Podcast Show Notes Generator

Turn podcast transcripts or audio files into a complete publishing package — SEO-optimized show notes, chapter timestamps, pull quotes, social clip suggestio...

Turn raw podcast transcripts or audio files into a complete publishing package. The agent acts as a podcast producer + content marketer, transforming a conversation into SEO show notes, chapter markers, pull quotes, social clip suggestions, newsletter blurbs, YouTube and Spotify descriptions, guest outreach copy, and tested title variations — everything a host needs to ship an episode.

Usage

Invoke this skill when you have a recorded episode (audio or transcript) and need to produce all the surrounding written assets in one pass.

Basic invocation:

Generate show notes from this transcript: [paste transcript] Here's an MP3 of episode 47 with Sarah Chen — give me the full publishing package Write SEO show notes, timestamps, and 5 social clips from this episode

With context:

Podcast: "Builder's Edge" — solo founders, weekly, 45-60 min episodes. Tone is conversational, audience is technical Guest is Marcus Lee, CTO of Vendr. Episode covers procurement automation We publish on Spotify, Apple, YouTube, and a Substack newsletter — produce assets for all four

The agent ingests the audio or transcript, extracts metadata, structures the output for every distribution channel, and returns a copy-paste-ready package.

How It Works

Step 1: Input Ingestion

The agent accepts two input modes:

Audio file → transcript:

Input: episode-47.mp3 (raw audio)

Pipeline:
  1. Run Whisper (large-v3 by default) for transcription
       whisper episode-47.mp3 --model large-v3 --output_format json --word_timestamps True
  2. Speaker diarization (pyannote.audio) to separate host/guest turns
  3. Word-level timestamps preserved for accurate chapter markers and clip cuts
  4. Punctuation restoration and capitalization for readability
  5. Light editorial pass — remove filler ("um", "uh", "like" when overused), preserve voice

Output: cleaned transcript with [00:14:32] timestamps and SPEAKER labels

Pre-existing transcript:

Input: transcript.txt or .vtt or Descript export

Pipeline:
  1. Detect format and normalize to internal structure (speaker, start_ts, end_ts, text)
  2. If timestamps absent, the agent estimates them from word count and average WPM
  3. If speakers unlabeled, the agent infers from context ("So, Marcus, tell me about...")

The agent will ask which mode to use and request file paths or pasted text. It handles transcripts up to ~3 hours of audio in a single pass.

Step 2: Episode Metadata Extraction

The agent reads the full transcript and extracts:

Field	How It's Extracted
Guest name	First introduction by host or guest self-intro
Guest title/company	Phrases like "I'm Marcus, CTO of Vendr" or host introduction
Guest bio (3-4 sentences)	Stitched from intro segment + any bio details mentioned mid-conversation
Episode title candidates	Top 3-5 themes by topic frequency and emotional weight
Key topics	NER + topic clustering on transcript chunks (typically 4-8 topics)
Notable mentions	Books, tools, companies, people referenced — for the resources section
Episode duration	From timestamps
Quotable density	Score of how many quotable lines exist (informs clip count)

Sample metadata output:

guest:
  name: "Marcus Lee"
  title: "Co-founder & CTO"
  company: "Vendr"
  bio: |
    Marcus Lee is the co-founder and CTO of Vendr, the SaaS buying platform that
    has saved companies over $400M on software contracts. Before Vendr, he led
    procurement engineering at HubSpot. He writes about negotiation leverage and
    procurement automation at marcuslee.com.
  links:
    twitter: "@marcuslee"
    linkedin: "linkedin.com/in/marcuslee"
    company: "vendr.com"

episode:
  duration: "52:14"
  topics:
    - "How procurement teams lose leverage in renewals"
    - "Why list price is a fiction in enterprise SaaS"
    - "Building negotiation playbooks from price data"
    - "The future of AI-assisted procurement"
  mentions:
    books: ["Never Split the Difference", "The Hard Thing About Hard Things"]
    tools: ["Vendr", "Coupa", "Ironclad"]
    companies: ["HubSpot", "Stripe", "Brex"]

Step 3: Timestamp Generation (Chapter Markers)

The agent generates chapter markers every 5-10 minutes based on topic shifts, not time intervals. It uses semantic chunking to find natural section boundaries.

