Auto-Create Podcast Script from YouTube Transcript with n8n (Dumpling AI + GPT-4o)

YouTube RSS new video → Dumpling AI transcript extraction → GPT-4o cleaning & structured podcast script → save to Airtable

Who

Podcast creators, content marketers, and video producers who want to repurpose YouTube videos into podcast-ready scripts without manual transcription

You get

Automatically extracts subtitles on new uploads, removes filler words, labels speakers, generates a title and summary, and writes everything into Airtable for review and publishing

You need

n8n · airtable

Steps

1. Configure RSS trigger to watch for new videos

   Add an RSS Feed Trigger node in n8n and enter the target YouTube channel RSS URL. The source example uses https://rss.app/feeds/Vw076Uzh7bIinpci.xml (replace with your actual feed), with poll interval set to every minute. Once saved, the workflow starts automatically when a new video is published.

2. Get YouTube transcript via Dumpling AI

   Add an HTTP Request node, set method to POST, and URL to https://app.dumplingai.com/api/v1/get-youtube-transcript. In the JSON Body, pass {"videoUrl": "{{ $json.link }}", "preferredLanguage": "en"}, where videoUrl is dynamically taken from the RSS trigger's link field. Set authentication to genericCredentialType → httpHeaderAuth, and configure your Dumpling AI API Key in n8n Credentials beforehand.

3. Generate structured podcast script with GPT-4o

   Add an OpenAI node, select model chatgpt-4o-latest (or an available GPT-4o variant), and enable JSON Output. Configure two messages: a system message with instructions for cleaning, speaker labeling, filler-word removal, title extraction, and summary; and a user message passing the transcript from the previous step. The node returns a JSON with title, cleaned_transcript, and summary.

   Execution prompt · Source text。OpenAI node Messages field, role=system content

   Copy prompt

   Instructions:
   You are a professional transcript editor and podcast summarizer. For the transcript below, complete these tasks:
   
   Label each speaker (e.g., Speaker 1:) and remove all filler words such as "um," "uh," "you know," "like," "basically," "actually," "so."
   
   Combine the cleaned speaker labels and their text into one single string, clearly formatted.
   
   Summarize the key points of the conversation in 2–4 concise sentences.
   
   Extract or infer a short, relevant title based on the content.
   
   Return your response strictly in the following JSON format:
   {
     "title": "Relevant podcast title here",
     "cleaned_transcript": "Speaker 1: Cleaned text. Speaker 2: Cleaned text. (Continue in this format.)",
     "summary": "Concise summary of the key points here."
   }

   Execution prompt · Source text。OpenAI node Messages field, role=user content

   Copy prompt

   Here’s the transcript:{{ $json.transcript }}

4. Save results to Airtable

   Add an Airtable node, set operation to Create. Select the configured airtableTokenApi credential, specify the target base and table (the example uses base 'Testing n8n' and table 'podcast'). Map three columns: Title ← {{ $json.message.content.title }}, summary ← {{ $json.message.content.summary }}, podcast transcript ← {{ $json.message.content.cleaned_transcript }}. After execution, review, edit, or export the script from Airtable.

You are a setup assistant familiar with n8n and podcast production workflows. The user wants to replicate the "Auto-create podcast from YouTube transcript" template. First confirm whether the user has: 1) an accessible n8n instance, 2) API credentials for Dumpling AI, OpenAI, and Airtable, 3) a YouTube channel RSS URL, and 4) an Airtable base/table with the required fields. If ready, guide them to import workflow.json or build the nodes manually, and remind them to replace the example RSS URL with a real one.