counting revisions stop giving honest feedback, and then the show drifts toward whatever the editor guessed.
If you want the full picture of how a file becomes a finished episode, how it works walks through the whole sequence. And if you are packaging episodes for YouTube, the thumbnail carries more weight than the first three minutes ever will, which is why ClickTheory exists as a separate studio rather than something we bolt on.
The thing I would leave you with is that cutting a tangent is a claim. You are saying the show is better without this person being themselves for four minutes. Sometimes that claim is obviously correct. Often it is not, and the reflex to tighten everything is the reflex that makes a hundred shows sound like the same show.
Keep the ones that teach me who somebody is. Cut the ones that just fill the space between questions. If you are ever unsure which one you have, send it over and I will tell you what I think. That part is free.