Hello, and welcome back to Pod News Weekly. I'm joined by a very good friend of the show. His name's Mark Askwith. He's the CEO and co-founder of Captivate. Mark, hello, how are you?- I'm all right, thank you, Sam. Good to be back.- Indeed, good to have you, my friend. Now, you boys have been busy again. I think True Founders is busy, but I sometimes wonder if Captivate's even busier than we are. You've just launched a brand new tool called Spark AI. So, what is Spark AI, Mark? And why is it different from every other AI we've seen hosts launch recently?- Well, I think in podcasting AI is pretty dull. It's pretty boring. And that's the beauty of Spark. It's why we've called it Spark, you know. Ultimately, you literally cannot start a fire under your brand, under your content as a creator without that little bit of a spark. But the difficulty is keeping that going. We know that fueling the fire is very tough and keeping the motivation, keeping the physicality of being a podcaster going can be really difficult. Spark, in my view, and certainly the way that we are positioning this within the podcasting industry, is it's the most useful podcasting artificial intelligence assistant that there is. And the reason it's the most useful is that, of course, it's fully integrated, which means that we can do much more within the ecosystem of Captivate. And I think, as well, given how different Captivate is overall, the guest booking platform, everything that we do with Amy, some of the things that we do around dynamic show notes, there aren't really that many hosting platforms that think as, number one, holistically, but without trying to become an all-in-one toolkit. We don't want to be an all-in-one toolkit. We want to be the right tools for the serious podcaster. And that's a real big distinction. So it always follows the same mantra as every Captivate tool. We'll either save you time, or we'll save you money, or we'll make you money. And this is where Spark comes in. And it's been fascinating to see it, because I'm a huge fan of doing things properly. We often do things first. We were the first 2.2 accredited host with the IAB. We were the first to put out proper dynamic show notes. We were the first to do various things, full transparency mode and guest booking integration, fully proprietary system for that. But we weren't the first to do AI. We weren't the first to do podcasting 2.0 stuff. But what we always do is we do it more thoughtfully, because I think there are certain times, little inflection points in podcast technology, where you've got to sit and you've got to think, well, actually, what can this technology genuinely do for people? What are the outcomes that we're genuinely trying to achieve? And that's how we approached Captivate Spark AI. We didn't want to just quickly put the most basic features in by integrating with another AI platform that is already serving podcasters. What we wanted to do was be more thoughtful. We didn't care about the feature list. You know, a lot of hosting platforms, it's one of my biggest points of contention in the industry. There are a couple of hosting platforms in particular that go like an inch deep on everything because they just want to get it on their website. And I really, I just think that's boring. I just think as a consumer, why would I want that? So the way that we approached Spark is we said, look, there are things that we're going to have to do. And there are things that maybe we can do that no one else can do or that no one else is considered doing. And it's all wrapped up in the ability to be the most useful AI for podcasters. And of course, remember, it is only phase one. So we'll get into some of the nitty gritty about it. But that concept, that mantra is why, I think not only the Spark AI, but why most things that Captivate does can be very different and it can be very, the feedback that we get is, this is the most thoughtful version of this that I've seen. Or very often we get the feedback you'll see on the live streams, why is no one else thought of this? And it's just the thoughtfulness. So we know that Spark has got that same level of thinking and that's been leveled back at us with the feedback that we're getting, which is really nice.- No, it's okay. So Spark is integrated into the Captivate platform. So when I use Spark, am I using Spark at every stage of my podcast production? So am I using it pre-production, post-production, promotion? So if I wanted to start most AI that I've seen within podcasting is post-production right now. It's, you've put your audio up, it takes a transcript, it creates a transcript for you, suggests speaker labels, maybe some titles, some episode breaks. So you've got your chapter markers, Matt, don't get me wrong. It was a massive leap forward from two, three years ago when I remember having to manually do that for Pod News Weekly to one click button. Now, because it's AI and it's early door still, it's never perfect. So you have to make some adjustments, but it was a pretty good step forward. So I saw the video you put out and obviously the first things you do are episode title suggestions. But so talk me through what you do at a basic level and then what you do at a additional level.