How to Automate Podcast Show Notes with Activepieces and Perplexity AI

This post is adapted from my YouTube video: Create Dynamic Podcast Content with Activepieces and Perplexity AI

Running a weekly podcast means staying on top of industry news, coordinating with co-hosts, and preparing structured show notes - all before you even hit record. For our trading card game podcast, "Stacking the Deck," I built an end-to-end automation in Activepieces that handles the entire content preparation pipeline.

Here's how it works, from the Saturday trigger to the formatted show notes email.

The Content Pipeline Overview

The workflow serves two purposes: it sources trending news from across the internet using AI, and it collects community-submitted stories from our audience. These two streams merge into a single, formatted show notes document that arrives in every co-host's inbox before recording day.

All of the programs in this workflow - Activepieces, FormBricks, Discord - are self-hosted on my own server. This is consistent with how I approach building infrastructure: own the tools, control the data.

Phase 1: Community Engagement Trigger

Every Sunday at 5:00 PM, the workflow kicks off with a scheduled trigger. The first action sends a Discord webhook to our community server - a friendly ping reminding members to submit any TCG and tabletop gaming news they've seen during the week.

The message includes a link to a FormBricks submission form (think Typeform, but open-source and self-hosted). Community members can submit news stories, links, and topics they want us to cover.

This is a two-channel approach to content sourcing: AI-powered internet research combined with genuine community participation. The community submissions add a human layer that pure AI research can't replicate. As a bonus, the form also collects marketing opt-in data - building a list of engaged audience members who are both consumers and potential clients.

Phase 2: Collaborative Document Creation

Next, the workflow creates a Google Doc with the podcast title and the current date as a differentiator (so each week gets its own fresh document). The initial content is just a placeholder - "insert copy here."

The workflow then updates permissions, granting each co-host editing access via their email addresses. An email goes out to everyone with the document link.

Here's the key timing mechanic: there's a 23-hour delay built into the workflow. During that window, each co-host adds their own news, topics, and discussion points into the shared Google Doc. This collaborative step ensures we're not solely relying on AI - every host contributes their perspective and expertise.

Phase 3: AI-Powered News Research

After the 23-hour delay, the workflow reads the Google Doc content (now populated with co-host contributions) and inserts it into a Google Sheet for logging. Then comes the AI layer.

The system sends a carefully structured prompt to Perplexity AI:

Find trending news from [current date] to seven days ago focused on trading card games from across the internet. Give me eight stories about different games, where two stories are about indie TCGs. Include as many details, data points, and drama as possible. Be sure to include the following stories: [Google Doc content from co-hosts]. Each story must be a paragraph long and feature a web URL. If the story doesn't have a verifiable URL, don't use it. Share why each story is noteworthy along with any relational insights.

Two critical elements make this prompt work. First, it incorporates the co-host submissions, ensuring their topics get researched and expanded by AI. Second, it includes a deduplication rule: "Do not mention any content from [previous week's Google Sheet row]." This prevents the AI from recycling last week's news, keeping the content fresh every episode.

Phase 4: Formatting and Delivery

The Perplexity output gets passed through GPT-4o for final formatting - converting the raw research into structured show notes with proper markdown formatting, segment breakdowns, and the specific dos and don'ts of our show format.

The workflow then updates the Google Sheet with the final output (for next week's deduplication reference), converts the markdown to HTML using GitHub-flavored markdown, and sends the finished show notes via email to all co-hosts.

The result is a professional, well-sourced email with headlines, story summaries, source URLs, and discussion prompts - ready for recording day.

The Deduplication System

One detail worth emphasizing: the Google Sheet serves as a rolling memory system. Each week, the current show notes are stored in a row. The following week, the Perplexity prompt explicitly excludes that row's content. This ensures the podcast never accidentally repeats last week's stories - a problem we hit early on before implementing this logic.

This is a lightweight version of the same principle behind RAG-powered content systems: giving AI memory and context to produce better outputs over time.

Technical Architecture

The complete flow in Activepieces:

Scheduled trigger - Sunday at 5:00 PM
Get current date - for document naming and prompt context
Discord webhook - community submission reminder
Create Google Doc - fresh document with date-stamped title
Update permissions - grant co-host access
Send email - document link to all co-hosts
23-hour delay - collaborative editing window
Read Google Doc - capture co-host contributions
Insert to Google Sheet - log for deduplication
Perplexity AI prompt - research trending news, incorporate submissions
Update Google Sheet row - store final content for next week
GPT-4o formatting - structure as show notes
Final row update - save formatted output
Markdown to HTML - convert for email rendering
Send email - deliver formatted show notes

Key Takeaways

Combine AI research with human curation. The best content pipelines blend automated research with community input and editorial judgment. Pure AI content feels sterile; pure manual research is too slow.

Build in deduplication. Any recurring content workflow needs a memory system to avoid repetition. A simple Google Sheet with a weekly row reference handles this elegantly.

Self-host when possible. Running Activepieces, FormBricks, and Discord on your own server gives you complete control over data and zero recurring SaaS costs. I covered this approach in detail in my multi-agent AI stack guide.

Use delays strategically. The 23-hour window isn't dead time - it's collaboration time. Not every step in a workflow needs to happen instantly. Building in human touchpoints makes the output better.

If you're interested in more Activepieces automation workflows, I've covered Twitter automation using a similar approach.

Edward Chalupa is a digital marketing specialist and founder of Whtnxt, a digital marketing and automation consultancy. Connect with him on LinkedIn or explore more at echalupa.com.