Half the new podcasts
are not human

Somewhere in the last few months, the machines took the microphone. On the day the Podcast Index counted, only 44.6% of new shows looked like they were made by people. Human is now the minority.

The Podcast Index runs a public new feeds report. Every show that appears in the open podcast ecosystem, logged, with a flag on the ones that look automated. In April it showed something the industry had been muttering about for months and had not yet said out loud.

In a 24 hour window, 44.6% of new shows were rated likely legitimate. 45.7% were rated potentially AI. Across a nine day sample, Bloomberg reported roughly 39% of all new feeds as likely machine made.

Not the episodes. The shows. Whole podcasts, born by the thousand, with names and artwork and hosts who do not exist.

01Nobody is listening to them either

Let us be clear about what this is. It is not creativity. It is arbitrage.

An AI network can produce an episode for about a dollar. Thousands of episodes a week. They are aimed at high volume search terms, the stuff people type into a podcast app at midnight. Health. Wellness. Celebrity biographies. Then they sit there hoovering up whatever programmatic pennies fall out of the pipes.

One outfit, Inception Point AI, published 325 new shows in a single day. That was nearly one in five of every new podcast launched that day. Podnews counted more than 8,000 of its feeds still sitting in the index. Twenty new wellness shows went up in under 75 minutes. Codependency. Panic attacks. Gambling addiction. Hosted by voices with names and no bodies, one of which opens by explaining that it is an AI, so it can discuss your mental health "without judgment, agenda or projection".

The quality is exactly what you would expect. One Spanish language biography episode ends with the audio saying, in English, "I'm sorry, I can't assist with that." Nobody checked. Nobody listened. Because listening was never the point.

02The thing that actually breaks

The instinct is to shrug. Slop has always existed. Bad podcasts were free before AI turned up.

Here is the difference. Discovery is a shared space and it is finite. Charts, search results, category pages, "listeners also enjoyed". Those are the shelves. When a machine can stack a thousand boxes on the shelf overnight, the shelf stops meaning anything.

Apple now asks creators to disclose when a material part of a show is AI. Spreaker has started labelling. Both of those are humans trying to tag content that is being produced faster than humans can tag it. That race has one winner and it is not the person doing the tagging.

So the old plan, publish consistently and let the algorithm find you, is dead. Not dying. Dead. You cannot out publish a machine that works for a dollar an episode and never sleeps.

03Meanwhile, the audience got bigger

Now the part everyone misses in the panic.

Ofcom published its audio research on 20 May. 27% of UK adults now listen to a podcast every week, up from 23% a year earlier. 93% of us listen to some form of audio every week, and that rises to 98% of 16 to 34s. And despite every studio in Britain buying cameras, 39% still prefer their podcasts audio only, more than the 27% who prefer video.

The car tells the truest story. Just over a quarter of live radio listening now happens in vehicles, up 56% since 2006. There are 11.7 million adults for whom the car is the only place they listen at all, up from 2.6 million twenty years ago.

Think about what that means. A person, alone, in a metal box, giving you their ears for forty minutes. There is no more intimate media buy on earth. And the machines are flooding the market for it because it is worth flooding.

04Scarcity moved

For twenty years the scarce thing in audio was production. A mic, an edit, a feed, a distribution deal. If you could make the thing, you were ahead.

That is over. Making the thing is free now. Anyone can generate a polished, competent, forty minute conversation about your sector before lunch, and roughly four hundred people already have.

So the scarce thing is what a machine cannot fake. A person with something at stake. A view somebody could disagree with. A room, a story, a mistake, a laugh in the wrong place. Specificity. The stuff that comes from having actually been there.

Machines can make output. They cannot make a reason to listen.

05What we would actually do

Stop competing on volume. You will lose. Publish less and make each one unmissable.

Put a human name and a human face on it. Credit the host, the producer, the guest. Say who made it and why. In a feed full of ghosts, a byline is a trust signal.

Cite your sources out loud. The slop cannot, because it does not have any.

Own the route to the ear. Your list, your site, your clients, your event. If your show only exists because search sends it strangers, you are building on the exact ground that is being paved over.

And record in a real room with real people. The flaws are the proof of life. That is the whole product now.

Everybody else is about to get very good at making audio nobody asked for. Be the one people choose.