There’s money to be made in podcasting. It’s getting harder, though — at least for some podcasters.
But while you may listen to podcasts, you might not be as familiar with the business side of podcasting. I am.
On the October 9 episode of Freakonomics Radio (episode #690), host Stephen Dubner welcomed Jenna Weiss-Berman — the executive producer of Good Hang with Amy Poehler, the CEO of Paper Kite Audio, and the cofounder of Pineapple Street Studios. He also had on Jeanine Wright, the cofounder of a company now called Liquid Studios AI (originally called Inception Point AI), which focuses on the business side of podcasting.
You’ve likely heard podcast ads, so you know that ad sales are a key part of the podcast industry. Not surprisingly, though, the largest shows get a disproportionately large amount of the ad dollars. The technology to serve the ads is limited. And scaling host-read ads is hard — there are many pain points between getting hosts to approve ads, getting product in their hands, getting them ad copy to read, getting them to record the ads properly, and then getting the ads in their episodes. Programmatic and pre-produced ads are a lot easier to scale, but a lot crappier to listen to, and thus a lot easier to ignore, and thus a lot less effective.
I sold podcast ads, and/or led teams that sold podcast ads, for about fifteen years. I started indie, then cofounded podcast ad sales company Midroll, which was eventually acquired by Scripps. (Scripps bought Stitcher a year later; Midroll renamed the whole business Stitcher, and eventually SiriusXM bought it, and later chopped it up for parts.) I left before the SiriusXM acquisition and went to ART19, a company focused on the technical side of podcasting (backend hosting infrastructure and dynamic ad insertion), which wanted to get into ad sales and originals. ART19 was the company Amazon acquired that led to my eventually working at Wondery for a year and a half.
First, some disclosures: I negotiated multiple deals with Stephen Dubner and Freakonomics when I signed him to Stitcher many years back, and he had the office next to mine. I negotiated multiple deals with Jenna Weiss-Berman when she was at Pineapple, including my favorite — a deal for the seminal podcast Missing Richard Simmons. And Jeanine Wright was eventually hired as my boss at Wondery after both Wondery and the company I was at, ART19, had been acquired by Amazon; I stayed at Amazon six months longer than I would have otherwise just because of how much I liked Jeanine. And I once ran an improv for enterprise workshop for her team, when the company she runs now was still called Inception Point AI.
So, yeah. I know many people in this story. Not Amy Poehler, though. I never met her.
The money in podcasting
Dubner also spoke to Ben Davis, his agent at WME (with whom I’ve done a slew of deals). As Ben references, an area that sets podcasting apart from, say, classic television deals, is this: On the ad-supported side, in both cases, the money comes from — surprise! — advertising. But in podcasting, from the very beginning, most deals were structured where the podcaster(s) got a cut of the ad revenue. So where a classic TV deal might be: We’ll pay you $250,000 an episode for this 22-episode season, podcasting deals usually take the form of we’ll pay you 70% of all the ad revenue this show generates. And very often, those deals get more nuanced: We’ll guarantee you the greater of this sum of money at a minimum or 70% of all the ad revenue — derisking the podcast host, and … *en*risking the company providing that minimum guarantee.
The episode covers the reality that the vast majority of podcasts make most of their money from ads. In the tech nerd podcast world, that’s still often true, but not always. Loyal fanbases often subscribe to “premium” versions of shows, which — at least for shows whose size means they can’t command as much ad revenue — can often mean memberships become the lion’s share of revenue.
So podcasters typically want as many ads as they can get in their show without alienating listeners. More ads is more money on which they can earn their revenue share. And sure, the common minimum guarantees that the biggest podcasters can command certainly help, but few creators are satisfied with just the minimum guarantee; they want to exceed it.
And it’s worth noting: The vast majority of podcasts that exist don’t get any ad sales at all. And of those that do get ads, most don’t have the power or leverage to score minimum guarantees. So the majority of money-earning podcasters are even more dependent on ads; if there are no ads in an episode, that episode isn’t directly generating revenue.
As Ben explains, most podcast ads are sold on a CPM basis, which means the advertiser spends some dollar figure per thousand downloads. (The “M” is for “mille” which means “thousand” and confuses people. Fun!) Industry standards in podcasting are around an $18 CPM for 30-second pre-rolls, and a $25 CPM for 60-second mid-rolls. So if a show can offer an advertiser 100,000 downloads, that spot might cost $2,500 (100,000 divided by 1000, multiplied by $25.) Over time, CPMs tend to trend lower — except for shows with incredibly lucrative audiences or incredibly famous hosts.
