How to turn a podcast episode into short clips

A two-hour episode has maybe a dozen clips in it. The work of getting them out is not trimming; it is finding the six moments that belong to the same thought, scattered across the conversation, and putting them in an order that makes sense on their own.

This is the job an AI editor is most obviously good at, so it is worth being concrete about what “good” looks like. Below is a real clip cut in Narrative from a real episode, and what was asked for to get it. You can open it on its timeline with no account.

The episode and the clip

The source is one file: a 137-minute podcast episode, shot in 4K, in which the host walks through his weekly training routine. The clip is 70 seconds, vertical, captioned, on one thread of that conversation: the training week itself.

It is made of six cuts, and the timeline names each by what it is: the lifting plan (11 seconds), compound movements (15), strength and cardio (23), cardio intensities (13), intervals and leg day (6), the rest day (3). Read those in order and you have the clip’s outline. None of them are adjacent in the episode. That is the finding work, done.

How the moments get found

Narrative transcribes the footage when you add it and reads the transcript to find the moments a request describes. That is why a request that names a topic (“the part about his training week”) works, and why the six cuts land on sentence boundaries rather than mid-word. The captions come from the same transcript, so they are timed to the words rather than added afterwards.

What it cannot do is decide for you which thread the clip is about. A 137-minute episode has twenty possible clips in it. The one above exists because the brief chose one.

How to ask for it

A podcast-clip brief has three parts. The first is the one people leave out.

• The thread. “The part about his training week.” “Everything she says about pricing.” “The story about the first customer.” One thread per clip.
• The shape. Length and format. “Under 90 seconds, vertical, captioned.”
• The framing rule. One host on camera is simple. Two people means deciding: follow the speaker, or stack them. (There is a separate guide on that.)

Cut a vertical clip, under 90 seconds, on the part where he lays out his training week. Captions on. Start on the plan, end on the rest day.

Then review the cut the way you would review a human editor’s: does the clip make sense with no episode around it? Is the first sentence a sentence someone would stop scrolling for? If the answer to the second is no, say which moment should open it. “Start on the compound movements line” is a one-line revision, and it becomes a version you can restore.

Several clips from one episode

The footage is uploaded once. When the first clip is done, export it, then ask for the next one in the same chat: “Now a clip on what he says about sleep, same format.” The timeline becomes the new clip and the first one stays in the version history, one click from being restored. If you are producing a set for the week, the fastest order is to write the threads down first, then ask for them one at a time, exporting each before you ask for the next.

What to check before you post

• The first two seconds. A clip that opens on a half-thought loses the viewer before the caption loads.
• Names and numbers in the captions. Transcripts are good and not perfect; a proper noun or a figure is where they slip.
• The ending. The clip above ends on the rest day, a natural full stop. A clip that ends mid-list feels cut off, because it is.

See it on a real timeline

Open the clip. The six cuts are on the timeline with their names; scrub between them and you are looking at the outline above. Then ask it for a change, “drop the rest day and end on the intervals” is a good first one, and watch what moves. The first three prompts are free; the editor runs on desktop today.