The MrBeast Rule That Quietly Decides Who Wins on YouTube | Rox Codes, CEO @ Flightcast

MrBeast said, if you make a video 10% better than mine, you will get 4 times the views. And that's true because of how YouTube works. YouTube picks the best video and the best video goes further and further and further. So every 10% better you can make this, it has the opportunity to have a 10x outcome. They test everything under the sun. Everything in the world is an A/B test. They've tested if the captions go on the top of the clip or on the bottom of the clip. They've tested if you should zoom in on the face first or have a voice say something first. They've tested if Steve should be on the right side of the thumbnail, the left side of the thumbnail, if Steve should be in the thumbnail at all. They've tested 100 thumbnails on Facebook, For every episode, we have you make a blank YouTube channel, and then we spam AI clips on the blank YouTube channel. Because if it was your channel, you would have 120 crappy AI clips spamming your channel, and you'd probably get banned. I will post 120 clips for the sake of knowing which 4 are good with no effort of my own. My eyes are closed, my hands are tied, the whole thing's on autopilot. The people who consume on podcasting platforms, retention's like 50% through to the end. Half of people listen to an entire episode of a podcast most of the time. Therefore, a stream on Spotify is worth 10 views on YouTube, trivially. And sometimes 100. It's always you versus you. Everybody sees all these big podcasts and thinks, I gotta get 10 million views like Diary of a CEO. No, if you got 1,000 views this week, you're going for 2,000 views next week. That logic holds forever. Just get better, pursue greatness, and be a student. Literally nothing else matters.

Today I have Rox. Rox worked with 2 of the biggest YouTubers on Earth, MrBeast and Stephen Bartlett. So now he's building Flightcast. It's the video-first podcast infrastructure that actually runs Diary of a CEO. And today he's gonna break down the DOAC playbook and tell us exactly which parts a normal creator like myself can actually copy. Welcome, Rob.

Hello. Thanks for having me.

So good to have you. How long have you actually been in the podcast world?

Ooh, podcast world. I've been in just over 2 years. Started May 2024 for Flightcast.

Why podcasting though? I know that, you know, of all the things you could build after MrBeast, right? And you're not a podcaster yourself.

No. Yeah, I got in touch with Steve. So I made my first company Thumbnail Test, which was A/B testing YouTube thumbnails. And that's what ended up giving me the job at MrBeast. But it's also how I got in touch with Steven from Diary of a CEO. He was a beta customer of mine. And then when I sold Thumbnail Test, Steve reached out. And he said, hey, we should do something together. I don't really know what, I don't really care. I think I've got 100 ideas of YouTube tools. So do you, let's just do something. And I said, all right, sure, let's try it. So I flew out to London. And we started working on some stuff. It wasn't exactly podcast-oriented originally. We messed around with a few different ideas. But then one day Steve said to me, hey, what do you think the best business we could ever start is? And I said, I think it's Probably podcast hosting. And, uh, the rest is history. We just kind of went for it.

Was it on Steven's list of ideas?

I think we had something that was like a Patreon on there. The original idea we came in with was called, was like a concept called Pre-Watch, which was, uh, showing people early edits of a video and then tracking retention on that. And then using that to re-edit the video to its final cut. So your viewers could be like, your superfans could basically tell you like, oh, this part was interesting. This part wasn't interesting. And then you could act on that feedback before. So like we've done small approximations of some type of a video platform. I still have a little bit of a feeling that I can maybe do a really cool take on like a Patreon from like an analytics perspective. But I don't think so. I don't think we had fight YouTube on the list. I think we mostly had win YouTube. That was mostly the goal.

That's a very cool idea, but it's very complicated to operationalize for a normal creator or early stage.

There's a lot of questions of content security and how interactable the data really is. I like building things that are very, very simple. I like the idea that any startup I build is basically like giving somebody a power drill that used to use a screwdriver. So I don't really have to explain it a lot. It's just like, hey, imagine that thing you were doing, but strictly better. And then they go, wow, that's awesome. I will give you money for that. As opposed to, you know, like all this crazy AI stuff people do now where it's like every business is so complicated and convoluted and it changes every day. And then people are getting, you know, replaced by competitors and ChatGPT shut down all these startups and everything. It's like, yeah, well, I made power drills, you know, and, uh, I feel pretty stable in being the power drill guy. Uh, so yeah, I'm really, I'm really glad with where we landed. I really, I really love what we're doing.

Do you think, uh, defining it as a power drill is limiting your audience or limiting your market because you're very, you're very niche-focused in a sense? I, I, I've interviewed someone, you know, focused on positioning and messaging. So I know there's pros and cons, whatever, wherever, whichever lane you're choosing.

No, I mean, I probably simplified a bit, but I think having AI not be the foundation of the product, but a piece of the product is a decision I absolutely still stand by forever. Because I, when we started the company, Steve was like, hey, let's do AI-first podcast hosting. And I was like, I don't think that's the win. I think it's video-first podcast hosting. So we tested it. And we did a bunch of testing, we ran a bunch of ads, and we looked at all the click-through rate and everything, and video first won every time. And so I was proven right. And we went with that. I think, I guess you could say, when I mean it's a power drill, I mean, it is a tool that does a simple thing everybody wants. But there are a lot of simple things people want. And there are some simple things people want that they don't know exist yet. And I think that that opportunity is still huge, where you could say like, hey, this is the first, I don't know, right angle bracket. And then people are like, well, I really need to put things in an L shape, but there's just no tools for that. And then I go, hey, I noticed you guys are doing all this L-shaped stuff. I made this L-shaped piece of metal. Is this useful? And they go, dude, hell yeah. And I'm not selling them crazy construction software to change exactly how you run your whole business, all this stuff. I'm like, no, let's just like, test clips better by default, automatically or something, right? Like, let's just take things that you're doing, and then just make them simply easier. And that's the whole thing. I'm just such a fan of like simple businesses, but the business being based on insights, right? So like other people are not curious enough to sit and stare at the construction workers for long enough to find out that most of the brackets should have been 90 degrees. So yeah, that's kind of my sort of thesis of, I don't know, I consider it both a boring business and a very hype business. It's the most exciting thing in the world to me. I don't know if it would be to everybody else.

