TechRadar: 39% of new podcasts AI‑made

Podcasting has a spam problem now — but in audio form. The new wrinkle is that the spam can sound polished, multilingual, and weirdly plausible for a few minutes at a time. Over the past couple of weeks, reporting built on Podcast Index data showed that a startling share of brand-new podcast feeds look machine-made, not human-made. That matters because podcast apps were built for a world where making a show took effort, not a script prompt and a voice clone. ### Where did the 39% number come from? The number traces back to Podcast Index, the open podcast directory run by Dave Jones and Adam Curry. In a roughly nine-day window, the index logged 10,871 new feeds, and about 4,243 of them were flagged as likely AI-generated. That works out to 39% at the high end, with some follow-on coverage noting a running level closer to 35.4% on average. ### What does “likely AI-generated” mean here? It does not mean somebody proved every episode was fully synthetic. Basically, researchers and industry watchers are looking for patterns — cloned voices, formulaic titles, repeated structures, bulk publishing behavior, and huge networks of niche shows that appear all at once. So the figure is best read as a strong signal of industrial-scale automation, not a courtroom-grade count. (bloomberg.com) ### Who is making this stuff? One name keeps coming up — Inception Point AI. The company has been described as producing around 3,000 podcast episodes a week across thousands of shows, with reported costs around $1 per episode. That is the important shift. A podcast used to be constrained by recording time, editing time, and whether a human host actually had something to say. Now the bottleneck is mostly distribution. (tech.yahoo.com) ### Why flood podcast apps this way? Because the economics can work even if almost nobody listens. If you can generate thousands of episodes about celebrity biographies, wellness advice, local weather, or search-friendly how-to topics, a tiny trickle of ad revenue across a giant catalog can add up. It’s the old web-content-farm playbook, just moved into audio. (thedonut.co) ### Why is this worse for podcasts than it sounds? Podcast discovery is already messy. Most listeners find shows through app rankings, search, recommendations, clips, and cross-promotion. Dump a few thousand synthetic feeds into that system and you get clutter fast. Human-made indie shows have to compete not just with other creators, but with software that can publish at industrial speed in every niche and language. (podnews.net) ### Don’t platforms just label the AI shows? Some do, a little. Apple Podcasts requires disclosure when a significant portion of a show is AI-created, and Spreaker has started manually labeling some content. But manual labeling is the key phrase there — humans are trying to tag machine-scale output. That does not scale well. (digitalmusicnews.com) ### Is all AI audio “slop”? No — that’s the catch. AI tools can help real podcasters with cleanup, translation, editing, or even generating companion versions of a show. The problem is not “AI touched this.” The problem is bulk synthetic publishing with little editorial oversight, where the goal is volume first and usefulness maybe never. (digitalmusicnews.com) ### So what changes now? Expect more pressure for “made by humans” signals, stronger disclosure rules, and harsher filtering in podcast directories. But the deeper issue is that podcasting just lost one of its old protections — friction. When making a show becomes almost free, trust becomes the scarce thing. ### Bottom line? The 39% figure is not really a story about robots replacing hosts. It’s a story about distribution systems getting gamed. (digitalmusicnews.com) And once that happens, the hardest thing to recover is not content quality — it’s listener confidence.