The ad technology
As someone with many years of podcast ad sales experience, it pained me when Dubner played a clip of Will Page, the former chief economist at Spotify, describing audience measurement in podcasts as “the wild west.” Podcasting, and podcast ad tech, has been described as “the wild west” for as long as there have been ads in podcasts.
“The wild west” label was the bane of my existence when I sold podcast ads full-time. Here’s why it gets thrown around:
Podcast listening essentially happens offline. Podcast hosting companies (Megaphone, ART19, etc) aren’t streaming; they’re serving up audio files for download. Your app (Apple Podcasts, Overcast, Castro) may well start playing the audio file before it’s fully downloaded, but you’re not streaming the episode: You’re downloading it.
And there’s the rub. We know you downloaded the episode. We have no idea if you listened to it.
So ad buyers get up in arms. There have been numerous revisions to the agreed-upon definition of what counts as a podcast download. It’s complicated, and you don’t want to dive deep into it. Apple caused industry-wide download reductions with iOS 17, which changed how automatic downloads worked in the Apple Podcasts app.
Translation: Apple stopped downloading new episodes when people weren’t listening to episodes, so total downloads decreased. Total listening, of course, didn’t.
Ad buyers want “real” numbers. Which is hilarious, because the numbers most ad buyers can see are all poppycock anyway; they’re just universally accepted bits of poppycock. Nielsen’s methods for measuring TV watching or radio listening are, in a word, hilarious. And even if I know for sure that your TV was on when a commercial was aired, I have no idea if you had left to use the toilet or grab a snack.
With podcast downloads as the key ad sales metric, we know one thing: You downloaded the podcast. That’s it. It’s a real number! But whether you listened to the episode at all, or skipped the ads if you did listen — that’s a black box.
There’s a more recent development on the business side of podcasting called The AMP Accords, spearheaded by Dan Granger, the CEO of Oxford Road, the largest ad agency spending money on podcast ads in the world. I’ve known Dan for as long as I’ve been in podcasting. I get this goal. I think the AMP effort is a futile one, because I can’t imagine a world where Apple or Spotify will start sharing data on ad consumption out from their apps, which The AMP Accords require as a baseline.
For now, podcast advertisers tend to rely on what they call pixels or prefixes. (Pixels is a silly name, dating back to display ads; they’re really just URLs.) A prefix is hit every time you download an episode of a podcast — it’s essentially a redirecting URL that sits in front of the actual audio file URL, so that someone, somewhere can count every download request.
“Pixels” are tracking URLs that are served up at the moment you download the byte range of a file containing the ad. That’s a mouthful. But essentially, your podcast app goes to download an episode — whether automatically, or because you pressed Play (triggering a progressive download) or just pressed Download — and the server starts sending you the file. With dynamic ad insertion, ads are stitched into the episode as you download it. And when an individual ad unit is stitched in, the “pixel” is fired, letting the ad delivery be measured by some third party.
It’s clunky.
Listeners worry that the point of such pixels is to track you, but tracking is somewhat limited. Advertisers want to use these pixels for attribution tracking (in other words, did you buy this product because you heard an ad on a podcast), but the attribution is, by necessity, fuzzy.
The server delivering the podcast download can see your IP address and your user agent, the latter of which may include the app you’re using, the OS you’re using, as self-identified by the app you’re using to download the episode. And that’s different for your podcast app than for your browser. So if your iPhone downloads a podcast using your home IP and the Spotify app, and then you check out a product you heard an ad for on your work computer — there’s no clean way for the pixel to attribute your purchase to your prior podcast download.
Also, if 20 folks on the same Wi-Fi in the same college dorm or office building all have the latest iPhone and all use Apple Podcasts and all download the same episode of SmartLess… It’s borderline impossible to measure that, or attribution from ads included in that podcast, with any accuracy. And the IAB v2 standards that the podcast ad sales industry relies upon treat those 20 people as a single download.
So everyone in the attribution space uses some measure of extrapolation and interpretation.
Like I said, it’s fuzzy.
The AI of it all
Jeanine Wright launched Inception Point AI (again, now Liquid Studios AI). When she was interviewed by Dubner for Freakonomics last May, Inception Point had 15,000 active shows, with 300,000 active episodes. The episodes were 100% AI-generated: the topics, scripts, hosts, and recordings were all AI.
The math was simple: If you could aggregate a slew of listens off content that was cheap to make with no human involvement, selling ads across those downloads (that could also be voiced by AI, or could simply be pre-produced programmatic spots) could be profitable.