Where is your insight coming from? Is it from working, like staring at how Steve works all day long?

It's a bit of everything. I have a lot of conversations. I spend so much time not talking to my customers exactly, but Talking to like just people in the industry. I talked to my friends that are Twitch streamers about how they're using YouTube Studio now. I talked to my friends who are big YouTubers about what algorithm changes they're seeing. I talked to my friends who are YouTube strategists about how their ideation process has changed. I talked to the biggest podcasts I can find about how they're doing clips and how they look at numbers differently. Like, there's so many new playbooks coming out every day. And I think it is my job To bake those into the software. So we call it the playbook from the diary of a CEO, which is one of many things that it is, but it's kind of like, it's meant to be every playbook, I guess. It's meant to be the modern video podcast playbook, whatever that is in the moment where you sign up for Flightcast, it should reflect that. It should know how the platforms work. It should know how Spotify is winning right now. It should know what the current state of the ad industry is. It should be almost a consultant for you as much as it is. a power drill is a power drill with instructions on how to drill things, uh, in case you're new to drill stuff, drilling stuff. Uh, yeah, so it's just a lot of talking. I go to like every creator conference. I talk to people all the time on Discord and I try and stay super tapped in. I think that's my number one job.

Do you feel in this process there's anything that's not replaceable by using AI?

I guess not replaceable versus shouldn't be is a different question.

Yeah.

But I feel like the number one problem I have with the way that a lot of people use AI to build products right now is Replacing your thinking, I think that's where you really lose yourself. AI is really good at implementing a concept. The best position you can be in is having AI be your hands as opposed to your brain, in my perspective. You have to have the brain. So I don't think AI can go all the way from, at least in my space, ideating a product, building a product, Shipping it out. Like, it can't do the whole thing for you. The reason Flightcast works is because I'm the only person with these ideas right now, because I'm basically taking a bunch of YouTube-based ideas and bringing them into the podcast world and then developing them through the tech of the podcast world into new layers of idea. I don't think I could do that without this level of insight. and kind of depth of understanding. And I don't think, I just don't think the AI can do that kind of like ideation. There's this weird thing I heard once where the guys who do a lot of like the lovable type apps say, well, don't be specific as to how you want the problem solved. Just tell the AI you want to solve the problem and kind of let it figure out how to solve it. And I hated that. I was like, this is— This feels like the worst advice I've ever heard. And maybe when you're first building and you have less of a picture, sure. But for me, like us knowing what we want it to look like, knowing how it should feel is part of the insight. It's not enough to just go and build a thing that does a thing. It's like you have to know every piece of it to deliver it flawlessly. And so I think that's just missed out on. AI lets you ship a thing if you know what you want to make really well. If you don't know what you want to make really well and you don't have a great take, I think you're going to build garbage anyway. And there's a lot of people putting out garbage and I don't think it helps the world. And I don't think it helps them either. But my friend who's a YouTube producer who was never able to code is now able to make her own YouTube producer toolkit with a bunch of really niche tools that really help her. And then I posted about it in a group chat of producers and they were all like, whoa, can I get access to that tool? And I was like, this is awesome. Look at her. She's crushing it now. And she couldn't, she can't code. Like she never knew how to code, but now she can vibe code. So if you know what you want to make, amazing. If you don't have the insight, it's not that great. So I think AI can't do the insights yet that deeply.

As a starting point, it's great. But part of it is I realized that when people are, if you have very concrete ask for AI, it may not follow exactly your instruction. It may not turn out to be what you want unless you actually know how to build it correctly. It doesn't come out from a prompt, like one prompt, you want a full-fledged app. It's not gonna happen. You can have a prototype, but—

Yeah. And I think that's part of it, right? If you know what you want, like AI is less good at delivering something to a spec, right? The more complex the spec, the harder it is for it to deliver. There's a lot, it's gotten a lot better, you know, God bless like this past couple of updates, man, between Fable and like even 4.6 or, uh, like GPT-Sol and everything. Like it, it checks its work, you know, it, it reruns things, but I'll still get it. Giving me a UI that has like no padding half the time. I'm like, all right, you know, we've missed so many fundamentals here. God help you to actually build a thing of use. And it's just, again, there's— it's like sculpting. You're just given this giant piece of clay and you have to know what you want it to look like at the end. And then just slowly just hit off each little piece, smooth every little edge. But if you don't know what you want, then you don't know that there are edges there. You won't understand that this thing is clunky. Inherently, and you'll have to go out and figure out that it's clunky by giving it to people, which is fine, but you'll move slower.

I love that analogy of engineers becoming sculptors, right? We're the sculptors of a final product, but it takes kind of effort.

Yeah. It's still a buttload of work. It's so much work.

It's so hard.

Flightcast was so hard to build. It's so complicated. It's such a nightmare, but God bless, we're getting somewhere.

You should share some of the challenges. What are the biggest challenges building it?

