Turn a Podcast into a Blog Post with n8n (Transcribe to Draft)

Every published episode is already a blog post, a set of show notes, and a week of social copy. It's just locked inside an hour of audio nobody has time to transcribe and rewrite. An n8n podcast to blog workflow unlocks it: drop in the episode, get back a draft article, show notes, and social variants, all from one transcription pass.

The workflows that rank for this tend to do one of three weak things. Some are blog-only and skip the show notes and social copy. Some hide the JSON behind "comment if you want it." Some lean on a paid summarizer API and never show the transcription step at all. The build below is vendor-honest end to end, transcribe with Whisper, write with a parse step, and stop at a draft for review instead of auto-publishing.

What one episode can become

A single audio file feeds several outputs in one run:

• A 1,000 to 1,500 word SEO blog draft
• Timestamped show notes with section headers
• Three to five social posts pulled from the best lines
• A title and meta description for the post
• Pull quotes ready for graphics
• A Sheets row logging the episode and its status

Generate the formats you publish. If show notes never go anywhere, drop that branch and save the tokens.

The podcast-to-blog pipeline

New episode (RSS / upload) → Transcribe (Whisper) → Structured writing prompt
       → Parse JSON → Draft to CMS/Doc → Review gate → Log episode

A new episode arrives. Whisper turns the audio into text. A single structured prompt returns every format as JSON. A parser splits it into fields. The article lands as a draft, a human approves, and the episode gets logged so it never runs twice.

1. Trigger on a new episode

The cleanest trigger is the show's RSS feed, an RSS node fires when a new item appears. For pre-publish drafts, a Webhook or a Google Drive trigger on an uploaded file works. Capture the audio URL and the episode title up front; both feed later steps.

2. Transcribe the audio

Send the audio to Whisper through the OpenAI node's audio transcription operation. Two things bite here. First, the file size limit, OpenAI's audio endpoint caps uploads, so a long episode needs splitting into chunks, each transcribed and joined. Second, the language, set it explicitly when the show isn't English, or the model occasionally guesses wrong on the first minute.

3. Write every format in one structured pass

Feed the transcript to an OpenAI Chat node with a prompt that returns a single JSON object: article, show_notes, social_posts, title, meta_description. One pass keeps the formats consistent with each other, the social posts quote the same lines the article highlights, because they came from the same call.

4. Draft, don't publish

Write the article as a draft, to a WordPress draft status, a Notion page, or a Google Doc, and email the approver a link. AI drafts from transcripts mangle names, misquote numbers, and invent product details just often enough that auto-publishing to a live blog is a real risk. The gate costs a few minutes and saves the occasional correction post.

5. Log the episode

Append a row: episode title, audio URL, draft link, status, date. The log is the dedupe guard, the RSS trigger can re-fire on feed quirks, and the log filter stops a second draft of an episode you already processed.

Implementation patterns worth copying

Pattern: reuse the transcript for clips

The transcription step is the expensive part, so spend it once. Store the transcript in a Sheet, and a separate workflow can pull clip timestamps from the same text without re-transcribing. Turn long videos into short clips with n8n covers that selection logic, it reads the same timestamped transcript this workflow produces.

Pattern: chunk long audio cleanly

Split on silence, not on a fixed byte count, so a chunk boundary doesn't land mid-word. An Execute Command node with FFmpeg can split on silence detection; then transcribe each chunk and concatenate the text in order before the writing step. Joining out of order scrambles the article. Tag each chunk with an index when you split, and sort by it before the join, so a parallel transcription that finishes out of sequence still reassembles correctly.

There's a cost angle worth flagging too. Whisper bills per minute, so a weekly hour-long show is a small, predictable line item, but a back catalog of two hundred episodes run all at once is not. Backfill in batches, cache every transcript, and the writing step becomes the only part you re-run when you tweak the prompt. That caching is what keeps an editorial rewrite from re-billing the whole archive.

Pattern: feed the distributor

Once the social posts are parsed, hand them to a distribution layer rather than posting inline. The Content Scheduler & Distributor reads a queue from Sheets and formats per platform, so the episode's social variants go out on a cadence instead of all at once.

n8n nodes you'll use most

Getting started

1. Connect the show's RSS feed or an upload trigger for pre-release drafts.
2. Transcribe with Whisper, splitting long files into sub-limit chunks.
3. Write one structured prompt that returns every format as a JSON object.
4. Add a Structured Output Parser, with a retry branch for bad JSON.
5. Write the article as a draft and email the approver a link.
6. Hand the parsed social posts to a distribution queue.
7. Log each episode so the feed never produces a duplicate draft.

For the rest of the repurposing chain, build an n8n content repurposing workflow shows the one-source-to-many fan-out the social posts flow through, and generate AI images with n8n covers the pull-quote graphics. When the article needs a cover, the Social Media Scheduler & Designer generates the caption-and-graphic pair.