Pricing Dashboard Sign up
Recent
· 9 min read · MDisBetter

Turn a Podcast Episode into a Blog Post with AI Transcription

You have a 60-minute podcast episode and you want a publishable 1500-2000 word blog post out of it. Doing it by hand takes 2-3 hours of listening, scrubbing, drafting. The AI-assisted version takes 20-30 minutes total, including final editing. Here's the exact workflow with the prompts that produce drafts you'll actually publish — not bland AI mush.

The four-step workflow

The whole pipeline:

  1. Transcribe the episode to Markdown.
  2. Open the .md in Claude or ChatGPT.
  3. Run a structured prompt to draft the blog post.
  4. Editorial polish pass (10-15 minutes), publish.

That's it. Each step has details that matter, but at the high level there's nothing more complex than upload-then-prompt. Let's go through each.

Step 1: Transcribe the episode

Open /convert/audio-to-markdown and upload the episode audio file. Most podcast hosts let you download the MP3 from the episode page; if your podcast lives on Riverside or similar, the recorded audio is in your project folder.

Wait 1-3 minutes for processing. Download the resulting .md file. The output has speaker labels ("Speaker 1", "Speaker 2") and H2 headings at topic transitions.

Critical cleanup pass (3-5 minutes):

Step 2: Open in Claude or ChatGPT

Both work. Practical differences:

For files over ~50KB of Markdown, attach the .md as a file rather than pasting the content. Both interfaces handle attached Markdown cleanly.

Step 3: The blog post prompt

Generic "summarize this transcript" prompts produce generic mush. Specific prompts produce drafts you can publish. The template that works:

You are an editor at a publication that runs substantive long-form podcasts as feature articles.

Convert this podcast transcript into a publishable blog post draft.

Requirements:
- 1500-2000 words
- Clear narrative arc — not a chronological transcript
- Lead paragraph: a hook based on the most counterintuitive or sharp insight from the conversation. No "In this episode..." framing.
- Body: 4-6 sections, each with an H2 heading that names the substantive topic (not a question)
- Pull 3-5 verbatim quotes from the guest as block quotes, attributed by name
- Where the guest cites a specific number, study, framework, or example — preserve those concrete details
- Closing paragraph: the most important takeaway in 2-3 sentences. No "To learn more, listen to the full episode..." CTA
- Tone: substantive, direct, written for a smart non-expert reader. No hype words. No buzzwords. No exclamation points.
- Do not invent quotes or details not present in the transcript

This produces a draft that's about 80% ready. The remaining 20% is the editorial polish pass.

The variations that produce different drafts

Sometimes the standard prompt isn't the right framing. Three reliable variations:

The contrarian framing

When the episode contains a surprising or against-the-grain insight, the contrarian framing front-loads that:

Frame the article around the single most counterintuitive thing the guest said.
Lead with the conventional wisdom in 1-2 sentences. Then introduce the guest's contrary view.
Build the rest of the article as the case for the contrary view, using the guest's reasoning and examples.

The practical playbook framing

When the episode is heavy on actionable advice:

Frame the article as a practical playbook. Pull every concrete tactic, framework, or step the guest mentioned.
Structure as: brief intro, then numbered or H2-headed steps, each with the guest's reasoning and an example.
Close with what to do first.

The two-sided framing

When the episode is two experts disagreeing:

Frame the article around the disagreement between [Speaker A] and [Speaker B].
Intro: the question they were debating.
Body: alternating sections, each laying out one side's argument with quotes.
Close: where the conversation actually landed, or what would resolve the disagreement.

Choose the framing based on the actual content of the conversation, not a default. The framing decision is 5-10% of the time investment and 30-40% of the article quality difference.

Step 4: The editorial polish pass

The AI draft is a starting point, not a finished piece. The polish pass takes 10-15 minutes for a typical 1500-word draft. Things to check:

Voice and texture

Read the lead paragraph aloud. Does it sound like something a real person wrote, or does it have the AI's slightly rounded-edge cadence? Tighten any sentence that uses three soft words where one sharp word would do. Cut any sentence that explains something the next sentence already shows.

Quote attribution and accuracy

For every block quote, ctrl-F the exact phrase in the source transcript. The model usually gets quotes right, but occasionally lightly paraphrases — which is fine for content but matters for editorial integrity. If a quote is paraphrased, decide: tighten the quote to a verbatim version, or unquote it and run as the model's framing.

Concrete details

The best transcript-to-blog conversions preserve specific numbers, named tools, and concrete examples. The AI draft sometimes generalizes "the team grew from 3 to 50 in 18 months" into "the team scaled rapidly". Hunt for these flattenings and restore the specifics from the transcript.