Yeah, I mean, platform inconsistency is huge. Like getting data from everywhere sucks. Data from YouTube, data from Spotify, data from Apple, data from RSS, and then creating different insights creates like a very scaled kind of data storage and aggregation problem. So we had to learn like a lot about that. Video distribution, because of how Apple does video now, basically Flightcast is a mini YouTube. Like we are doing the video delivery. That means the video is streamed from my servers to Apple. Which is basically like from YouTube, the videos are streamed from YouTube, from Spotify, they're streamed from Spotify, for Apple, it's streamed from Flightcast or whatever hosting platform. So that created a whole magical nightmare pipeline of HLS re-encoding, make a 1080p version, 720p version, and then it's just been so much work. And we've had to like bring in subject matter experts. Luckily in our network, we had a lot of the right people, right? Like my CTO did a whole streaming analytics platform before this. So he knew ball when it came to analytics. And then we had a friend who ran like a streaming video delivery service. So we were able to like bring him in and say, hey, we need you to help us with this Apple stuff. This is so complicated and so involved and we're just losing our minds trying to make it happen and then trying to make it happen at no cost, right? So a lot of the competition has charged additional money for video if you're doing certain amounts of video or certain scales of video. And we don't, and we made the decision not to. And that required an astounding amount of engineering effort to make that even feasible, to not charge bandwidth. And we pulled it off, but we had to rewrite our entire infra probably 5 times just to get to the point where we could do— like analytics is on v5, I think video's on v4, and there's still going to be iterations beyond this. But all the pages load fast now, all the data stores fast now, all the analytics are right and the ad system runs well. Oh my God, dynamic ads are so complicated. You have to like in seconds or milliseconds, really like hundreds of milliseconds, stitch together an audio ad file with a core 2-hour audio file. So you have to like slam the audio files in where the ads go before you send it. So as opposed to YouTube where it would like cut away to an ad, in audio, the ad has to be baked in. So you have to like at request time bake it in and it's a nightmare. God, it was just, it took so long. We made our own cut of FFmpeg. Like we literally ripped our own version of FFmpeg, which is like the default kind of video editing CLI tool. And we just tore it down to like only using audio. And then I think even that wasn't fast enough. So then we started just like slamming together like the raw memory of the files. I don't know. It got so complicated. Every single thing we've done has been complicated. And that's why I love this business because that's the part that I'm not really worried about anybody else coming in and doing.

Yeah.

Nobody aspires to build a product super fast. That's really annoying to build. Everybody wants to build simple things.

What can you share about the playbook for DOAC? At their level, a point system of, let's say, 1 to 100, I'm assuming they're gonna be, you know, level 100. What can you share about what they do?

They do everything, man. They go nuts. Uh, they test everything under the sun. They— everything in the world is an A/B test. They've tested if the captions go on the top of the clip or on the bottom of the clip. They've tested if you should zoom in on the face first or have a voice say something first. If it, Clip should have a title or no title, if the clip should link back to the episode or if it should not link back to the episode. They've tested if Steve should be on the right side of the thumbnail, the left side of the thumbnail, if Steve should be in the thumbnail at all, if there should be a face, if there should be a picture in the middle, or if there should be words. They've tested 100 thumbnails on Facebook for every episode, and then they test the top 6 as like 2 separate A/B tests on YouTube. And then they test different titles. They do the same thing, seeing like click-through rate on like paid And then, and then, and then everything that could ever be tested has been tested over there. Everything that could ever be learned, they have pursued the knowledge of. It's why Steve and I get along really well. Because I started in this space with like A/B testing, right? Like A/B testing thumbnails was my default. That was my whole business. And Steve was a beta customer. Steve PayPal'd me $20 2 years ago. No, 4 years ago. He PayPal'd me $20 because I didn't even accept credit cards yet to use my software. Because I was not as good at programming back then. Within a month of me launching it, so that tapped in, right? Steve on Twitter somehow saw my rinky-dinky tweet saying, hello, I am beta launching this thumbnail software. I think I had, I don't know, 3,000 Twitter followers at the time. And somehow Steve saw that because that was his algorithm, because that was all that he cared about. So we've started to build a lot of that into Flightcast. There's a few kind of core playbooks that we took from it. The first is obviously like a growth mindset first. Every podcast hosting platform is very focused on ads and monetization. We are very focused on growth. I don't care about ads. I have all the tech for ads, but other people make most of their money on ads. I make most of my money on SaaS subscriptions. And I would like to keep it that way. It makes my business very simple. Power drills. So first, one of the first things I built was clips testing. So whenever you upload an episode to Flightcast, it AI generates clips just like Opus Clip or anything else. And what we have you do is we have you make a blank YouTube channel. And then we spam AI clips on the blank YouTube channel. Because you shouldn't do that.

My channel?

It's not your channel.

Okay. Okay.