- Yeah, good question. So you're right. The thing that's boring about AI is the stuff that it's already taken for granted for, which we know about, but it's like anything. We take leveling and mastering our audio for granted, but it still needs doing just because it's become so obvious in commonplace, doesn't mean that it's still not valuable. And that's the thing that we considered here with Captivate Spark. Yeah, we've got to do episode titles, but unlike a lot of players, we'll actually do you several and you can keep regenerating them until they land close or exactly as you want them. So that's a big tweak. We'll also give you the summary in the intro. So what we're doing is leading into things like promotional aspects, but also the YouTube world. We understand that people are going to take some of these assets, use them on YouTube. That's fine. So we didn't want to preclude anyone from using that. We didn't want to tell people actually, it's audio only, podcasting is this. We understand that we can probably be helpful using Spark. So we create you a summary in intro. We'll create you three variations of show notes and guess what you can regen them as well if you want. Just keep regenerating, it's cool. And then we'll create you a few things, of course, chapters, which can be copied into your show notes or this is the integration coming in beautifully. You can click a button and they will be added to the relevant places to your ID three tags. Podcasting 2.0 support or with one click of a button. The same with your transcript. Obviously you can generate, you can edit your transcript, but that is automatically added to all the right places. And we do a few things as well at that post-production level that you generally wouldn't always think of. So things like as people want to monetize their back catalogue or as people get ready for monetization but didn't think that they were ready for it. And actually this got a little bit surprised by this aspect that I can do programmatic or I can put my own ads in. But I didn't leave any ad slots. So what Spark will do is it will analyze the audio analyzer transcript and contextually tell you where some good places are for ad breaks. And the context is important. And yeah, it's not going to get it right all the time. It's AI, it gets you 80% of the way there. We know that. It's why it's called an assistant. It's not called a replacement. And again, I think that's really important. So what we'll do is we'll analyze the context of the audio and we'll say, look, put an ad marker here. The reason is that the hosts finish talking about a specific topic. They finish talking about a specific thing. And this is a really good place to put an advert. Or it's a segue between two segments or whatever. So that's really important. We also generate search engine optimized content. So we'll give you content around obviously a title, you SEO description, but we'll also give you YouTube keywords. And again, we're not going to just stand on our high horse and say, do you know what? The keywords have been deprecated in iTunes. Because we'll call it iTunes because we're talking about keywords. But actually, we want you to use them on YouTube. We know that they're still important. Just get on top of it. So we'll give you the keywords as well. So it's important for us to think about the ecosystem that someone is going to be working within. As I said, this is a phase two. And that post-production aspect of it is only going to get bigger and better. We've got a lot of really interesting ideas for that one I'll get killed if I talked about. But I think the power that comes with this is really, really for me, I think it's in the training that you can do around the persona. And I think it's in the way that you can really start to understand how to grow your show by capitalizing on two things-- what's happening with your audience and what's happening in your industry. So these are two big things that Spark does that no one else does. So first and foremost, we've seen these tools now where it's like, put your content in. And you can test whether this is written by AI. And because I get that, people don't want to always see AI content. So what we said is, look, because Captivate is a multi-show platform, you, Sam, me, Mark, we can have multiple podcasts on one account. But each show level, we've got to appreciate that each show is probably their own brand. It's probably its own tone of voice. There'll be similarities, of course, if the hosts are the same. But ultimately, there will be separate brands. They will start to feel a bit different as they grow in different directions. So we created something that we're calling persona training. We'll give you a couple of presets that's formal and informal. Let's go real basic. But the power comes in, this ability to set these custom personas. And what you can do is you can say, I want to be-- OK, I want to add some humor, and I want it to be witty humor. And in my writing style, my episode titles, and so on and so forth, I want it to be technical or I want it to be simplistic. I want it to be complex or not complex. I want it to be first person, third person, and so on and so forth. And what we'll do is we'll save that profile, and every time Spark does something, it will remember that. It will work that into your assets. But we take it even further. We say you can pop in any written content that you've previously created, paste those links in, and Spark is going to go check the tone of voice and save that against your show. So that you can start to write things like you. And obviously, this is going to get better and better. So that's a