From Podcast to Published: How to Turn Audio Into Articles

A one-hour podcast contains roughly 9,000 spoken words. Most of it never gets read by anyone, because nobody publishes a raw transcript and expects people to wade through it. That recording is full of usable material: arguments, examples, strong lines. The work is turning loose speech into something a reader will finish. Here is a workflow that does that reliably, whether the source is a podcast, a webinar, or a recorded sales call.

Start with a clean transcript

You cannot edit what you cannot read, so transcription comes first. Automatic transcription is good enough now that hand-typing is a waste of time, but the output still needs a pass.

A few things matter more than people expect:

• Speaker detection. A transcript that labels who said what is far more useful than an undifferentiated block. Diarisation (separating speakers) lets you attribute quotes correctly and keeps a two-person conversation readable. Austen can transcribe audio with speaker detection and draft directly from the result, which removes the manual labelling step.
• Timestamps. Keep them at least every paragraph. When you fact-check later, you will want to jump back to the exact moment something was said.
• A quick clean-up pass. Strip the worst filler ("um", "you know", false starts) but leave the phrasing alone for now. You are tidying, not rewriting.

Budget 15 to 20 minutes to read the cleaned transcript end to end before you do anything else. You are listening for what the conversation was actually about, which is rarely what the episode title claims.

Find the throughlines

A good conversation wanders. A good article does not. The hardest step is deciding what the piece is about, and a single recording usually contains two or three separate articles, not one.

Read the transcript and mark every passage where the speaker:

• makes a claim they clearly believe
• tells a specific story or names a real example
• disagrees with conventional wisdom
• explains a process step by step

Group those marks. You will start to see clusters: maybe ten minutes on hiring, fifteen on a pricing mistake, a tangent about a tool. Each cluster is a candidate article. Resist the urge to cram all of them into one piece. A focused 1,000-word article on the pricing mistake will outperform a 3,000-word "everything we discussed" post every time.

Pick the strongest cluster for your first article. Save the others.

Structure before you write

Speech is organised by association; writing is organised by argument. The gap between them is why straightened-out transcripts still read badly. Before writing a word of prose, build an outline from the cluster you chose.

A reliable shape for an interview-derived article:

1. A short hook that states the central claim or tension
2. The context the reader needs (often something the host asked that you can fold into a sentence)
3. The main argument, broken into two to four points, each anchored by a story or example from the transcript
4. The objection or nuance the speaker raised
5. A concrete takeaway

Now slot the marked passages into that outline. Some will move far from where they appeared in the conversation. That is fine. You are reordering thought into a sequence a reader can follow, and the speaker's best line about the ending might have come three minutes in.

Keep the speaker's voice

This is where most audio-to-text repurposing goes wrong. Editors smooth the transcript into generic prose and lose exactly what made the conversation worth publishing: the person's actual way of putting things.

Some rules that help:

• Keep their distinctive phrases. If someone says a bad hire "costs you twice, once when they're there and once when they leave", keep it close to verbatim. Do not paraphrase it into "bad hires are expensive".
• Preserve sentence rhythm where you can. Short, blunt speakers should read short and blunt on the page.
• Convert speech to readable prose, not formal prose. Remove the repetition and the false starts, but do not raise the register. A relaxed expert should not suddenly sound like a press release.
• Quote directly for the best lines, paraphrase the connective tissue. Use real quotation marks around the sentences worth preserving exactly, and write clean linking text around them.

If the article is written in the third person about the speaker, decide early how much you quote versus summarise, and stay consistent. Mixing the two carelessly is what makes repurposed content feel stitched together.

Fact-check before publishing

People are loose with numbers when they talk. They round, they misremember a date, they overstate a statistic for emphasis. None of that is a problem in conversation and all of it is a problem in print, because in writing it reads as a checked claim.

Go through the draft and flag every:

• statistic or percentage
• date, name, or company
• "studies show" or "research says" claim
• superlative ("the first", "the biggest", "nobody else does this")

Verify each against a source, or soften it to what you can actually support. If the speaker said "around 70 percent", and you cannot confirm the figure, either find the source or write "most" instead of inventing precision. This step protects both your credibility and the speaker's, since their name is on the quotes.

This matters more than it used to. Articles that get cited by AI assistants and search tend to be the specific, verifiable ones, and a single wrong statistic can undermine an otherwise strong piece.

Split one recording into several pieces

Remember those other clusters you set aside. A single good interview can become a small content run rather than one post:

• The flagship article from the strongest cluster
• A second article from the next cluster, published a week later
• A short how-to from any step-by-step section
• A handful of standalone quote cards or social posts pulled from the best lines
• A newsletter intro that links to the article

The economics of this are worth saying plainly: you already paid for the recording. Getting four pieces from one hour of audio changes the cost per published article dramatically, and the pieces reinforce each other because they share a voice and a point of view.

A practical takeaway: next time you finish a recording, do not file it. Transcribe it with speaker labels that same week, read it once to find the throughlines, and commit to shipping the single strongest article before you touch the others. One focused post from a conversation you already had beats a blank page every time, and the leftovers are still sitting in the transcript when you are ready for them. (If you want the transcription, speaker detection, and first draft handled in one pass, Austen can take the audio and produce a structured draft you then shape.)