Because if it was your channel, you would have 120 crappy AI clips spamming your channel, and you'd probably get banned. And the quality of your channel will go down so much, everybody would be so pissed. And they'd be like, Angela, this is crap content. Why are you doing this to us? So we make a separate channel. And we use that as basically the testing ground. And then every day, we generate clips off of your new episodes. And we put out up to 4 clips a day, like, let's say like 120 clips a month. And we just spam them at this other channel. And we just track the views. That's it. Very simply, we track the views, likes, comments, all the stats. And then we show you which clips did well on a blank channel. And this is an approximation of what Diary of a CEO has done with clips. And Largely everybody has, which is how much can I scale clips channels? How much can I scale the same content being repurposed? But generally people have used clips as a sense of like growth and views. We are using this separate channel to learn. And the idea is I will post 120 clips for the sake of knowing which 4 are good. And I will, A, learn what makes a good clip quite quickly through that. With no effort of my own, my eyes are closed, my hands are tied, the whole thing's on autopilot. And then B, I can choose to, if I want to, take those 4 clips and put them on my main channel and know I'm only getting the good stuff. I have pre-validated that this is good content, right? So we've basically approximated— Diary of a CEO has 5, 10 clips channels, right? Run in different ways by different people testing all these different things. This is the simplest, simplest boiled-down version of that, which is just, I will make you clips. I'll post them and we will track the views and then you will learn. And then eventually we feed that all into our AI chat. So you can also talk to the AI about what's working with your clips and what's not, if there are any patterns in the transcripts getting pulled and not. And then you can kind of see very quickly how that could spin up to a much larger sort of content intelligence base when you combine that with all of the long-form data. So that's like probably the most clear example of a playbook that we've taken. But being video first is a playbook, like being video first on YouTube, being video on Spotify, what analytics you look at just being growth first at all in the podcast space is weirdly unusual because podcasters are not used to having an algorithm. On YouTube, we all think about the algorithm. Any video can go viral on YouTube. Not true on Spotify. Not true on Apple Podcasts. Everything can't just go viral on those platforms. They don't really have a concept of virality. Because to go viral, there has to be an algorithm to make a thing viral, an algorithm that optimizes for that somehow. So at best you have a light recommendation engine, but half of what goes on your homepage on Spotify is editorial. Someone at Spotify decided that would go on your homepage as opposed to a computer 50% of the time. It really changes the context. You also don't get like click-through rate on Spotify. So you don't know if you can make a better thumbnail. You also don't get, well, actually you do get retention now, but like on Apple Podcasts, similarly, you wouldn't get click-through rate. Apple Podcasts doesn't even have thumbnails. They just have cover art. So what even could you test, right? On YouTube, when something goes wrong, you can change something, right? You can change the thumbnail, you can change the title if you see that it's not doing well. On audio, you don't necessarily have those same levers in the same way. It's not as obvious that the thumbnail's bad. Is the thumbnail bad or is the episode bad? Is the title bad or is the episode bad? Is the intro bad or is the episode bad? Are all questions Flightcast answers right away. We tell you what the click-through rate is on YouTube. We tell you If there's drop-off early in the episode, because it's actually a bad intro. So we like will, for example, cut 30-second retention as its own number. And then we present that number to you. So you can analyze that over time and see if this episode had a good intro or above average or below average. So we started to take this like very data-focused approach that Diary of a CEO has done, and bring that sort of YouTube growth mindset into podcasting to say, hey, this is how you should be thinking about this. And that's like the huge, huge ethos. And that's the like early levels of podcast growth strategy and just, I guess, generally content growth strategy that we took from Diary of a CEO and from all my YouTube friends, and we baked it into the software.

I'm doing my own version of the testing, for instance. However, I think the best testing is from the market, like getting market reaction rather than whatever rubrics or frameworks I use to measure the potential popularity of any, any moment of a video. I can use AI to make some judgment call. I can hire a producer to help me find the best moments of a video, right? But you're saying this can be done automatically through this proxy channel. And then all I need to do as a podcaster is just upload my video and I don't have to do anything. And I can just read the results.

It is at least a start. It is not as great as a human would be. The clips won't be as good, right? You could always make way better clips with a human and paying way more. But it is absolutely level 100 of let me start testing, let me start learning, and let me do it with my eyes closed and no effort. And my hope is to build a lot of those kinds of opportunities in. So yes, if you're doing it all yourself, it's probably as good as what you're doing right now.

The challenge is this. Sometimes I question myself, am I really picking the best moments of this conversation between me and Rox? But locating that, that moment. I think that's harder.

It's always hard. And then knowing that you're right. The problem with making clips is, are you right? The problem with making clips is, do I know that this is the best 30 seconds? Do I know that this is going to do well? And the magic of testing is you just find out and you lower your risk. So you get the answer without having to risk a low-view video, without having to risk the audience being disappointed. That's like the core thesis of the kind of like experiment channel. Diary of a CEO will test it live, right? They'll just run 2 versions of the same clip on the same account, or do like paid spend against it and see if like the click-through rate's better or the watch time's better. But we don't all have those resources, those time, that amount of time. We don't all have that capability to A/B test. You know, I have many iterations I want to do on this clip system to make it much more complex and more towards what they've done. But this was the first step, like, because like you said, for the average creator, Doing it by themselves, even with an editor sometimes, just keeping track of all these stats is hard. Having theories about what works is hard. Picking good moments is hard. Every single piece of this is hard, but it doesn't have to be, at least to start. And then you get to learn and get intuition from the A/B testing channel. So now you will have some intuition about what works without having to put in effort to gain that intuition. All you have to do is watch your top 5 clips every month and you say, okay, cool, I'm seeing a little bit of a pattern here. I understand what's probably going to work. And then you get free intuition, you get free ball knowledge. That is the pursuit, the pursuit of ball knowledge.

Do you have different dimensions that you're testing? You mentioned some, but what's available on the product right now? The thumbnail cover for each clip, you have the title, you have the first 5 seconds. Do you have all those available just yet?

Halfway there. I'm building to it. I'm building to it. I want to do everything. My aspiration is to build kind of a system where you can just Really double and triple down on like, find a good thing and then find every iteration of a good thing and then, and then, and then. So test the titles and everything on like one clip. At the moment, we haven't done that yet. There's just not, there's not yet enough appetite for it. I know if I build it, people will love it. And I've had a couple of my enterprise guys go like, hey, I got some crazy clips ideas. And I'm like, I got some crazy clips ideas. But right now it is very much hip fire. It is the simplest system it could possibly be. It is generate as many clips as we can possibly think of, post literally all of them, and then see what worked and what didn't.

How does clip testing affect the longer form video?