big thing, the persona training. And then things like the persona training feed into something that Captivate is very unique in, which is the episode planning. Captivate is the only platform that allows episode planning. You can clip research links from across the web and save them to your Captivate dashboard by going in, if you're a big news show, like something like the news agency, if you're a day-to-day show, if it's a pod news weekly, whatever it might be. I can go into my episode planning inside Captivate, I can attach the guests that I've booked on with all their details. I can attach research links that I want to talk about, add production notes, anything that I want to do. And Captivate, of course, generates the show notes based on that, generates an e-mailable PDF document for all of your talent or for your host or whatever you're doing, OK? The way that Spark AI now powers that is it takes the persona training and it will generate you two types, if you want it to, two types of episode ideas. The first one is, let's say, OK, Spotify announces that they're making a big push into video. I need to get an episode out, I can tell Spark, I want to talk about today's press release from Spotify. Here's a link to it. Click a button, give me some guest suggestions, and what Spark will do is it will go and analyze everything that's going on. And it will say, all right, here's an episode outline. Let's get it out, let's do it, let's get it done. Oh, and here's some guests if you want to reach out to them with their contact details, their Twitter handles, their Instagram handles, whatever it can find and whatever we can build into that. That's really good for reactive stuff, like really good for reactive stuff. Perhaps the thing that is more exciting, even more exciting, to be honest with you, is the idea then that I can use Spark to create completely fresh episode idea. That is press one button, and it's going to go away. It's going to look at your turn of voice, look at your style. It's going to analyze your own analytics to see what has worked in terms of titling, content, phrasing, talking points, what has worked, where did you get your peaks, where did you get your troughs, what do you do well at. And then it's going to look at your industry, at your niche, what's going on the internet in your niche. And it's going to bring everything together. And it's going to generate you. Here are three episodes, here are some titles, here are the talking points, and here are some guests. And that is just, it's ridiculous, like I use it all the time now. It's ridiculous to be honest with you. But it's so good, and the kicker, the big kicker is that thoughtfulness and the integrated thoughtfulness. Take my analytics, look what I'm good at. Take my industry, look at what's happening and bridge the gap. So yeah, I'm excited about it, as you can tell, man.- Let's break that back down. So persona sounds really good, different styles, different tones. Customized personas, I like that. And then the ability to generate content by using AI at what it's probably best at, going out and finding content and summarizing that for me, or starting with fresh ideas. I like that whole aspect of it. That's the pre-production element. I look at it, and this is not a negative, because I don't think it is. Could you just go that one stage further and have an AI voice and then we're out of it all together? I go, "Wow, okay." So Pod News Weekly, James could just get the AI to go and do the research I do, and yep, he could go and get a voice to replace me, and suddenly bang, I'm done, I'm gone.- That's the thing, and I think that's where we always sit with AI, is that AI, in the podcasting context, is very much about being complimentary to the creator. It's very much about being the assistant. It's very much the kind of technology and human collaboration, which I think is important. And yeah, people are messing around with AI voices, and there's the whole ethical argument around that. There's a lot of gray area in there. But you and I are a bit weird, aren't we, because we work in podcasting. But 99% of people in podcasting don't work in podcasting. Like, the last thing they'd want to do is stop doing the thing that they got into podcasting for, which is to talk about the thing that they love talking about. It's only going to be the people that treat this as a job and they've got to get something out that would even care about AI voice. Outside of maybe, like, let's correct some things, you know? Outside of that, I don't think AI voice is... For 99% of people, I don't even think it's anything that they'll even consider, because that's the reason they got into it.- I think brands might, that's my worry. I can see agencies going,"Oh, yeah, we can pop 10 of these out quite quickly." But I think the proof will be in the pudding, as in, will people connect deeply with an AI voice as they were with a human voice? So, one of the things that I wanted to also touch on was you've told me how you've got potential guests and episode subjects. What, in terms of any A/B testing, one of the things that I'm curious as, and it may not be an answer, but, for example, you say you suggest titles, right? And you suggest show notes, and it's all about SEO. How do you know what works? How do you know what is the right one? If I'm not an SEO expert, then I'm like,"Oh, that sounds like a great title." And then I put it out, but how do AI as a creator know that this is the right SEO to use?