From One 60-Minute Interview to Show Notes, a LinkedIn Post, and a Tweet Thread: The Otter + Claude Workflow

Last March I recorded a 60-minute interview with a CMO at a Series B fintech. By the time the Zoom call ended, I had a clean transcript waiting in Otter. Two hours later, the show notes, a LinkedIn post pulling out her sharpest line, and an 11-tweet thread were all live on the channels. The interview itself took 60 minutes. The whole content pipeline took about 90 minutes after that.

That used to be a full week of work. Here's the pipeline I run now on every interview — no agency, no editor, no "we'll send the notes next week" delay.

The math that changed my mind

One interview, used to mean one blog post. Maybe a quote pulled out for Twitter if I felt generous. The rest of the conversation — and the best 80% of it usually lives in the "rest" — died in the recording.

I started counting: a typical 60-minute interview contains roughly 8,500 words. A solid LinkedIn post needs 150 to 250. A tweet thread of 10 to 12 tweets is 2,500 characters. A standard set of show notes is 600 to 1,000 words. Combined, those three assets use maybe 10% of what's in the transcript. The other 90% gets thrown away.

That's the actual opportunity. You're not creating three pieces of content from one. You're recovering three pieces of content you already recorded and forgot about.

The pipeline, end to end

Four steps. Each one has a single tool and a single decision point.

Step 1: Record clean audio. Step 2: Run Otter against the file. Step 3: Paste the transcript into Claude with a show-notes prompt. Step 4: Run two more Claude prompts for LinkedIn and Twitter.

Total time, in real numbers from my last three runs: 60 minutes recording, 12 minutes Otter, 40 minutes Claude + editing, 8 minutes posting across platforms. The Otter and Claude steps are where the time savings live.

Let's walk through it.

Step 1: Record the interview so the transcript doesn't fight you

The single biggest variable in transcription quality is the recording itself. Otter's accuracy is genuinely good — but only when the audio is decent. Bad audio in means garbage timestamps, misattributed speakers, and a transcript where you spend 20 minutes fixing "the" into "they" before you can paste anything into Claude.

Three things that move the needle:

• Two separate tracks, if you can. Zoom's "Record separate audio file for each participant" gives Otter speaker labels that actually work. Otherwise the transcript says "Speaker 1" and "Speaker 2" and you waste ten minutes renaming them.
• Headphones on the guest. Open-back headphones cut the echo that makes transcripts come back as "and then she said—" with three words missing. Ask for this in the calendar invite, not on the call.
• No music intro. If your podcast has a 20-second theme song, record the actual interview starting after the music fades. Otter doesn't know what a synth pad is. It tries anyway.

I also keep a 5-minute buffer at the start of the call for small talk. Otter transcribes it, and Claude will dutifully include "so how's the weather in Austin" in the show notes if you don't trim. The first edit I do on every transcript is deleting the warm-up.

Step 2: Let Otter do the typing

Otter's free tier gives you 300 transcription minutes a month, 30 minutes per conversation, and 3 lifetime file imports. That 30-minute cap is the catch — for anything longer than a half hour, you need a Pro plan or you need to upload the file directly. Upload is the workaround I use. Upload a 60-minute MP3 and Otter transcribes it as one file; the per-conversation limit only applies to live recordings, not imports.

When I upload, I do three things in Otter before I export:

1. Rename the speakers. Click on the speaker labels in the sidebar and assign real names. This carries into the export and saves Claude from a guessing game.
2. Highlight the three best quotes. Otter has a highlight feature. Tag the moments I want to remember. These become the source for the LinkedIn post.
3. Export as .txt, not .docx. Plain text pastes into Claude cleanly. The .docx export adds formatting Claude has to ignore.

The transcript arrives in my inbox about 10 minutes after I upload a 60-minute file. By the time I'm back from making coffee, the bottleneck has moved downstream.

Step 3: The show notes prompt

This is the Claude prompt I use for show notes. It produces 600 to 900 words, structured the way podcast platforms and Google like to see them.

You are a podcast show notes editor. Below is a verbatim transcript
from a 60-minute interview. The host is [NAME]; the guest is
[GUEST NAME, TITLE, COMPANY].

Write show notes in markdown with these sections, in this order:

## Episode summary
3 sentences. What the episode is about, who it's for, and the
one specific thing the listener will learn.

## Key takeaways
A bulleted list of 5-7 takeaways. Each takeaway is one sentence.
Each takeaway should be specific to this conversation, not generic
podcast advice.

## Notable quotes
3-5 direct quotes from the guest, verbatim. Include the timestamp
in [HH:MM] format pulled from the transcript. Choose quotes that
either make a sharp claim, share a number, or capture a memorable
frame.

## Chapters / Timestamps
A list of 5-8 chapter markers with timestamps, in HH:MM format.
Use natural topic shifts, not every 5 minutes.

## Guest bio
3 sentences. Pull from the transcript if the guest introduced
themselves. Otherwise write a placeholder I will fill in.

## Links
Anything the guest mentioned — books, tools, articles, their own
website. List them as a bulleted list at the end.