The theory is the more you understand, the better you do everything. The pursuit of Flightcast is to help you learn what works and what doesn't. So you may notice when you go and clip this episode, probably that the clips that do the best are clips where I started by repeating the question you asked me because it made the clip more concise, right? And I almost gave you a hook every time. And you will— and then we post 40 clips from this episode and it's like, oh, well, the top 4 or the top 5, 3 of them had Rox saying kind of a hook thing at the beginning inherent to the way he answered the question. So Maybe I should be giving my guests a little one-pager about how to speak well in interviews that says, hey, I'm gonna ask you a question, and then I would like you to reframe it as you would imagine a clip would go. And like I did this kind of thing at a talk that I did not too long ago at Open Sauced, where we were talking about like, are editors cooked or something? And I started every— and they told us they were clipping the things. So I started every single answer to my question With something that felt like it would be a hook, or like, how do you do better? And I was like, okay, like, so it would be like, so how do you grow on YouTube? And I was like, all right, if you're trying to grow on YouTube, you have to blank, right? As opposed to just answering the question, there's like half of a restate to give the hook, right? All of these little things, you start to notice patterns. The patterns aren't always positive. Sometimes the pattern you notice is that like, oh, when I had a guest on that people disagreed with politically, Those clips did really well because the comments were all people saying how much they hate that guy. And unfortunately, you have learned something. It's not a positive learning, but you did learn something, right? And I have seen these patterns show up on these channels randomly. Like just, you would constantly, constantly be amazed at what goes more viral or what does well, what gets the comments, what gets the likes. And once you have those patterns, you can then inform your content strategy. So Flightcast is meant to make you a better creator. Flightcast is meant to teach you all of the things that work and don't work, however you may find them to take shape. And then you can use that knowledge to do better. You can use that knowledge to make a better long form, make a better clip, make a better thumbnail, whatever that looks like. And then hopefully you grow because of it. That is very fundamentally what we're trying to do.

I started grading myself from my first interview almost 100 episodes ago. I devised like 5 dimensions on every podcast, and I'm grading like the conversation, warmth, and insight, then, you know, depth and shareability, and I also grade myself how I perform, but it's also from, you know, from a researcher perspective. It's not from the market reaction perspective, but you're saying that actually, you know, you can get some market signal and then feed that back to how you can improve yourself and how you can, you know, Guide your— guide my guests to react on camera, right? So this is very interesting.

You can do anything. MrBeast said in an interview recently, probably my favorite MrBeast quote, MrBeast said, if you make a video 10% better than mine, you will get 4 times the views. And that's true because of how YouTube works. YouTube picks the best video and the best video goes further and further and further. So every 10% better you can make this, Every 10% better this podcast can get, every 10% better the title, the thumbnail, the clip, it has the opportunity to have a 10x outcome. And therefore, all we can do is pursue every 1% improvement, every 10% improvement that we could possibly ever get.

It's so hard. So hard.

It's so hard. It's so impossibly hard. And that's part of why Flightcast is a thing, is because I want to make that one thing less hard for people that understand that that is the one thing.

Today I feel it's getting harder and harder because all those AI-generated video and more and more people are doing podcasting. I also noticed a lot of companies are doing podcasting for, you know, different purposes, maybe relationship building, building relationship with your potential clients, things like that. So it's getting very crowded. How do you stand out?

If you're starting a podcast from zero, the top 2 things I recommend doing are number one, You have to come out with a bang, with an episode that feels like a good YouTube video, not just a good podcast. Very common, people will say, oh, I did a great interview. But if the thumbnail sucks, or the title sucks, or the pitch isn't good, then it doesn't matter because you're competing on YouTube with everybody else on YouTube. This episode needs to be better than most YouTube videos by default. It needs to pitch itself better than most YouTube videos. It needs to have an intro better than most YouTube videos. And the second thing is, I would optimize the way that you're asking the questions and the format to pull out some clips, because clips are such a, such a, such a huge thing. Now, every time I go into a podcast, if I'm going to interview somebody, I kind of know what they're going to say already. I have some semblance of an understanding. And I'm asking questions to pull out of them what I believe will be a meaningful clip. Because I think this guy has cool insights, I've researched everything this guy has ever said. I know he'll probably say something along these lines. And I know that that's gonna be something people wanna hear and that'll tie them back to the show. In general, short form to long form is very hard, but still right now it is make a video that's better than a YouTube video and make a format that lends itself well to clips. And then just do great research and make sure that you are feeding those clips well yourself, that you are like pre-filling a great question that can make a great clip or teeing people up to give you that opportunity. If you do those 2 things, you're starting off 10 times more likely to succeed than the average show just going out on Spotify, going out on Apple, publishing on YouTube with a thumbnail that they made on Canva in 5 minutes, and closing your eyes and praying.

These days, clips is more important than the long forms.

This is a point of contention. I don't think so. But I think it's different. What I've learned about podcast clips is so many people say that clips are the whole industry now. Clips are a different audience. There are people who will watch only your clips. And then they'll say, I love TwoSet.ai. I love that show. And they have never seen a full episode of it. They've never listened on Spotify. They've never watched on YouTube. But they consider themselves a fan. They consider themselves a viewer because they watch your TikToks. And that is a different form of the same content. So to some extent, it is just a different audience base. But at the end of the day, right now, So much time being spent on short form. It is one of the largest levers of growth that anybody has, because being really good at long form is very hard. And the way that people interface with podcasts is very different. Like Spotify is almost closer to a Patreon than it is to a YouTube. People will be coming back week after week. So acquiring that viewer, that listener on Spotify or Apple is such a big deal that it is worth putting out the clips. But a 10 million view clip might get you like a few hundred to 1,000 listeners, maybe, maybe. So the scale of what has to be done on clips to win is so difficult. Podcasts, long form is so important because it's what keeps the person coming back. So I think at the end of the day, the long form is still the most important, but you probably won't grow if the long form can't be appreciated in some short form capacity.