I like Jeanine. I did not like this model. Neither did Jenna Weiss-Berman, who told Dubner, “I think they’re the devil, and I’m happy to say that.”
I do not hate LLMs; I’m amazed at what they can help me build, and all the projects they make possible. But I absolutely think creativity should be left to humans.
Liquid Studios AI is a pivot for Jeanine’s company. It turns written text into audio. It turns audio into video. It translates audio into other languages. It generates clips for social media sharing. These things I understand, and I think they make a lot more sense than creating AI-generated podcast content from scratch.
I know of others in the AI space who are quietly working on different ways to use synthetic voices for ads. Can we get a host-read ad in the host’s voice that doesn’t require the host’s time? Sure, in theory, if the host is willing to let you use their voice like that. I doubt the ad will sound as good or compelling as one actually voiced by said host. I know of sneakier tech that fixes mispronounced words or substitutes in new or custom per-show promo codes in the ad reader’s voice. Those are much harder to detect. That’s just the reality.
What did Freakonomics get wrong, exactly?
I stand by my distaste for that “Wild West” analogy.
I also think Dubner gets it wrong when he goes deep on the word “podcast” and how bad he thinks it is. I do appreciate the idea of just calling podcasts “shows,” but I don’t mind that there’s a word that refers specifically to a format we all immediately know without any context (which “show” doesn’t do). Yes, “podcast” is outdated since the “pod” came from “iPod.” But the iPhone’s still an iPhone even though we barely use it for phone calls. What matters is we have a term we all understand.
Also, in his section talking about those AMP Accords, Dubner says “the podcast industry is trying to mature.” I don’t think that’s exactly right. I think that podcast ad buyers are looking to get ever more hard data on who’s listening to ads. I get that. The folks spending hundreds of millions of dollars on ads want more intel. But podcast ads are mature; that’s how the industry has grown so much already. We already know the ads work from promo codes and vanity URLs alone. I understand that buyers want even more data, but I don’t think they’ll get it.
So what is the future of podcasting?
I’m sorry to disappoint you, but I don’t have any shocking prognostications to share.
Just as with blogs and video creators and everything else, there are tiers. There are enormously famous and successful podcasters — Poehler, Conan O’Brien, the SmartLess guys, Bill Simmons, etc. They’ll continue to do just fine. Advertisers can buy ads on those shows and have hugely famous people voice those ads — while talking directly to their fans.
Mid-size and smaller podcasters who at least sometimes run ads will continue to face increasing challenges. It’s easier and more scalable for an ad buyer to run fewer ads on bigger shows. Hosts who reach niche, or affluent, or otherwise desirable listeners will continue to get some ads, but if you’re not doing hundreds of thousands or more downloads per episode, getting custom ads for your show will only get harder.
That may well mean you end up running more radio-sounding programmatic ads. These ads, in turn, alienate listeners, who will increasingly train themselves to press the 30-second skip button a few times every time ads come up. That will make ads perform worse, which will make the rates for ads decrease, which may lead to some creators cramming even more ads into their shows to make up the shortfall, and everything will get worse.
Technology like Jeanine’s will help, in some ways: Every podcaster wants their show to be bigger. A bigger audience translates directly to more revenue, if you’re monetizing your listenership with ads, subscriptions, merch, or even live events. You need more fans to make more money. So if Liquid Studios can help you create more easily shareable social assets from your podcast, that may help grow your listener base. Automatic translations may help too, but if you’re looking for US ad dollars, a French localization that sounds like you but serves listeners overseas is a bit less monetizable.
In effect, I think the biggest podcasts will get bigger. I think more famous people will launch podcasts, and some of them will be successful. I think hobbyist and prosumer podcasts will increasingly rely on subscription revenue as ad revenue gets more challenging to earn.
But I think people who are passionate and have things to share will at least occasionally continue to find audiences, and find follow-on success from there.
PS: Oh yeah, video
Dubner didn’t go deep on video, so I didn’t either. They say comedians want to be rock stars and rock stars want to be comedians, and apparently now huge swaths of people in podcasting also want to be YouTube stars. Okay.
Audio podcast consumption is different from video podcast consumption, is my hot take. Audio listeners are completionists; video watchers have shorter attention spans and are more prone to multitasking. (There’s a reason YouTube changed its counting methodology.) There’s a place for video podcasts; meet your audience where they are. But television didn’t kill radio. There’s still plenty of room for audio-only success.