On Tuesday, October 13th, Tom Webster and Glenn Rubenstein will co-present a new analysis of The Podcast Atlas built around podcasting's heaviest buyers: the Super Spenders. Who are they? How do they consume podcasts differently - or do they? Where do they show up across audio, video, clips, and social? And which behavior identifies them most reliably in the study? Together, they'll connect those patterns to what agencies see when real budgets are on the line - and what it means for how podcast advertising should be bought, sold, and measured. Registration for The Super Spenders: Who Really Buys From Podcast Ads is open now!
This week I am in Denver for an unusual setup: Content Marketing World, LIONS and The Market Research Event, all under one roof, sharing an expo floor and a keynote stage. I was there to speak at CMW, which I’ve done enough times now that I think I’ve earned the five-timers jacket. (I’m still waiting on the jacket. I’ll take a tote bag.) I also did a book signing for my podcasting book, The Audience Is Listening. A rigorous test of self-esteem, I can tell you that.
Putting marketers, creatives, and researchers in the same building should produce three different conversations. It produced one. I walked the expo floor multiple times and nearly every booth was selling an AI-augmented something: AI for writing, AI for briefs, AI for surveys, AI to summarize what the other AI wrote. If you’d dropped in blindfolded, you could not have told me which of the three conferences you were standing in.
That’s an interesting snapshot of this moment. It’s also a moment I hope passes soon. To be clear, I mean the HYPE, not the technology. The technology isn’t going anywhere. But someday an AI marketing conference will sound about as cutting-edge as a fax machine conference, and I’d like to get there quickly.
The average machine
For content marketing in particular, I get why AI took over the room. It’s an existential question for the field, because the latest models can do the job of the average content marketer. I mean every word of that sentence. There is a LOT of terrible content marketing out there, and statistically, most people are average.
Look at LinkedIn right now, if you don’t believe me. Lately it seems like over half my feed was basically written by Claude with no customization, distinction, or, you know, weirdness. I like weirdness. But it’s a poor potter who blames the clay. AI didn’t approve and post those updates. A human did. I am still waiting for the next updated model of human, but for now, I think we work with what we have.
Producing the average is not a side effect of these tools. It’s how they work. A large language model generates text by predicting the most probable next word, over and over. It is, by design, a machine for producing the most likely answer. The most likely answer is, almost by definition, the average one.
That’s wonderful when you need the average. Most of us have plenty of average work to do every day, and I’m thrilled to delegate it to the robots. I do, constantly. I have built a pretty ambitious infrastructure right at home, with my own box, models, and skills, to enable me to do more and do better work. It’s the Red Queen theory of evolution from Alice in Wonderland – you need to run faster just to stay in one place.
But nearly every AI tool marketed to marketers is pitched on making the job EASIER. Easier was never the job. Peter Drucker said the purpose of a business is to create a customer. One of my business school professors added the part I’ve carried with me ever since: the job is to delight a customer, profitably. Average doesn’t delight anyone. Nobody has ever been delighted by the most statistically probable outcome.
The Market Research Event had its own version of this, and as a recovering market researcher, it’s the one that worries me more. The hot topic was synthetic respondents: AI models standing in for real people in surveys and focus groups. Faster, cheaper, no incentives to pay, no one dropping out halfway through the questionnaire.
The pitch makes sense if you think the point of research is to produce a number. It isn’t. Research earns its keep by finding what you DIDN’T already expect: the outlier, the segment nobody modeled, the listener who behaves in a way the conventional wisdom says she shouldn’t. A synthetic respondent is built from the aggregate of what’s already been written down. Ask it a question, and it hands you the consensus back, neatly tabulated.
In other words, it’s the average machine again, wearing a lab coat. It’s useful for testing a questionnaire or pressure-testing a hypothesis before fielding. But if your whole business is finding the signal everyone else missed, a tool that returns the most probable answer is working against you.