I don't want to say that doing a long, the longer video is just the baseline base asset for producing the clips. There are a lot more within a long form. Versus the, the shorts.

That's what makes it hard. It's the pursuit of a long form that is standalone and beautiful and well-made and valuable that simultaneously is clippable. And that is kind of the difficult pursuit of making a great podcast.

Uh, there are research prove— proving that people's attention span are shortened these days from 7 seconds or something like that, which makes me feel like, what's the point of having a longer conversation anymore? What if people only watch us the first 7 seconds?

Yeah, that's why I like this space. The people who consume on podcasting platforms, retention is like 50% through to the end. Half of people listen to an entire episode of a podcast most of the time, which means that whatever you are making is being consumed deeply much more often. And therefore a view in podcasting, in my opinion, is worth like a view on YouTube or Spotify, a stream on Spotify is worth 10 views on YouTube. Trivially for me, and sometimes 100. So that pursuit is so valuable. Once you have a podcast audience, if you truly convince somebody, I am making good stuff that you care about, they will come back week after week after week. And it is probably the most valuable view you'll ever get, because one view becomes so many more. And that is a person who's a true fan of your audience, they'll probably subscribe to your newsletter, they'll probably buy something if you recommend it. And that depth and trust is what makes the pursuit worth it. And the opportunity, once you have a big podcast, to then have massive sponsors that pay a lot of money because of that trust that you've earned creates this gigantic mountain to climb, but an amazing view once you're at the top.

What are the most important metrics we should care about as podcasters?

Same thing as usual. I like default YouTube metrics, click-through rate and retention and views. Those are kind of your big 3. For podcast platforms, you can't really dig as deep. I would look at returning versus new. You don't want too much of either. If you have all returning and no new, then you're not growing. If you have all new and not returning, then you made something kind of clickbait that doesn't really relate to your audience. And that isn't actually helping you. If you get both new and returning, that's the dream. You should be looking at who is showing up. You want probably as close to an even split as you can get. On podcast platforms, it's going to be heavily returning. But let's say on YouTube, you really want to aim for maybe at least 50/50. That is probably the deepest, easy stat. There are other stats that we look at in Flightcast that we've built up. I also look at overlaps between episodes of precise audience. That's like a graph that we made. So you can see how many people that watched episode A also watched episode B. And that lets you kind of infer If these topics overlap well. So it's kind of like a deeper view into that, like returning versus new, that goes more precise to exactly who watched as opposed to just people who have been here before versus new guys. That's been really interesting for us and has helped inform a lot of content strategy. But generally, I think you want an even split of new and returning. If not, you're either not attracting enough new people or your current audience doesn't care. And then otherwise, it's your classic YouTube 3.

Do you have actually guidance on content strategy from these data? There's so many data being collected, but it's hard to interpret.

It's very hard to interpret data as a creator, especially the bigger that you get. Part of what Flycast does is start to be that kind of consultant and interpreter for you. Every graph that we have has an explain this to me button that has AI analyze it for you and kind of use our own baked-in breakdown. If your click-through rate is low, change your title or your thumbnail. If your retention is low, make the episode better. Do tighter editing. Look at all the drop-offs in the retention graph on YouTube, or if you're in Flycast, you can see all of them overlaid. Like find those moments of drop-off. Any moment where some more than 2% of people dropped off is something you should do better on. It means people got bored and they left. Look at your first 30 to 60-second retention deeply. That is your intro retention.

Mm-hmm.

Every person that you didn't keep through the intro, Did not stay for the rest of the episode. Sounds obvious, hard to deal with emotionally. Make a better intro almost always, especially if your first 30 to 60 seconds is bad, or like that's below average, I should say. And then on podcasting, again, shoot for 50/50 new and returning. If it's all returning, change your title format completely, change your thumbnail format completely, probably make more clips. You're not getting a new audience at all.

Right?

And you need to do something to attract them, which might mean that your titles are something only your existing fans will click on. It might mean your thumbnails are good for your existing audience, but not attracting new people. So maybe they are recognizable, but it's not a great format. So if you're seeing nobody new, change the thumbnail format, change the title format completely, and probably change the episode format or style to bring in some other people. That one not definitely, but consider that maybe the topics you're covering are not interesting to a wide audience.

Should I consider like reupload the episode if I tighten the editing or improve it?

I generally don't like reuploading. I usually recommend the best way to learn is for the next one. Just always trying to make the next episode better. But the question is always just how much can I learn from this one? Right? Like look at your top 5 or 10 metrics, figure out if your intro's good, if your thumbnail's good, if your title's good, and if you're bringing in the right people. And when you have those 4 answers, if any of them are no, Then pursue solving that in the next episode. Ask yourself, what can I change? What can I do here to make this better? What can I do to make that first 30 seconds better? How can I tighten that intro? How can I pitch people harder? How can I— whatever it is, or maybe it's no intro, maybe you go straight into the conversation. Maybe the intro actually killed you because you tried to do an intro. Test stuff, find out. That's it. Just if you keep making your next episode better than the previous one, in 4 years, you'll be one of the best podcasts in the world.

4 years?

Well, that's how long it took Diary of a CEO, several years. But even in 6 months, you'll be 2 to 3x as good if you're uploading every week, right? And if your podcast is, imagine in 6 months being 3 times as good at this. That's a huge jump. Give it another 6 months, you're 6 times as good. Crazy. Or 9 times as good.

