Exploring the Impact of AI on Podcasting

What’s the role that AI plays in enhancing podcast content and brand engagement?

When it comes to podcast AI, we’ve got some questions. 

1. How is AI disrupting podcasting as a service industry? 
2. How effective is AI in podcast production? 
3. Will robots take our jobs?

Read this 100% human-made blog post to find out.

In case you haven’t noticed, AI is a thing right now. Natural Language Processing Tools such as ChatGPT and Bard, DALL·E 2, Midjourney, and many more, are fast becoming household names. Many view artificial intelligence as a threat to civilization. Others see it as a way to heal the sick. Still, others see it as a kind of Pandora’s Box full of poorly understood challenges to the status quo of various creative industries such as podcasting. One thing’s for certain: When your grandparents start talking about AI, it means the Pandora’s Box is wide open — and you’d better start dealing with what’s inside. 

Inside JAR’s A Tale of Two Podcasts experiment

At JAR Podcasts, we produce award-winning, high-performing podcasts for brands and organizations. In this capacity, it’s our job to stay on top of industry trends, like AI, to figure out how to help our clients’ podcasts succeed. That’s why we decided to create an experiment that would put AI podcasting tools to the test. We called it “A Tale of Two Podcasts” and presented our findings at this year’s Podcast Movement in Denver. 

You can view our presentation deck here. 

Can AI actually make a good podcast?

Here’s what we did: 

1. First, we challenged ourselves to learn as much as possible about contemporary AI tools in the media space. JAR’s very own “RED Team” immersed itself in tools like ChatGPT, DALL·E 2, Midjourney, Descript Voice Cloning, DeCIFR, Adobe Podcast Speech Enhancer, RunwayML, Melville, ElevenLabs.io and Vidyo.
2. We decided to create two podcasts: One entirely human-made with full creative freedom, and one largely made by AI tools, with minimum human intervention. 
3. We tested these two products on a group of unsuspecting listeners — without identifying which podcast was which — to find out what approach to content creation was ultimately going to be most successful. 

I’ll say this now that it’s all over — we were a bit nervous. There’s a great deal of hype around AI right now, some of it for good reason. In my heart of hearts, I know that human creativity matters, but I confess I had my doubts about whether others would feel the same. 

How we built an AI-generated podcast

Here’s how our experiment worked: 

Our RED Team prompted ChatGPT to “invent a client similar to AirBnb.” After a few false starts, ChatGPT invented a company called “Staycationary,” which we thought was a pretty good name. Next, we asked the AI to “list several topics for a podcast created by that client.” A number of travel-related topics were thrown out in about 25 seconds. Of those, our producers thought “Slow Travel” had the most potential. So we ran with that. ChatGPT was then prompted to “write a script for a host and two guests about slow travel for a (fake) company called Staycationary. The goal is to inspire people to travel in a way where they’d stay long enough to really get to know a place.” ChatGPT could not create a full-length script without help, so it had to be done in sections. After several tries, we had a script. 

Meanwhile, the human-powered podcast team was given the exact same mission and set loose on the project. Producer Tori Weldon drew on her personal experience backpacking as a young whippersnapper to weave a personal narrative. She interviewed author, writing instructor, and “slow travel aficionado,” Robin Rivers, on-site in a local nature park, drawing a physical connection to the theme of “getting to know a place.” Sound Designer/Editor Sam Seguin pulled out all the stops to create a rich sonic tapestry to support Tori’s writing. 

Listen to the human-generated podcast [here]. 

Where AI podcast production started to fall apart

Back with the RED Team, they pressed on with the AI podcast, embarking on a steep and somewhat time-consuming “prompt engineering” learning curve. After several tries, they managed to get the AI to suggest some hosts and come up with names for some guests. The first set of hosts that ChatGPT coughed up was unsuitable. Later attempts (after some more prompting) suggested a few interesting people with connections to travel, but who were not necessarily accessible for a “fake podcast.” For this reason, the decision was made to use a full AI treatment for both the host and guests. 

This is where things got weird. 

ChatGPT insisted that both guests should be women named Olivia. A real producer might have fought that choice to avoid listener confusion, but our producers went along with it, thinking “When in Rome…”

We used elevenlabs.io to create all three voices. We used “off-the-shelf” AI voices for the guests. There were a few false starts as our team weeded through sexist-sounding voice quality choices like “Playful,” “Alluring,” “Ideal for Adult Books,” and “Whiny Middle-Aged Female.” 

