Jeanine is CEO and co-founder of Inception Point AI — this interview has been lightly edited for style and readability
Jeanine Wright: Inception Point AI designs AI generated personalities and then launches them into the world as 360 creators. And we create AI generated content on their behalf across multiple different platforms. We’ve started and grown up in audio, and in the process, we built what we believe to be the world’s largest independent podcast company by show count.
Editor: For more on the background of Inception Point AI, read this piece in The Hollywood Reporter.
James Cridland: From the reviews that that I’ve seen of your work - are you being judged a bit too much on the early stuff you’ve produced?
Jeanine Wright: I think that’s fair. That’s part of what I said in the piece that we posted. At this point we’ve created more than 160,000 episodes, and we’ve been doing it for than two years. We were very early in experimenting with AI - early in our own capabilities, and early in AI’s capabilities. A lot of the stuff at the beginning - frankly, it really sucks. You know, we go back and listen to it and we cringe. But I understand this is a very common thing - I’ve been working with creators in the podcasting space for many years, and pretty much without fail, everyone is embarrassed about their first episodes, because it takes a while to practice and to get good.
But, I’ve also been candid that even some of the stuff that we make today might not be good. We still do a lot of experimentation. And some of it lands, and some of it doesn’t. Some of the stuff I really love, and some of the stuff doesn’t do it for me. But I think it’s better.
James Cridland: Is this an SEO play? Are you flooding podcast directories with content and hoping that people find that content - or are you seeing people coming back again and again?
Jeanine Wright: We’re definitely seeing people coming back again and again, which is why we’re leaning into many of our personalities. I don’t know if it’s fair to say “flooding with SEO” - there are lots of businesses that make content based on what people are wanting and searching for.
Perhaps that’s been a problem with the way that we’ve been doing podcasting in the past. At some of the companies that I was at previously, you would make this beautiful, brilliant, incredible show, and then title it something that nobody has ever heard of. And then you would have to spend huge amounts of money to invest in telling people that you made this show and this is the title of this show.
Then listeners would have to type in to, candidly, these rudimentary search directories that a lot of these podcast platforms have. You would hope and pray that they typed in exactly what you titled this show so that they could try and find it. That was part of what drove up the content production marketing and talent costs of making content under this traditional model. That has resulted in widespread - I don’t mean to be dramatic - extinction events across our industry where we’re seeing layoffs at Stitcher, and Gimlet, and Wondery, and Audacy. These companies are just not finding sustainable business models to continue to produce original content, or even partner content, profitably.
If I’m going to make a podcast about whales, we title it ‘Whales’ because, that way, when people are looking for content on whales, they can find our content.
James Cridland: Yes, the Max Cutler method of titling a show. He had a serial killers show, which was called Serial Killers.
Jeanine Wright: Exactly. Because our time to production is dramatically less, and our costs of production are less, it means that we can surf trends much better than traditional podcasting organizations.
So, recently when Charlie Kirk was shot, we had a new show about Charlie Kirk and his biography and his his assassination up within an hour. So when people typed Charlie Kirk into Apple and Spotify, we were three of the top five shows - even beating out the Daily Mail. The only one that was beating us was The Charlie Kirk Show. So that means that we’re able to meet a need for content that people want in that moment. That’s not really ever been part of the commercial podcast business model until now. When his assassin was arrested, we had a new episode up within fifteen minutes.
James Cridland: Did a human being check that first?
Jeanine Wright: When we’re producing political or news related content, especially when it’s new shows that we are launching, we listen to the content before it launches.
It’s part of the reason why we haven’t leaned in to news and politics quite yet is because I think there still needs to be human review. AI is not so great at always capturing the tone of the moment, and when you are crafting personalities that you want to be complex, you have crafted into these personalities things like senses of humor and flaws and weaknesses and things, so sometimes they don’t always treat sensitive topics with enough sensitivity. They don’t strike exactly the right chord. So, as we think about news and politics we are figuring out how do we build the right infrastructure that still leverages AI and still is consistent with our model, but mitigates the risk of getting it wrong. That probably means that we need to have a well-known news person who is reviewing that content when it is released and is making sure that we’re striking the right tone.
James Cridland: Human review there is important, but human review for a podcast knitting or gardening isn’t necessarily as important?