Math. Yeah. That sounds like the North Star metric that we should be shooting for. Like, can you improve yourself? Like, can you 2x yourself?

It's always you versus you. Everybody sees all these big podcasts and thinks, I gotta get 10 million views like Diary of a CEO. No, if you got 1,000 views this week, you're going for 2,000 views next week. That logic holds forever. And that's how every single guy grew. Just get better. That's all you have to do. Just get better and pursue getting better. Pursue greatness and be a student. Literally nothing else matters.

That's my clip.

There's your clip.

Yeah. Thank you. Anything from MrBeast does that you think is applicable for podcasting?

It's everything that we just said. Like the Beast mentality is always just make the best video possible. That's the only sentence they ever say. You have 150 people dedicated to making the best video possible. They care, right? Care too much. I, I am such a proponent of this. MrBeast says make the best videos ever. Diary of a CEO says get every 1% improvement you possibly can. Just care. Care so much about making the best damn thing. Not enough that it makes you not upload the episode, but care so much about everything that could be better. And you will suddenly find all of these unturned stones. And every time you do that, all the other people who did not will go below you, and you will rise above them to the next level. And that's how you get from level 1 to level 2. And the same logic is how you get from level 2 to level 3. And all those giant YouTubers and giant podcasters that you see just did this a billion times, or they got really lucky one time and they've just been doing it for 20 years and they've aggregated a huge audience. It is one of those two things always.

Do you feel that this is the same as building a company?

I think it's similar. The cool thing is I get to do 1% improvements every day. In the same way one would ideate a new episode, I ideate a new feature. I'm often getting to sort of re-edit my episode though. And I'm very lucky in that I can just update a feature to be better. I can just make a thing easier. So I don't have to get it right the first time, which you don't really get that opportunity with YouTube. Once it's out, it's out. That's the end of the story. And then you have to make a new thing to do it again. So the same logic applies to why I think Fycast does well. I am a student of the industry. I am a student of the playbooks. I Pursue at all moments of every hour of every day. What's working? What are people doing? What's new? What can I give to people? What new drills and brackets can I create? In the same way that somebody else could look at what intros are people doing? How is thumbnail style changing? What topics are doing well on YouTube right now that I can talk about? It's all being a student in some way. And just Being a student and being an iterator. That's it. Those are just the 2 things, I guess, that stay the same no matter what. It's just keep moving forward, keep making the thing better no matter what, in whatever way that takes shape. So it's different in some ways. I think it's probably easier on my end because I get to re-edit the video in a way. But yeah, I guess it is similar.

I'm assuming you have paying customers right now? How did you get your first, let's say, 10 paying customers?

Whoa. I think the first 10 was Twitter. Yeah. Usually, 'cause I have a little bit of an audience on Twitter and I had a lot of people when I sold Thumbnail Test that were like, hey man, you ever make any other tools, you call me. Like you let me know what it is, which was really cool for me. So I had like a handful of like a dozen of sort of lifetime Rocks product fans. And so those were Who I reached out to first. The first customer is a friend. The second customer was somebody that saw my tweet that had been following me for a long time. The next few were people from Thumbnail Test. And the next, probably that was probably the first 5, and the next 5 or 10 were just people who saw the tweets, people who had podcasts. We got some pretty decent ones pretty early. Like Greg Isenberg has been with us for a really long time. And that's just because more of a Twitter guy than a LinkedIn guy, you know, but it was easier this time because this is my second company. When it was Thumbnail Test, I did the same thing, but I had friends who were YouTubers that I had made over years or friends that were Twitch streamers. And that was where I got my first 10. And then I just, usually it's just talk about my thing. That's usually been my strategy. Post about the thing and make cool things. And I don't know, usually I've been lucky enough that if you build it, they will come. Although now I do actual marketing. So maybe I'm finally past that stage in life.

What kind of marketing are you doing?

All of it, dude. Sponsoring, I'm doing paid ads, I'm doing emails, I'm doing funnels, I'm doing conversion rate optimization offers. I have like 12 different landing pages. It's everything. Because Fightcast is big and complicated. And I have learned that Thumbnail Test was really easy to sell. It was, do you want to A/B test thumbnails? Yes or no? Yes. Great. Give me $20. And this is like, do you want to grow your podcast with better tools and save time when you're publishing and do all this cool clips testing stuff and have better ad tech and, and, and like I built a whole like operating system. And so it's a lot harder to pitch. And I realized that requires much more of like an education timeline, like almost micro demos over retargeting. And it's just more involved now. So a lot more to do. I'm learning.

How big is the market?

Depends who you ask, man. The ads market's a few billion in podcasting. The hosting market, I think, is probably like a couple hundred million, maybe $100 million in terms of like raw, like SaaS fees being charged right now. But it's complicated because again, most of these companies are making their money on ads. I'm not making my money on ads. So very different for me. But I also am usable by just like YouTubers. There are YouTubers who— excuse me, There are YouTubers who use Flightcast just to repost their stuff to Spotify and make money. Like there's a part of Flightcast that's just a syndication platform. So my TAM is a little differently calculated potentially, but I'm going through that exercise now. I'm recalculating my TAM, but I can tell you there are around, I think it's 600,000 active podcasts right now. So that have posted in the last 30 days. Wow.

Yeah.

Period. Globally. And then you take those, you go down to how many are doing video. And then you go down to how many have $50 and very quickly my TAM is much smaller, but we've gotten pretty big on enterprises lately. Like we just signed a 150-show network. We just signed another small network that has like one of the top 50 shows in the US. So things are, things are going very well on that side. We've built very advanced tools.