Rules:
- Use only information from the transcript. Do not invent quotes,
  numbers, or background facts.
- If the guest was vague, write "guest estimated" rather than
  inventing a precise number.
- Keep the tone professional, third person. No "I" or "we."
- Do not include small talk, intro pleasantries, or sponsor reads.

[PASTE TRANSCRIPT]

Three things to notice about this prompt:

• It tells Claude the role, the format, and the length, but it doesn't tell Claude what the show is about. That's intentional. If the prompt says "this is a podcast about B2B SaaS," Claude will over-fit every takeaway to B2B SaaS and miss the parts of the interview that were actually about something else.
• The "use only information from the transcript" rule cuts hallucination in half. Without it, Claude cheerfully invents the guest's college degree and a previous employer.
• The chapters are a separate section, not a running list. Most show note templates blend chapters into the takeaways, which makes both sections harder to read on a phone.

The first draft is usually 85% usable. The 15% I edit is almost always the timestamps (Claude occasionally invents a plausible-looking one) and the takeaways that came out generic. Fix those, paste into your CMS, ship it.

Step 4: LinkedIn and Twitter, in two more passes

The show notes are the source of truth. The LinkedIn post and the tweet thread should not be rewrites of the show notes. They're different formats with different jobs. A separate prompt for each is faster than asking Claude to do all three at once.

The LinkedIn post prompt. LinkedIn rewards one strong claim, one specific story, and one line the reader can quote. Here's the prompt:

You are a LinkedIn ghostwriter. Below is a podcast transcript.
Write a 200-word LinkedIn post in the first person from the
podcast host's perspective.

Structure:
- Line 1: A specific, surprising claim from the guest. Not
  "guests say X." The actual quote, in quotation marks.
- Lines 2-5: Why this claim is interesting. One paragraph,
  three sentences max.
- Line 6: One example or number from the conversation.
- Line 7: A question to the reader. Not "what do you think."
  Something specific: "What's the last campaign where X bit you?"

Rules:
- No hashtags in the body. Add 3-5 at the end.
- No "I recently had the pleasure of speaking with..." openers.
- No emoji.
- Plain text, no markdown.

The "no 'I recently had the pleasure of'" rule is the one that fixes the most LinkedIn posts. It's the default Claude reaches for when it doesn't know how to open, and it's the line that makes every corporate post sound identical.

The tweet thread prompt. A thread is a different beast. The first tweet has to do all the work alone — most readers never click "show this thread." The rest of the tweets have to be standalone enough that someone scrolling the replies still gets value from each one.

You are writing a tweet thread based on a podcast transcript.
Return 10-12 tweets, each one its own line, separated by "---".

Rules:
- Tweet 1 is a hook. It must work without the rest of the thread.
  Use a specific number, a counter-intuitive claim, or a question.
- Tweets 2-11 each make one point. One idea per tweet. End each
  tweet on a complete sentence, never a clause.
- Tweet 12 is a "if you found this useful, repost the first
  tweet" CTA. Keep it dry, not begging.
- Each tweet under 270 characters.
- No "🧵" emoji in tweet 1. Twitter's UI already says "Show this
  thread" — the emoji is filler.
- No hashtags in any tweet.
- Do not invent quotes or numbers. Only use what's in the transcript.

The "one idea per tweet" rule is the one that makes threads actually read. Without it, Claude packs three ideas into tweet 4 and the reader's eye slides off. With it, each tweet can be screenshotted and shared on its own, which is how threads spread.

What I don't do anymore

A few things I tried and dropped.

• Asking Claude for the social posts and the show notes in one pass. Output is always worse. Each format has different constraints, and the model averages them out.
• Posting the LinkedIn write-up to Twitter. Cross-posting reads lazy. The thread is built for the thread reader; the LinkedIn post is built for the LinkedIn reader. They share an interview, not a draft.
• Editing the Otter transcript by hand before pasting into Claude. I used to fix every "the" → "they" typo. Then I realized Claude reads the typo'd version fine, and my edit was adding 20 minutes for zero quality gain. I do still delete the warm-up, the sponsor read, and any "sorry, I was on mute" tangents. Everything else stays.
• A second human review. My co-host reads the LinkedIn post and flags anything that sounds like a brochure. That's the only review pass. Show notes get a self-review over coffee, no second pair of eyes.

The actual cost

For one interview, this is what runs through it:

• Otter Pro: I pay $8.33/month (annual plan). For one interview a week, that's the only fixed cost.
• Claude Pro: $20/month. Used for this workflow and a dozen other things.
• My time: 90 minutes after the recording.

If I billed that 90 minutes at even a junior freelance rate, the pipeline pays for itself the first time I use the LinkedIn post to land a client conversation. The tweet thread is the one that compounds. Six months in, the threads I posted in March 2025 are still being found by search and quoted in other people's threads. The show notes, meanwhile, are still pulling organic traffic to the podcast page.

The whole reason to record a podcast is the conversation. The whole reason to have a workflow like this is to make the conversation do more work than the recording did.

If you only run one improvement this quarter, run Otter + Claude on your next interview. Skip the "let me transcribe it myself this weekend" plan. Ship the show notes on Tuesday. Post the LinkedIn on Wednesday. Save the thread for Friday morning. Your future self, the one who's scrambling for content in six weeks, will thank you.