Chapter generation logic:

1. Sliding window over transcript (3-minute windows, 1-minute overlap)
2. Embed each window, compute cosine distance to neighbors
3. Identify peaks in distance (topic shifts)
4. Snap to nearest natural pause (longer silence in audio, or paragraph break in transcript)
5. Generate 5-9 word chapter title that summarizes the segment
6. Reject chapters under 3 minutes (merge with neighbor)

00:00 — Intro & Marcus's path from HubSpot to Vendr
03:42 — Why list price is a complete fiction
09:18 — The renewal trap: how vendors capture you
16:05 — Building a negotiation playbook from price data
24:30 — When to walk away from a vendor
31:12 — AI agents in procurement: hype vs. reality
40:55 — Marcus's biggest negotiation loss (and what he learned)
48:20 — Rapid fire: tools, books, and what's next

These timestamps are formatted for YouTube (MM:SS — Title or HH:MM:SS — Title) and Spotify chapters (which use a different API but the same human-readable format).

Step 4: SEO Show Notes Structure

The agent produces show notes with deliberate SEO structure:

# [H1: Episode Title with Primary Keyword] — Episode 47 with [Guest Name]

[Lead paragraph: 2-3 sentences. Include primary keyword in first 100 words.
Hook the reader with the most surprising or valuable claim from the episode.]

## About the Guest [H2]

[Guest bio paragraph from Step 2 metadata.]

## What You'll Learn [H2]

- Why [counterintuitive insight #1]
- How to [actionable insight #2]
- The [framework/playbook] Marcus uses to [outcome]
- [Surprising data point or claim]
- When NOT to [common advice the guest disagrees with]

## Episode Highlights [H2]

[3-4 paragraph narrative summary, written for skimmers and SEO.
Use H3 subheaders for each major topic. Each H3 includes the timestamp.]

### List Price Is a Fiction (3:42) [H3]
[2-3 sentence summary of this segment, with one pull quote inline.]

### The Renewal Trap (9:18) [H3]
[2-3 sentence summary.]

[... etc for each chapter]

## Timestamps [H2]
[Full chapter list from Step 3]

## Resources Mentioned [H2]
- [Book: Never Split the Difference by Chris Voss]
- [Tool: Vendr — vendr.com]
- [Article referenced]

## Connect with [Guest] [H2]
- Twitter: [@handle]
- LinkedIn: [profile]
- Website: [url]

## Subscribe & Support [CTA H2]

If this episode helped you, the best thank-you is a 5-star review on Apple Podcasts —
it takes 30 seconds and is the #1 thing that helps new listeners find the show.
[Apple link] | [Spotify link] | [Newsletter signup]

SEO checks the agent runs:

• Primary keyword appears in: H1, first paragraph, one H2, meta description, image alt
• Long-tail keyword variations distributed naturally through H3s
• Internal links to 2-3 related past episodes (if catalog provided)
• External links to guest's primary owned property (their site, not just social)
• Schema markup hint: PodcastEpisode JSON-LD snippet provided as bonus

Step 5: Quote Extraction (Pull Quotes)

The agent surfaces 5-10 pull quotes optimized for social sharing. Selection criteria:

Criterion	What the Agent Looks For
Standalone clarity	Quote makes sense without surrounding context
Counterintuitive	Challenges a common belief — drives engagement
Quotable length	8-25 words ideal; under 280 chars for Twitter/X
Specific	Has a number, named example, or concrete claim
Voice-preserving	Sounds like the speaker, not generic
No filler	Edited to remove "you know", "I mean", false starts

Sample pull quote output:

QUOTE 1 (best for Twitter/X):
  "List price in enterprise SaaS is a fiction. Nobody pays it.
  The real price is whatever you can negotiate before they need
  to close the quarter."
  — Marcus Lee, Episode 47 [9:42]
  Format: text card with podcast cover, timestamp, attribution

QUOTE 2 (best for LinkedIn):
  "Procurement teams lose leverage the moment a department head
  has already picked the tool. The negotiation is over before
  it starts."
  — Marcus Lee [12:08]
  Format: text-only post, expand into 3-paragraph LinkedIn essay

[... 3-8 more quotes ranked by virality potential]

Step 6: Social Clip Suggestions (60-90 sec)

The agent identifies the best 3-5 audio/video moments to clip for short-form social (TikTok, Reels, Shorts, X video). Each clip suggestion includes hook + arc + CTA reasoning.