Headings

The model tends to write generic H2 headings ("The Importance of X", "Why Y Matters"). Replace with substantive ones ("Why customer interviews stop being useful past 30", "The hiring filter that broke for Series B"). Specific headings rank better in search and read better to scrolling readers.

Closing line

The model's default closing tends to be a tepid summary. Often the best closing is a single quote from the guest that captures the article's argument. Surface one from the transcript and use it.

Sample input → output (abbreviated)

To make the workflow concrete: a typical chunk of input transcript looks like this after speaker renaming:

**Host:** So you've talked about how most early-stage companies are
thinking about hiring wrong. What specifically?

**Guest:** The mistake we made — and I see basically every founder make —
is hiring for what the company needs in 18 months instead of what it
needs right now. We had a head of growth at 12 employees. There was
literally nothing to grow yet. He sat there for four months trying
to find product-market fit through marketing channels, which is a
category error.

The model converts a section like this into something like:

## The hiring trap that breaks most early-stage teams

The most expensive hiring mistake [Guest] sees founders make isn't
bad candidates — it's hiring on the wrong horizon.

"We had a head of growth at 12 employees," he says. "There was
literally nothing to grow yet. He sat there for four months trying
to find product-market fit through marketing channels, which is a
category error."

Notice what changed: the H2 heading is substantive, not a question. The intro sentence is sharper than "the mistake is...". The block quote preserves the guest's actual phrasing. The paraphrased setup sentence does work the quote alone wouldn't.

Cross-feature: cited articles and references

If the episode references a study, a paper, or another article, you'll often want to link to those references in the blog post. Convert the cited URLs to Markdown via url-to-markdown for content creators and you have clean source material to fact-check against and link from. The combined workflow — audio for the conversation, URLs for the references, optional PDF for any cited reports — gives you a complete source pack for the article.

SEO setup for the post

The blog post is also the SEO asset. A few setup steps that matter:

Pair this with publishing the full transcript on the episode page itself (separate URL) and you have a two-page asset per episode: the curated blog post on the topic, and the full transcript for long-tail SEO. See your audio content is invisible to Google.

What about distribution beyond the blog post?

The same transcript and AI workflow extends to other formats — Twitter thread, LinkedIn essay, newsletter blurb, quote graphics. Run the same kind of structured prompt for each. We cover the full repurposing pipeline in podcast repurposing takes hours.

Common failure modes

Generic AI voice in the draft. Cause: prompt too vague. Fix: include explicit voice constraints (no hype words, no buzzwords, etc.) and run the prompt again. Also helps to give the model a 1-2 sentence example of the voice you want.

The article reads like a recap, not an article. Cause: model defaulted to chronological transcript order. Fix: explicit "don't go in transcript order — find the argument and structure around that" in the prompt.

Quotes don't match the transcript. Cause: model lightly paraphrasing. Fix: in the prompt, require verbatim quotes only. In the editorial pass, ctrl-F every quote.

The intro tries too hard. Cause: model's default "hook" patterns are stale. Fix: rewrite the intro yourself based on what struck you most when you listened. The model can rough out structure; the human-written intro tends to land harder.

Time budget

Realistic time per episode-to-blog-post workflow:

Total: 25-40 minutes per blog post. Compared to the 2-3 hour manual baseline, that's an 80-85% time reduction.

Recommendation

For weekly podcasts, integrate the workflow as a standard post-production step. The first few episodes through the workflow take longer; by episode three or four, the whole flow is muscle memory and consistent. The compounding effect on your blog's organic traffic over a year of consistent publishing is significant — especially if you also publish the full transcript on the episode page (the SEO play covered in audio content invisible to Google).

Frequently asked questions

Should I publish the AI draft as-is or always edit?
Always edit. The draft is 80% there; the last 20% is what makes it sound like a real publication and not generic content. The 10-15 minute editorial pass is non-negotiable for anything you put your name on. Without it, readers immediately register the AI texture and trust drops.
How do I avoid sounding the same across multiple episode-to-post conversions?
Vary the framing prompt. Some episodes get the contrarian framing, others the playbook framing, others a profile-style framing focused on the guest's story. The transcripts are different so outputs differ; varied framings keep the surface texture from feeling identical across your blog.
Is it ethical to publish AI-generated blog posts from podcast transcripts?
It's editorial-assisted writing, not pure AI-generated content. The substance is the guest's actual words; the AI is doing the structural conversion that you'd otherwise do manually. Most publications treat this the same way they'd treat a human transcriber producing a draft from notes — disclosure isn't typically required, but the edit pass and accuracy check are.