I’ve seen myriad examples of this throughout my entire career. In another life, I did a lot of research in the Smooth Jazz format for broadcast radio. It used to be called “New Adult Contemporary” (NAC) until ONE woman in a focus group in Chicago said: “You know, it’s like Jazz, but it’s Smooth.” Gold with a sample size of one.
What this means for podcasting
Podcasting has its own expo-floor version of the AI conversation, and the dividing line is the same: AI used for the average tasks makes you better, and AI used to make the product itself makes you average.
- Show notes, transcripts, chapter markers, clip pulls. Delegate them without a second thought. Nobody ever subscribed to a show because the chapter markers were lovingly handcrafted.
- Ad operations. Trafficking, reporting, matching inventory to briefs. Same answer. Let the machines do it.
- The ad read itself. Here it gets dicey. The host-read ad works because a specific human with a specific relationship to the audience vouches for something. A synthetic voice reading generated copy is the most probable ad read, and the most probable ad read is exactly the one listeners have learned to skip.
- The show itself. You can now generate thousands of podcast episodes for roughly the cost of a coffee. Some companies are doing exactly that. I’m sure some of them will find an audience. What they will not find is the thing that makes podcasting worth advertising on in the first place: an audience that feels like it KNOWS the person in its ears.
We are going to have a special research report in a few weeks about how podcast consumers react to AI and use AI as a discovery surface for the medium. One thing I can tell you in advance: there are limits to what the audience will accept. And above all? They don’t want to be fooled. The intimacy of audio, in particular, is a competitive advantage. It’s also the one thing an average machine cannot fake at scale. Let’s not automate our moat.
Reset your thermostat
We all have an internal thermostat for what counts as good, or good enough. Some days we fall a little short of it, some days we beat it, but over time we revert to our set point. That’s not a character flaw. It’s how people get through a week without burning down.
Notice what that means, though. Your thermostat and a language model are doing the same thing: settling at the most comfortable, most probable level of output. If your set point is “good enough,” you are now competing directly against a machine whose ENTIRE job is good enough, and it doesn’t take lunch.
So here is the deal AI is offering, whether we asked for it or not. Hand it the average work. All of it. Then take the time you just got back and spend it on the things the average machine can’t do. I’m not suggesting you’ll spend all day being extraordinary. Nobody does. But the ratio has to shift, from mostly average with a little extraordinary to something much closer to the reverse. The people who thrive in the next few years will be the ones who actually reinvest those hours instead of filling them with more average.
That’s what AI is asking of us. Demanding, really. Turn the thermostat up past the default set point and keep it there. It will be uncomfortable, because the whole point of a set point is that it’s comfortable. Exciting, too. For the first time in a while, the average work has someone else to do it.
Just a tool
We’ve been here before. There was a time when “mobile marketing” was its own conference track, with its own gurus and its own perennial “this is the year of mobile” keynote. Then mobile became the water everyone swims in, and now it’s just marketing. Social media went the same way. AI will too. It’s a tool, and tools stop being the topic once everyone has one.
When that happens, the expo floor will go back to arguing about the things that always mattered: whether the work is any good, whether anybody cares, and whether it made someone a customer and kept them happy at a profit. The tools will be invisible, and the only thing left to judge will be where you set your thermostat.
If you’d like to know more about this, be sure to haul out your fax machine and subscribe to my faxinar.
New Partners
New Partners
Sounds Profitable exists thanks to the continued support of our amazing partners. Monthly consulting, free tickets to our quarterly events, partner-only webinars, and access to our 1,800+ person slack channel are all benefits of partnering Sounds Profitable.
- Overlap is a platform that builds and learns clip accounts automatically.
- Ozen.fm is a curated audio monetization network for multicultural and multilingual content, connecting podcasters and streaming radio stations with advertisers across Latin America, the U.S. Hispanic market, and international markets through programmatic and direct sales.
Want to learn more about partnership? Hit reply or send us an email!
