How to automate podcast show notes with AI workflows

Podcasters spend an average of 3-4 hours per episode on post-production tasks, and writing show notes is one of the most tedious. If you publish weekly, that's 150+ hours per year on summaries, timestamps, and links — time better spent on content. Here's how to automate podcast show notes end-to-end using AI workflows that handle transcription, summarization, and publishing without manual intervention.

TL;DR

Feed your audio into a CodeWords workflow that transcribes via Whisper, generates structured show notes with timestamps and key takeaways, extracts guest links, and publishes directly to your CMS or hosting platform. Total hands-on time: under 2 minutes per episode.

Unlike generic AI automation posts, this guide shows real CodeWords workflows — not just theory. You'll get a production-ready pipeline that handles the entire show notes lifecycle from raw audio to published content.

Why manual show notes don't scale

The math is brutal. A 60-minute episode produces roughly 8,000-10,000 words of transcript. Distilling that into useful show notes means:

• Listening back (or skimming transcript) for key moments
• Writing a compelling summary that isn't just a recap
• Adding timestamps for navigation
• Extracting and verifying guest links and resources
• Formatting for your specific platform

At 2-3 episodes per week, this becomes a full part-time job. Most podcasters either skip show notes entirely (hurting discoverability) or produce thin, unhelpful ones.

Tools like Descript and Riverside offer basic transcription, but the summarization and formatting still falls on you. That's where workflow automation fills the gap.

Set up your transcription pipeline

The foundation of automated show notes is accurate transcription. CodeWords workflows can process audio files through OpenAI's Whisper API with speaker diarization, giving you a clean transcript with speaker labels.

Here's the workflow structure:

1. Trigger: New audio file uploaded to Google Drive or received via webhook
2. Transcribe: Send audio to Whisper with word_timestamps=true for precise timing
3. Diarize: Apply speaker identification to label who said what
4. Store: Save raw transcript to your Airtable base or database

The transcription step runs in an ephemeral sandbox, so you're not managing infrastructure. Audio files up to 25MB process directly; longer recordings get chunked automatically.

Generate structured show notes with LLMs

Raw transcripts aren't show notes. The next stage uses LLM orchestration to produce structured output. CodeWords gives you access to OpenAI, Anthropic, and Google Gemini without managing API keys.

Your summarization prompt should extract:

• Episode summary (150-200 words, conversational tone)
• Key timestamps (major topic transitions, memorable quotes)
• Guest bio (pulled from transcript introductions + web search)
• Resources mentioned (URLs verified via Firecrawl web scraping)
• Key takeaways (3-5 bullet points)

The LLM processes the full transcript in context, using the timestamps from Whisper to generate accurate time markers. No more scrubbing through audio to find "that part where they talked about pricing."

Verify links and enrich metadata

Show notes with dead links are worse than no show notes. Add a verification step that:

1. Extracts all URLs mentioned in conversation (the LLM catches verbal mentions like "check out our site at...")
2. Scrapes each URL to confirm it's live and grab the page title
3. Searches for guest profiles on LinkedIn, Twitter, and their company site
4. Formats everything into your show notes template

CodeWords' web scraping capabilities handle this without rate-limiting headaches. The AI Web Agent can even find resources that were mentioned by name but not by URL — "that blog post John wrote about serverless" becomes an actual link.

Publish to your podcast platform

The final step pushes formatted show notes to wherever you host. Common destinations:

• WordPress/Ghost: Publish via REST API with proper formatting
• Transistor/Buzzsprout: Update episode descriptions via their APIs
• Notion/Airtable: Store in a content database for review before publishing
• Slack: Send a draft to your team channel for quick approval

With CodeWords' 500+ integrations, you can hit multiple destinations in parallel. Publish to your hosting platform AND update your website AND post a teaser to social — all from one workflow trigger.

For teams that want a human-in-the-loop, add a Slack approval step: the workflow sends a formatted preview, and a thumbs-up emoji triggers publishing. Check out workflow templates for pre-built approval patterns.

Schedule and monitor your workflow

Set your workflow to trigger on a schedule that matches your publishing cadence. Options include:

• Event-driven: Fires when new audio lands in a specific folder
• Scheduled: Runs at a set time (useful if you batch-record)
• Manual: Trigger via Slack command or webhook when you're ready

Monitor success rates and processing times through CodeWords' built-in logging. If transcription fails (corrupted audio, unsupported format), the workflow alerts you via Slack or WhatsApp instead of silently failing.

For pricing details on workflow execution, check CodeWords pricing.

FAQs

How accurate is AI-generated transcription for show notes? Whisper achieves 95-98% accuracy on clear English audio. Speaker diarization adds another layer of usefulness. For heavy accents or poor audio quality, accuracy drops to ~90%, which the LLM summarization step smooths over since it's generating summaries, not verbatim quotes.

Can I customize the show notes format? Yes. Your summarization prompt defines the output structure. Want a specific template with branded sections? Include it in the prompt. The workflow outputs whatever format you specify — markdown, HTML, or plain text.

What about episodes with multiple guests? Speaker diarization handles multi-guest episodes. The workflow identifies distinct speakers and attributes quotes accordingly. You may need to map speaker labels to names in a post-processing step for first-time guests.

How long does the full pipeline take? Typically 2-4 minutes for a 60-minute episode. Transcription is the bottleneck (~1 minute for 60 minutes of audio). LLM summarization adds 15-30 seconds. Link verification runs in parallel, adding negligible time.

Start automating your podcast workflow

Stop spending hours on show notes that a well-designed workflow handles in minutes. CodeWords gives you the transcription, LLM access, web scraping, and integrations to build this entire pipeline without stitching together five different SaaS tools.

Browse automation templates for podcast workflows, or build your own from scratch with serverless Python functions. Your next episode's show notes could write themselves.