Podcast transcript turned into an emailed digest

Podcast transcript turned into an emailed digest

An n8n workflow that turns the transcript of a podcast episode into an email. You paste the transcript in, an OpenAI model
summarizes it, picks out the topics and writes questions to think about, and an agent looks up each topic on Wikipedia and
explains it. Gmail then sends you the summary, the topics and the questions as one HTML email.

What is inside

• workflow.json: the workflow as n8n exports it, nineteen nodes. Fifteen of them run and four are sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• The file carried the full transcript of one philosophy podcast episode, written by someone else. We removed it. `Podcast Episode
  Transcript holds the line Paste the episode transcript here.` now.
• The email went to a named person at n8n, and its greeting named that podcast. The address is <YOUR_EMAIL_ADDRESS> now, and the
  greeting says "this week's episode".
• The three model nodes had names that began with the old model, GPT 4 or GPT3.5. They are OpenAI Extract Model, OpenAI Research Model and OpenAI Summarize Model now, since the model changed.
• The models changed too. OpenAI's deprecations page, read on 12 November 2026, shows gpt-3.5-turbo and gpt-4 as shut down on
  23 October 2026, with gpt-5.6-terra and gpt-5.6-sol as the substitutes. The two GPT-3.5 nodes called gpt-3.5-turbo-16k, a
  model of the same line. OpenAI Extract Model calls gpt-5.6-sol, which replaces gpt-4. The other two call gpt-5.6-terra.
• The credential links are gone.
• Nothing else. The prompts, the chunk sizes, the schema and the formatting code are as the file had them.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Gmail account for n8n to send from, signed in through a Gmail credential.
• The transcript of an episode, as plain text. If you have only the audio, get a transcript first, from the podcast's site or
  a transcription tool.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Open Podcast Episode Transcript and replace the text between the backticks with your transcript. The text must not contain a
   backtick or the characters ${, which end or break that string.
3. Select an OpenAI credential in OpenAI Extract Model, OpenAI Research Model and OpenAI Summarize Model.
4. In Send Digest, replace <YOUR_EMAIL_ADDRESS> with the address that should get the digest, and select a Gmail credential. The
   brackets go.
5. Press Test workflow. The run ends when Gmail has sent the email, so check your inbox.

How a run goes

1. When clicking "Execute Workflow" starts the run by hand.
2. Podcast Episode Transcript is a Code node that returns the transcript as one field, transcript.
3. Workflow Input to JSON Document reads that field. Recursive Character Text Splitter cuts it into chunks of 6000 characters
   with 1000 characters of overlap.
4. Summarize Transcript is a summarization chain of the refine type. It reads the chunks in order and refines one running
   summary, using OpenAI Summarize Model at temperature 0.
5. Extract Topics & Questions asks OpenAI Extract Model for questions and further topics that fit the summary. A
   Structured Output Parser makes the reply a JSON object with questions and topics, each with a short reason.
6. Topics splits the topics list into one item for each topic.
7. Research & Explain Topics is an agent that runs once for each topic, with OpenAI Research Model and a Wikipedia tool. It is
   given the topic and the summary and writes a short explanation.
8. Format topic text & title turns the summary, the topics with their explanations and the questions into HTML.
9. Send Digest sends one email with the subject Podcast Digest, with three headings: Episode Summary, Topics Discussed and
   Questions to Ponder.

Change the look

• The email: the message in Send Digest, and the HTML in Format topic text & title.
• The summary's length and detail: the chunk size and overlap in Recursive Character Text Splitter.
• What the topics and questions are like: the prompt in Extract Topics & Questions.
• The models: the three OpenAI model nodes.
• Where the transcript comes from: replace Podcast Episode Transcript with a node that reads a file, a feed or a web page, and
  keep its output field named transcript.
• Where the digest goes: swap Send Digest for Slack, Notion or another node.

Good to know

• The refine chain makes one model call for each chunk, so a long episode means many calls in a row. A run on a long transcript
  takes a while.
• The agent calls Wikipedia for each topic and the model may pick a page that is not the one you meant. Read the explanations
  before you pass the digest on.
• A model writes the whole digest from the transcript. A name or a claim can come out wrong.
  Check anything you will quote.
• A transcript may be under copyright. Use one you may copy, and check a podcast's terms before you share a digest of it.
• The agent is an OpenAI Functions Agent at version 1 of the Agent node. Leave it at version 1 as the file has it, and test the
  workflow again before you move it to a newer Agent version.
• Where the data goes: the transcript goes to OpenAI in pieces, the topic searches go to Wikipedia, and the digest goes to your
  Gmail account and to the address you set.
• Billing: OpenAI bills one call for each chunk, one for the topics and questions, and one or more for each topic.