Clip selection logic:

1. Find moments with high quotable density (multiple quotable lines back-to-back)
2. Verify the segment has a self-contained narrative arc (setup → tension → resolution or surprise)
3. Check that it can be cut cleanly (starts and ends near pauses)
4. Score on retention markers: question opener, contradiction, specific number, named entity

Sample clip output:

CLIP 1 — "The $40K mistake" (78 seconds)
  IN:  [16:32] "So I was negotiating with this vendor..."
  OUT: [17:50] "...and that's how I learned never to skip the redlines."

  HOOK (first 3 sec): "I lost forty thousand dollars in one negotiation."
  ARC: setup (rookie negotiator) → mistake (skipped redlines) → consequence
       ($40K loss) → lesson (the rule he never breaks now)
  CTA: "Full story and the playbook on Builder's Edge episode 47."

  WHY THIS CLIP: Specific dollar figure as cold-open hook, vulnerable
  admission earns trust, ends with actionable rule. High completion rate
  predicted.

  PLATFORM FIT: TikTok ★★★★★, Reels ★★★★★, LinkedIn ★★★, X video ★★★★

CLIP 2 — "List price is a fiction" (62 seconds)
  IN:  [3:42]  ...
  OUT: [4:44]  ...
  [hook / arc / CTA / fit as above]

[... 1-3 more clips]

Step 7: Newsletter Blurb (3 Versions)

The agent writes three blurbs for different newsletter contexts:

Short (Twitter-style, ~50 words):

New episode: I sat down with Marcus Lee, CTO of Vendr, on why
list price in enterprise SaaS is a complete fiction — and the
playbook he uses to save companies millions on renewals.
[Listen: link]

Medium (newsletter section, ~120 words):

This week's Builder's Edge: Marcus Lee, co-founder of Vendr,
joins to break down the procurement game most founders don't
know they're losing.

We covered:
- Why list price is a fiction (and what to negotiate instead)
- The renewal trap that captures 80% of SaaS buyers
- His framework for walking away — and when to actually do it
- Whether AI agents will replace procurement (his honest take)

If you've ever signed a SaaS contract and wondered if you got
played, this one's for you.

[Listen on Apple] | [Spotify] | [YouTube]

Long (standalone newsletter issue, ~400 words):

[Full standalone email: hook anecdote → problem → guest credibility →
4-5 bullet teasers → 1-2 inline pull quotes → CTA → P.S. with bonus]

Step 8: YouTube Description Optimization

[First 150 chars — visible above fold, must hook + include keyword]
List price is a fiction. Marcus Lee, CTO of Vendr, breaks down the
procurement playbook that's saved companies $400M on SaaS contracts.

[Subscribe CTA]
► Subscribe for new episodes weekly: [link]

[3-paragraph description with primary + long-tail keywords woven naturally]

[Timestamps — YouTube auto-detects these and creates chapters when first
timestamp is 00:00 and format is consistent]
00:00 — Intro & Marcus's path from HubSpot to Vendr
03:42 — Why list price is a complete fiction
[... full chapter list from Step 3]

[Resources block — links open in new tab, drive SEO juice]
🔗 Vendr: https://vendr.com
🔗 Marcus on Twitter: https://twitter.com/marcuslee
🔗 Books mentioned: [Amazon affiliate links if applicable]

[Hashtag block — last line, max 15 hashtags, first 3 most relevant]
#Procurement #SaaS #Negotiation #StartupOps ...

