Exploring the bland side of AI-generated podcast content

When the Drink button was pressed it made an instant but highly detailed examination of the subject’s taste buds, a spectroscopic analysis of the subject’s metabolism and then sent tiny experimental signals down the neural pathways to the taste centres of the subject’s brain to see what was likely to go down well.

However, no one knew quite why it did this because it invariable delivered a cupful of liquid that was almost, but not quite, entirely unlike tea.

– Douglas Adams, The Hitchhiker’s Guide to the Galaxy

AI has been heralded as our salvation and our damnation. Chatbots can save us time by creating reams of seemingly sensible text, but when everyone has access to it, aren’t we just creating a race to the middle? I don’t think there’s a simple right answer, so let’s have a play and see what happens.

Letting a computer create a podcast

When given this prompt:

I’m interviewing James Thomson for my podcast. James is the creator of PCalc. Douglas Adams, the famous author, beta tested one of his applications called DragThing. What questions should I ask James that he hasn’t been asked before, so we can have a great podcast interview?

ChatGPT came back with these questions:

1. Can you tell us about any particularly challenging or interesting technical problems you had to solve while developing PCalc or DragThing?
2. How did Douglas Adams become involved in beta testing DragThing, and what was it like working with him?
3. Can you share any insights or advice for independent software developers just starting out in the field?
4. Can you tell us about any new or upcoming projects you’re working on, and what we can expect from them?
5. Can you speak to the evolution of PCalc and how it has changed over time since its initial release?
6. Can you share any interesting stories or anecdotes from your time developing and releasing PCalc and DragThing?
7. Can you talk about any design decisions or features you added to PCalc that you are particularly proud of?
8. Can you discuss the role of PCalc in the context of the broader calculator app market and how it has differentiated itself from competitors?
9. Can you talk about any significant changes you have made to the PCalc app over time?

ChatGPT also had thoughts on “how to upload a podcast to Spotify”, starting from the technical underpinnings of RSS to finding a suitable home for the audio.

To follow my experiments with AI – and to see the images, descriptions, and scripts the robots have generated for three entirely new podcasts – join Origin Circle. You’ll get my behind-the-scenes blog series, plus access to my catalogue of videos, eBooks and resources to help you save time and become a more effective podcaster. All for £9 a month, or £90 for the year.

Letting a computer handle the metadata

Capsho is a service that employs a robot to write, for each episode:

• Show notes
• Transcripts
• Social media captions
• Email newsletter content
• Blog posts
• LinkedIn articles
• YouTube descriptions

It also finds notable quotes from each episode. But it does a terrible job of fact-checkin.

The thing about formats and formulas

Having a structure is good. But it often gets filled with clichés like “Are you suffering from problem X? Then this episode is for you!”, or “Get the inside scoop”.

This will get better

With more understanding of sentiment, better contextual awareness, we’ll get to a place where AI really will be able to take on the job of creating the supporting material so we can focus on building human connection with our voices. We’re just not there yet.

Future hopes and predictions

I anticipate we’ll be able to feed our style and tone-of-voice into the computer and let it generate text that sounds like us. As they get better at understanding the emotional content of what we offer, then we can save time we’d be spending on show notes and LinkedIn articles, and use it to grow our shows… or spend time with other organic lifeforms!