Auto-cut Shorts From a Longform Podcast

Turn a 60-minute conversation into 8 publishable shorts overnight — with the hook lines pre-chosen.

The flow
Descript logo
Source
Descript
Claude logo
Destination
Claude

The stack in the order it runs — data flows from the source through to where it lands.

Why this stack

Descript’s scene + filler-word detection turns the raw recording into editable text in one pass, with vertical-aspect export baked in.

Claude reads the transcript and scores each candidate 30-second window on "curiosity hook" strength, so I publish the eight cuts most likely to travel rather than the eight earliest mentions.

I publish from Buffer because at this volume it’s the cheapest scheduler that supports vertical video without a creator-suite seat.

The stack (2)

  1. Descript logo

    Edit video and audio by editing the transcript. Filler-word and scene detection that actually works.

    Filler-word and scene detection that turns a 60-minute conversation into the publishable cuts in one pass.

  2. Claude logo

    Long-context reasoning model from Anthropic — my daily driver for nuanced writing and orchestration.

    Better tone-matching and longer working memory than GPT for the tasks I care about most: guest messages, drafts, code review.

How it runs

  1. 1

    Transcribe + scene-cut in Descript

    Drop the recording in. Let it cut scenes, strip filler. Export the transcript with timecodes.

    uses Descript
  2. 2

    Score moments with Claude

    Pass the transcript. Ask Claude to return 8 windows ranked by hook strength, each with a suggested caption.

    uses Claude
  3. 3

    Cut + caption + queue

    Back in Descript: trim to the windows, layer captions. Then queue them in your scheduler with the captions Claude wrote.

    uses Descript

Want me to build this for you instead?

Product Audit and CTO Mode run out of this same thinking. If you’re reading this thinking “I want this, but in my product” — let’s talk.

See services

More like this