[Channel boilerplate — keep consistent across episodes for branding]

Step 9: Spotify Chapter Optimization

Spotify uses different chapter mechanics than YouTube. The agent produces:

• Chapter list in Spotify's Anchor/Megaphone format (timestamps + titles, max 50 chars per title)
• Episode description (max 4000 chars, supports basic HTML — <a href>, <b>, <i>)
• Episode-level keywords (Spotify uses these for "More like this" recommendations — 5-10 single-word tags)

SPOTIFY CHAPTERS (titles trimmed to 50 chars max):
  00:00 Intro & Marcus's path to Vendr
  03:42 Why list price is a fiction
  09:18 The renewal trap explained
  16:05 Building a negotiation playbook
  24:30 When to walk away from a vendor
  31:12 AI agents in procurement
  40:55 Marcus's biggest negotiation loss
  48:20 Rapid fire round

SPOTIFY DESCRIPTION (HTML-friendly, under 4000 chars):
  [Lead hook]
  <b>What you'll learn:</b>
  <ul>
    <li>...</li>
  </ul>
  <b>Resources mentioned:</b>
  <a href="https://vendr.com">Vendr</a>
  ...

Step 10: Guest Outreach (Promotional Swipe Copy)

The agent writes copy that the guest can copy-paste to share the episode on their own channels. This dramatically increases the chance the guest actually promotes — friction kills sharing.

Sample swipe copy package:

=== FOR THE GUEST TO COPY-PASTE ===

[Twitter/X — 280 chars]
Talked with @builderspod about why list price in enterprise SaaS
is a fiction, and the playbook I use to save buyers millions on
renewals. If you've ever signed a SaaS contract and wondered if
you got played, this one's for you. [link]

[LinkedIn — 1,300 chars optimized]
[Full LinkedIn post written from guest's POV — first person, conversational,
ends with CTA to listen]

[Email signature line]
🎙️ New: I joined Builder's Edge to break down procurement leverage. Listen ›

[Slack/Discord — short, casual]
Hey — was on Builder's Edge this week talking SaaS procurement.
If anyone's prepping a renewal, the playbook segment at 16:05
might save you a chunk. [link]

[Instagram caption — 2,200 char limit]
[Caption with line breaks every 1-2 sentences for mobile readability,
3-5 hashtags, link in bio reference]

Step 11: Tag/Keyword Strategy

The agent produces a keyword strategy spanning all platforms:

PRIMARY KEYWORD (use in title, H1, first paragraph):
  "SaaS procurement"

SECONDARY KEYWORDS (use in H2/H3 and body):
  - enterprise software negotiation
  - vendor renewal strategy
  - procurement playbook
  - software contract negotiation

LONG-TAIL KEYWORDS (use in H3 subheaders and FAQ):
  - "how to negotiate SaaS renewals"
  - "what is list price in enterprise software"
  - "when to walk away from a SaaS vendor"

APPLE PODCASTS CATEGORY:
  Primary: Business > Entrepreneurship
  Secondary: Technology

SPOTIFY TAGS (single words, 5-10):
  procurement, saas, negotiation, startup, founder, b2b, enterprise

YOUTUBE TAGS (mix of broad + long-tail, max 500 chars):
  podcast, saas, procurement, negotiation, "how to negotiate saas",
  "marcus lee", vendr, "enterprise software", b2b, ...

INSTAGRAM/TIKTOK HASHTAGS (mix of size, 5-10 per post):
  #saas (large) #procurement (medium) #foundertips (medium)
  #enterprisesoftware (small/niche) #negotiation (large)

Step 12: Episode Title A/B Testing (5 Variations)

The agent produces 5 distinct title variations with reasoning, optimized for different goals:

TITLE A — Curiosity Gap
  "Why List Price in Enterprise SaaS Is a Complete Fiction"
  WHY: Strong contrarian claim. "Why" + counterintuitive = high CTR.
  RISK: SEO weak — no keyword for buyers searching "negotiate SaaS".
  BEST FOR: Twitter/X discovery, podcast app browse.