Eventually, we found some voices that sounded okay, if a little robotic. Further, we found the consistency of these voices varied as the podcast went on. 

Here’s how Guest #1, “Olivia Chambers,” sounded:

Not terrible, but a little stiff. This AI woman has clearly never “unplugged” in her life. 

Adding idioms in an attempt to make the guest sound more natural didn’t really help, either. It turns out, fake-relaxed is almost worse than fake-uptight. 

Here’s how Guest #2, “Olivia Bennet,” sounded: 

You can hear what the AI is trying to do, but the “uh” sounds too studied. 

For the host, things ended up getting complicated. In the end, we cloned the voice of our RED Team producer (and JAR’s Audience Growth Specialist) Matthew Stevens. We used about five minutes of a voice sample to create the AI host.

See if you can figure out which voice is the REAL Matthew…

We do see the potential of this voice-cloning tool for aspects of podcast production, such as small corrections or “pickups.” However, there are serious ethical implications here, as well as glitches within the process which require time and attention to fix. 

For instance, our AI host was inexplicably taken over by demons: 

In addition, we had serious structural challenges with the AI podcast. Our AI host kept repeatedly throwing to breaks. 

If we were to release this podcast on behalf of a client, all of these problems would have to be mitigated, which would take a considerable amount of time. 

You can listen to the AI-generated podcast [here].

Finally, we experimented with a new BETA-stage AI tool to create a full-length video for Youtube, supplemented by images “matched” to the content by the AI. The result was interesting, but would not meet typical brand standards for visual assets. The images pulled by the AI were at times suitable and at others wildly off-base. 

Still, it’s early days with AI, and so any conversation about its role in the podcasting industry needs to be tempered with the understanding that these tools — and our mastery of them — will continue to improve. Already, AI tools can make some aspects of the podcasting process more efficient. It can be useful for brainstorming titles and fixing small glitches in host performance. Overall, however, it generated a podcast that was structurally flawed, mediocre in terms of content, and ultimately not very engaging, at least to our ears. 

We also did a survey to see how audiences felt. Here’s what we found: 

• 96% of respondents said that the human-made podcast made them “feel most inspired to try slow travel.”
• 96% of respondents said that the human-made podcast “gave me the most positive feeling towards the sponsoring brand, Staycationary.”
• 96% of respondents enjoyed the human-made podcast the most.

Here’s the thing: The main advantage of AI is supposed to be “efficiency.” Leaving aside the fact that getting an AI podcast up to an acceptable level takes time and human intervention, there’s a bigger question we should be asking, and that is this:

Is the purpose of podcasting (or art, for that matter) efficiency? 

Absolutely not. 

The purpose of podcasting is connection. It’s expression. It’s thought leadership. It’s awareness. It’s brand lift. It’s entertainment. It’s dialogue. It’s social advancement. These are the things that human beings bring to the process, and that no amount of AI monkeys with keyboards will ever replace. 

5 key takeaways on our AI podcast experiment:

1. Human Touch Prevails: The experiment highlighted that despite AI’s advancements, human-created podcasts still resonate more effectively with audiences, offering inspiration and a positive brand connection.
2. AI Efficiency vs. Human Creativity: While AI tools offer efficiency, they still require significant human intervention to reach acceptable quality levels, underscoring the irreplaceable value of human creativity and connection in podcasting.
3. Challenges with AI Podcasting: AI-generated podcasts faced structural and content quality issues, indicating that current AI technology still struggles with producing engaging and coherent podcast content on its own.
4. Ethical and Technical Considerations: Voice cloning and other AI tools raise ethical concerns and technical challenges that need careful consideration and management, especially regarding authenticity and listener trust.
5. The Future Role of AI in Podcasting: AI holds potential for assisting in certain podcasting tasks like brainstorming and minor corrections, but its role needs to be critically assessed against the backdrop of podcasting’s core values like connection and expression.

Frequently asked questions about AI in podcasting

AI is changing the podcast industry quickly, but not always in the ways people expect. From scriptwriting and voice cloning to audience trust and content quality, these are some of the most common questions brands ask about the role of AI in podcasting.