It looks like it's time to introduce a 'hosted by humans' badge for podcasts: Bloomberg reports that some 39% of new podcasts are now "likely" generated by AI, as per data from the Podcast Index that tracks the ecosystem.
This shouldn't come as a surprise really, as producing AI-hosted audio is now incredibly simple, and very fast. You can make your own AI podcast through a tool such as NotebookLM in just a few minutes — feed in a few sources to work from, and you get two AI hosts chatting through them. You can even interject with your own comments.
The AI podcast revolution
As part of the AI podcasts article, Bloomberg's Ashley Carman spoke to Jeanine Wright, the co-founder of Inception Point AI. Wright said that their company is now producing hundreds of shows a day with AI, is currently managing more than 10,000 different shows, and is looking to increase that output over time.
Head to the Inception Point AI website, and you'll see the company promises "the future of storytelling". The sales blurb says "we don't just make content, we craft characters" (which sounds a lot like the empty phrasing AI would come up with), and that there's a roster of AI 'characters' ready and waiting to host shows.
It's all rather baffling, but as Inception Point AI points out, this tech is cheap, quick, and easy to scale up. Not so long ago, a company promising "a full slate of AI talent" for producing podcasts would have come across as satire — but now it's turned into a very serious (and growing) business.
Meanwhile, data entrepreneur Adam Levy launched an Epstein Files podcast earlier this year — on the topic of Jeffrey Epstein — which was generated by AI from millions of source documents. What's more, the podcast was a big hit, and has passed the 2 million mark in terms of downloads.
This brings up all sorts of questions about the ethics and accuracy of AI-generated output, whether it's a text report, a video, or a podcast. How is the AI choosing what to include and what to leave out from its source material? How is it joining the dots between different pieces of information? Can we trust what we're hearing?
Questions aside, the Epstein Files shows that these AI productions can be hugely successful — though perhaps in this case it's more to do with the topic than how engaging or otherwise the AI hosts are. It doesn't look like AI podcasts are going away anytime soon.
Staying away from 'podslop'
I'm a keen podcast listener, via the excellent Pocket Casts app, and am currently signed up to 14 different podcasts (including the TechRadar podcast) — and I'm also behind on all of them. I stick podcasts on when I'm driving somewhere, walking somewhere, dozing off to sleep, and getting jobs done around the house. They're my regular companions.
The podcasts I listen to cover tech, movies, TV, music, current affairs, and general banter, and they're pretty varied in terms of style and approach. Something they all share, though, are presenters that are genuinely engaging and interesting to listen to, no matter what the topic is that's being covered.
Replacing those presenters with generic AI speech with all the edges smoothed off just doesn't work. At all. Yes, these AI podcast hosts sound very real, but when you've listened to a few of them, you get to recognize the tell-tale signs of AI — just like you do when looking at AI images or reading text that's been generated by AI.
These AI podcasts are produced by averaging out vast amounts of real audio, spoken by humans, which means everything — the script, the cadence, the changes in pitch, the little "ums" and "ahs", the back-and-forth of the hosts — comes off as bland and as something that's been built to a template.
There's no life here, no personality, no tangents, no mistakes, and no idiosyncrasy. It's worth restating (as it often seems to get missed) that AI has never seen a movie, heard a song, held a real conversation, or done any thinking beyond trying to organize words, sounds, and pixels into an algorithm-driven pattern.
AI content is increasingly being referred to as "slop", which seems a good term for it: mass produced, and of very little value. I'm going to stick to listening to real people having real conversations, no matter how many AI-made podcasts are pumped out. Now, don't get me started on variable speed playback...
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