TITLE B — Outcome / Number
  "How Vendr Saved Companies $400M Negotiating SaaS Contracts"
  WHY: Specific dollar figure builds credibility. Brand mention helps SEO.
  RISK: Sounds like a Vendr ad — may suppress organic shares.
  BEST FOR: LinkedIn, business publications.

TITLE C — Question
  "Are You Getting Played on Your SaaS Renewals?"
  WHY: Direct address ("you") + emotional stakes (getting played).
  RISK: Clickbait-adjacent; may hurt trust if overused.
  BEST FOR: Newsletter subject line, Apple Podcasts browse.

TITLE D — Guest-First (SEO + Authority)
  "Marcus Lee (Vendr) on Procurement, Renewals, and SaaS Negotiation"
  WHY: Surfaces in searches for guest's name, brand, and topics.
  RISK: Boring — low CTR for cold audiences.
  BEST FOR: SEO, returning audiences who recognize the guest.

TITLE E — Hybrid (recommended)
  "Marcus Lee: List Price Is a Fiction (and How to Win SaaS Renewals)"
  WHY: Guest name for SEO + curiosity hook + outcome keyword. Balanced.
  RISK: Slightly long (~70 chars) — may truncate on mobile.
  BEST FOR: Default publish title across all platforms.

RECOMMENDATION: Ship Title E as the canonical title. Use Title A in
Twitter promotion, Title C as the newsletter subject line.

Worked Example

Input — transcript snippet (3 minutes of a 52-minute episode):

[00:09:18] HOST: So Marcus, you mentioned earlier this idea of the renewal
trap. Can you break that down for someone who's never heard it framed that way?

[00:09:30] MARCUS: Yeah, so here's the thing. Most companies, when they buy
SaaS, they think the negotiation happens at the start. You sign the deal,
you get your discount, you're feeling good. But the vendor knows something
you don't, which is that ninety percent of their gross margin comes from
renewals, not new logos. So they're playing a long game and you're playing
a short one.

[00:10:12] HOST: Wait, ninety percent?

[00:10:14] MARCUS: Ninety percent of gross margin in mature SaaS, yeah.
And this is why every contract they write has language that quietly tilts
the renewal in their favor. Auto-renewals, price escalators capped at
some number you didn't push back on, usage tiers that ratchet up but never
down. By the time you hit your second renewal, you're locked in. The
switching cost is too high, the data migration is too painful, and they
know it. So they raise prices fifteen, twenty percent and you pay it.

[00:11:02] HOST: And what do most procurement teams do wrong here?

[00:11:05] MARCUS: They wait. They wait until ninety days before renewal
to even start the conversation. By then it's over. The vendor has all
the leverage. The right time to negotiate your renewal is the day you
sign the original contract. You write the renewal terms in then, when
you still have leverage, when they still want to win the deal.

Output — partial publishing package for this segment:

chapter_marker:
  time: "09:18"
  title: "The renewal trap: how vendors capture you"

pull_quotes:
  - quote: "Ninety percent of gross margin in mature SaaS comes from
            renewals, not new logos. They're playing a long game and
            you're playing a short one."
    timestamp: "10:14"
    platforms: ["twitter", "linkedin"]

  - quote: "The right time to negotiate your renewal is the day you
            sign the original contract."
    timestamp: "11:18"
    platforms: ["twitter", "linkedin", "instagram-card"]

social_clip:
  title: "The 90% rule vendors don't want you to know"
  in: "09:30"
  out: "10:48"
  duration: 78
  hook: "Ninety percent of SaaS gross margin comes from renewals."
  arc: "claim → why it matters → how vendors weaponize it → what gets locked in"
  cta: "Full negotiation playbook in episode 47."
  fit: { tiktok: 5, reels: 5, linkedin: 4, x_video: 4 }

show_notes_section:
  h3: "The Renewal Trap (9:18)"
  body: |
    Most buyers think the negotiation happens at the start. Marcus
    explains why that's exactly backwards: 90% of mature SaaS gross
    margin comes from renewals, which means vendors quietly engineer
    every original contract to tilt the next one in their favor.
    Auto-renewals, uncapped escalators, ratcheting usage tiers — by
    the second renewal, switching costs lock you in and prices climb
    15-20% per cycle. The fix: negotiate renewal terms on day one,
    when you still have leverage.