Jeanine Wright: There’s all sorts of levels of risk associated with different content. We think about which ones need to be more curated and where we need to be much more careful. We also examine and stay on top of what are the capabilities of the tools: AI still struggles with numbers - I wouldn’t rely on AI to do your math homework. So when we’re thinking about current events, what are the topics that we want to cover? We lean into the things that we know that AI is really good at, where we are less concerned about any potential downside risk. If it’s giving health and wellness advice we’re going to want to make sure that we’re doing more to pull from multiple different sources and mitigate against hallucinations, and that somebody is involved in listening and reviewing the content before it’s released.
I’m releasing 3,000 episodes a week and I have eight people on my team. There’s no way we’re listening to the overwhelming majority of our content before it’s released.
James Cridland: There’s one podcast directory out there removing AI generated podcasts from its directory. What’s your thought on that?
Jeanine Wright: I think that it’s an antiquated way of thinking. I think that in a very short amount of time - if I can make yet another bold prediction - less than 24 months and probably less than 12, AI will become the default tool for creating content.
This conversation around “is it AI or not” will go the same way that this happened with digital photography and photo editing, right? Now, nobody makes a disclosure like “this picture was taken with a digital camera and was edited after the fact”, right? Everybody knows people use digital photography and photo editing. In fact, nowadays you make a disclosure when something is captured traditional 35mm and is not edited, right? That’s been the thing that’s special about it.
There was a survey on YouTube where 92% of creators said that they’re using AI in their creative process.
James Cridland: There’s a difference between using AI as part of your creative process, and using AI as the only part of your creative process, though?
Jeanine Wright: We’re not using AI as the only part of our creative process. I think that is a very nuanced conversation that people don’t realize.
Right now the standard is that if you use AI as a material portion of your content, you need to include a disclosure. So we include disclosures in all of our content.
James Cridland: You didn’t do that at the beginning, and now you do. Why did you change your view there?
Jeanine Wright: We were limited because we have so many episodes, to be able to dynamically update all of our content via the API. But we were able to build out the process in order to do that. So now if you go through and you see disclosures in all of our show notes.
James Cridland: But you didn’t make disclosures when you started. So why did you change your mind and start disclosing that it was AI generated?
Jeanine Wright: You mean two years ago when we first started making content? Well, to be clear, not all of the content that you hear from two years ago is AI generated. Much of it wass voiced by people who are on our team. Some of it, we went back and then replaced with synthetic voices and personalities that we designed. At the time, I don’t even think people were thinking about how would you make an AI disclosure. So we’ve adopted and evolved as the industry evolved.
I think we’re really proud of the way that we leading the way in transparency. For all of our personalities, we have some clever prompt engineering around how we have them do their AI disclosure - so they say that they’re AI, but they do it in a way that sells the benefit of having a host that is an AI personality.
And from our data, we find very little, sometimes immeasurable, drop-off after our AI disclosure. So our read is that if people like the personality and they like the content, they don’t care that it’s AI.
James Cridland: In the same way as there’s an explicit tag in podcasting, where you can mark if you use saucy language or adult discussions, some people have been calling for an AI tag to help audiences or advertisers programmatically filter out majority AI content. What’s your view on that?
Jeanine Wright: First of all, I think that some people are trying to capitalize on the fear around this conversation to make news for themselves. If you have people who are really thoughtfully engaging in this conversation, I think it’s much more complicated - because if you’re using ChatGPT to make a script, but you’re actually reading that into the mic, is that materially AI generated? If you write the script, but you use a synthetic voice, is that materially AI generated? If you’re doing an interview and it’s a real person that’s interviewing a digital person, is that materially AI generated? It’s getting to the point now, where it’s so nuanced.
I have people behind the scenes emailing me saying, “I work at this company and we’re using AI in our process.” And in most podcast conversations that I’m doing, they’re going for the jugular, asking me the tough questions about controversy in this space - but then we get off the call, and then they say, “How do you think that I can better use AI in my process to make my process more efficient?” Or, “Could you recreate my voice? And what would that sound like? And maybe I could actually take a vacation sometime,” right?
This is the most powerful creative tool that the world has ever seen. And creators are going to be leaning in to this technology very quickly. And this conversation is going to go by the wayside.
But in the meantime, we will follow the guidelines to give as much disclosure as you want us to have. The reality is, we’ve not had it as we’ve not heard any pushback from any advertiser, ever. And we are not finding that we are missing out on audiences.
James Cridland: Jeanine, thank you.