- Yeah, it's a good question. About 10 years ago, man, I wrote down that we should build into Captivate. Before we even built Captivate, we knew it was called Captivate, A/B testing on titles. Literally, I've got thousands of ideas like that written down. And inevitably, what becomes the problem with that is that you need exposure to things before you get A/B testing feedback. And in podcasting, that doesn't work because of the nature of RSS. Trying to get an episode back into the cage is just not doable. So, it's very difficult to do that. So, instead, we approached it from the perspective of every one of these assets, whether it's a description, a title, a set of keywords, whatever, will be better than someone with no SEO training will do. Will be better. And that's the point of AI, in my view. It's in this context, it's to raise the bar for those that have got no interest in raising it themselves. And that's not a negative. I get into a podcasting to talk about Star Wars. I don't wanna learn marketing. I don't wanna learn SEO. It's not why I got into it. But do I wanna be left behind because I'm not into it? Well, not if I can press a button. So, the way that we've designed everything, the way that everything's built behind the scenes is that it will lift the level of quality from, let's, I don't wanna say zero, but let's say that everyone's naturally can write a two out of five star SEO title. We'll raise you to a three and a half or a four. And then the SEO experts will probably take you to a five if you ever needed it. And sure, some of the stuff that's generated will be five. But not everything, because it's AI. We've got to be pragmatic about it. So that was the goal was to try and raise all ships raised and all that sort of cliche. We've got to pull everyone up a level. And I think that's where I can really play a part like this.- I agree. I think we are seeing the tools that mean that as a podcaster, I can get all of this done quicker to focus on the thing I love doing, which is actually creating the audio. What's this cost then, Mark? Is it part of the standard Captivate? Is it an additional feature now? How do I get hold of it?- Yeah, so it's just a quid an hour basically or a book an hour, depending where you are for the transcription. Everything else is based on the transcription. And that's the only thing that we get charged for. So as always, we'll just keep it as low as we can. And we just pass that on essentially. So a book an hour, we'll just do you the transcription. All the other assets come from that transcription. So the AI element of it, in essence, is actually not chargeable, but you need the transcript to do it. And obviously we've got to charge a book an hour for that one. So it's really manageable. It's really affordable for people. Imagine that when we got into podcasting. So once I give us a book per episode or a book an hour, you would have done it instantly'cause it's such a time-saver. So yeah, it's really well-priced, man. And I think the good thing as well is it's all part of Captivate as well. So the transcription will power everything else in the future. So this is only gonna grow and grow. So yeah, real strong stuff.- Noel's question, is this under the hood, is this a third party that you've licensed and then adapted? Or is this something that you've built in-house using your own LLMs and AI tech?- A little bit of both. There's a lot of API usage that we use behind the scenes, but not with any one specific partner. This is the benefit of not, for example, white labeling an AI platform that is already working in podcasting. So what we've done is we've worked really closely at an enterprise level with various, like you said, LLMs and providers of LLMs. And we've taken the best bits of each one of them, which is why I think it's the tightest version of AI we've seen yet in podcasting, because we've not been beholden to one supplier. It's one of the benefits of standing apart a little bit. And then we do have certain elements, not really LLMs, but I can't really say too much, but we are working on pieces of that puzzle ourselves for the future phases, which I'll be able to talk about in a little bit more depth probably over the next, I don't know, probably next year, 18 months. So yeah, we pull a lot of different APIs together, a lot of different models together and use the best models for each part of the job, if that makes sense. You'll get that more than most will because of how into this you are.- Can I ask one favor of you? Can you ask the news agents to now include transcripts and chapters within their podcast, please? Give them a copy of it.- No comment on the transcripts. The chapters will be bloody handy though, I have to admit. And guess what? You can do it now with a click of a button.- Indeed. Emily, just click the button. That's all you have to do.- I think if Emily ever logs in to Captivate, I'd be delighted.(laughing)- Mark, ask if it has ever a joy to talk to you. Congratulations on SparkAA, my friend. And just remind everyone where they need to go to find out more.- Yeah, for sure. Thank you, always a pleasure catching up. Yeah, always fun and always insightful. So Captivate.fm, if you scroll down the homepage, you'll see there's actually a banner there now that talks about Spark. Clicking on that, take you to the full page and you can go deep on that. But Captivate.fm is the place. X.