newsletter_short_blurb_excerpt: |
  Vendors aren't making their money on the deal you just signed —
  they're making it on the renewal you haven't thought about yet.

youtube_chapter_line: "09:18 — The renewal trap: how vendors capture you"

spotify_chapter_line: "09:18 The renewal trap explained"

guest_swipe_tweet: |
  Talked with @builderspod about the renewal trap — why 90% of SaaS
  gross margin comes from renewals, and how vendors engineer your
  original contract to lock you in. If you've got a renewal in the
  next 12 months, this segment alone is worth your time. [link]

This pattern repeats across all 8 chapters of the episode, producing the full package.

Output

The agent delivers a single bundle containing:

• Episode metadata (guest, bio, links, topics, mentions)
• Chapter markers (timestamps with titles, formatted for YouTube and Spotify)
• SEO show notes (full markdown, ready to paste into hosting platform)
• 5-10 pull quotes (ranked by virality, formatted per platform)
• 3-5 social clip suggestions (timestamps, hook, arc, CTA, platform fit)
• 3 newsletter blurbs (short, medium, long)
• YouTube description (optimized first 150 chars + chapters + resources + tags)
• Spotify description + chapters + tags
• Guest swipe copy package (Twitter, LinkedIn, email sig, Slack, Instagram)
• Tag/keyword strategy (primary, secondary, long-tail per platform)
• 5 A/B title variations with reasoning and recommendation

Common Scenarios

"I have an MP3, give me everything"

Provide the audio file. The agent runs Whisper transcription, then produces the full package. Allow 3-5 minutes for a 60-min episode end-to-end.

"I already have a transcript from Descript/Otter/Riverside"

Paste or upload the transcript. Skip the audio step. The agent works directly from text.

"I only need timestamps and pull quotes"

Specify which assets you want. The agent produces only those, faster.

"Multi-episode batch — give me show notes for the last 10 episodes"

Provide all transcripts in a folder. The agent processes them in series and also produces a meta-analysis: top recurring themes, best guest by quotable density, suggested cross-promotion between episodes.

"Optimize an existing show notes page that's not ranking"

Provide the URL or current copy plus the target keyword. The agent audits structure, suggests rewrites, and produces an updated version.

"My audience is non-technical / B2C / a niche subculture"

Provide the audience description. The agent adjusts tone, vocabulary, and clip selection accordingly. Same skill, different voice.

Tips for Best Results

• Provide a show style guide if you have one (tone, banned words, brand voice) — the agent will match it
• Share 2-3 past episodes' show notes so the agent learns your house format
• For audio input, clean source audio (not heavily compressed phone recordings) yields better Whisper accuracy
• Mention distribution platforms explicitly — assets differ for Apple/Spotify vs. YouTube vs. Substack
• Specify affiliate link patterns if you use them, so the agent inserts them in the resources section
• Provide guest's social handles to avoid the agent guessing or omitting them
• For SEO-critical episodes, supply a target primary keyword — the agent will weave it through structure deliberately

When NOT to use

• Live unedited audio with significant overlap or crosstalk — Whisper diarization struggles; clean the audio first
• Episodes with NDA / confidential content — the agent will produce shareable assets that may surface things you intended to keep private; redact the transcript before passing it in
• Languages with weak Whisper support — for low-resource languages, transcription accuracy degrades; use a native-language transcription service first, then pass the cleaned transcript
• Highly produced narrative shows (Serial, Radiolab style) — the structure assumes interview/conversation format; documentary-style episodes need bespoke editorial, not templated show notes
• Episodes shorter than ~10 minutes — the chapter logic and clip selection assume enough material to segment; for short episodes, ask for "summary + 2 quotes" instead
• When you need fact-checking of guest claims — this skill produces marketing assets, not editorial verification; pair with a separate fact-check pass for sensitive topics
• When you need legal review — the agent will surface controversial pull quotes that may need legal sign-off if the guest made claims about third parties; do not auto-publish without review