Yeah. Congratulations on the path. And I think this episode serves as one of the educational points because people need to know how hard it is. And, you know, especially, and a lot of the advice you give, you know, in this session, I think is towards maybe not the, not the new players.

It's hard. I can't, I'm not as good at new podcasts. I'm not. I build advanced tools for analyzing a ton of data, which is really hard when you don't have a lot of data. So if you have like 4 episodes, it's hard to know what works and what doesn't. You need to get through 15 to like 50. Before you really have signal through the noise. Clips are really helpful for that. Clips let you get a lot more shots on goal. But Flycast is generally built for existing podcasters. I am trying to build the tools that a good podcast needs to become a great podcast.

I love it.

Yeah.

How big is your team now?

Dude, big. It's like 12 people. It's crazy.

Bigger than I thought.

Yeah, I know. That's what everybody says.

How did you find your first employees?

Twitter. I had a really great Twitter thread. It went a little viral. It got like a few hundred thousand views and like maybe 1,000 likes or something. It was like a very simple Twitter thread where I just described the company. And I think I started it with, I'm hiring founding engineers for my new company. If you want to work on some of the craziest tech in the world, and work with the biggest creators in the world, work harder than you've ever worked, and have the most fun you've ever had, please read below. And then below was like kind of me just saying, this job's gonna suck. This is gonna be really, really hard. You're gonna work so goddamn hard. Life's gonna suck. Everything's bad. You probably don't want this job. But if you've read all that and you're like, screw you, I do want this job, then you can apply here. And I made my own website. It was startupdevs.com. I think it's broken now, but I, it was a giant like questionnaire, kind of like my own sort of Typeform of asking basically like 20 questions on what I thought made a good builder and what I thought made a good founding engineer. And then people would apply through that and then I would interview people through that. And then that's where I got like 80% of my hires, probably, probably 80% of the people I've hired to this day came through Twitter. And only recently do I have like an actual recruiter helping me find like more senior talent. That's a little bit harder to dig through. Also because now I'm in person in LA. So that made things a lot less easy. Twitter's more global.

Can I ask you to share your questionnaire for my audience? Because I think I have a lot of software engineers and machine learning engineers on the channel.

It got harder, I will say, with all this AI stuff. All right, we're going to speedrun. I'm going to tell you every question that I asked.

Let's go.

Okay. So here's the intro to the platform. It says, hey, this is Rox, founder of flightcast.com. I need a couple founding engineers, so I made this platform so you can apply to the job. Job applications are annoying as hell, so I figured I wanted to make mine more worthwhile. If you fill out this application, it won't just go to me. It'll go to all the people who ask me if I know some cracked devs, which happens all the time. When I give this platform to other startups, they'll be able to search your profile against basic filters. So be sure to chuck a ton of keywords and stuff you've worked on in there. This will take about 30 minutes. Please don't use a ton of AI. I'd rather you copy paste your resume. If it's AI, I will skip. Ready? Let's do this. And then the intro is, before you spend a bunch of time here, I made a quick set of questions to get a sense of if you'd be a good fit as a founding dev. These are subjective and based on my own experience. It will not stop you from progressing, but it will give you a sense of how likely you are to match the jobs coming from this platform. If you feel strongly about being a founding engineer, at the end of this, you can apply anyway, regardless. These questions are mostly for you to decide, do I want this kind of job? And then it's 10 questions that have a 1 to 5 rating. You rate yourself 1 to 5 in each thing. And then at the end, it tells you if you like didn't rate yourself very highly, it discourages you from applying. It's literally like, if you rated yourself like a 3 on all these things, it's like, you will not have fun here. These jobs will not be happy. These do not do this. And if you answer 5 to all of them, it's like, you're in. So the questions were, and it's strongly agree to strongly disagree. So 5 is strongly agree. I know one tech stack very deeply. I can build in 48 hours what takes many people weeks. I've built several entire products by myself. That's probably the most important one. I watch or read about programming in my free time. I'm interested in starting my own startup. I'm capable enough to be CTO of a company. That was another big one.

Yeah.

I'm willing to give up all side projects for the right job. Very startup thing.

Right.

I usually code 60 hours a week or more. So it's 60+. I've worked at several startups before. I've been an early employee or founding engineer before. I'm willing to take low salary in exchange for high equity. That one became less important after we raised. And I can pick up new stacks or languages easily. Those were the core questions. That was my founding engineer quiz. This is not get a default software engineer job. This is get a I am the greatest engineer of all time. I will change your company. I can do anything, anywhere at this company. And those are the people that I wanted to hire for founding engineer, because I didn't know what we were all going to do. You had to be able to build a Chrome extension one day and crazy deep video tech the next day. So there's a ton more in here that I can talk about if you're interested. But those were like the biggest high-level questions of if you answered yes to all of these, please apply. But it was like, you rank yourself 1 to 5. And if you didn't say 5 to, I've built several entire products by myself, very quickly that became like one of the biggest filters because I noticed a huge pattern between people who had built a ton of projects themselves and people who had not and their technical capabilities. But not just, oh, I did a little basic, it's one API endpoint weekend hackathon project, like products that you launched that had customers, that took a month or more to build that had complicated architecture or whatever. So there's my little, the dev side of things, my big 10 questions or 12, I don't know.

Is it self-identifying themselves as incapable and wanting to join this high-growth company?

All self-identifying. I'm not ranking you. I'm not asking you to prove any of these things. I just tell you that this is before you're even allowed into the platform. It's just rate yourself 1 to 5 on all of these. And if you get like, below a 70%, it's like, hey man, I don't think you're going to enjoy this job. Like, this is going to suck for you. But if you do enjoy it and it's like you get a good enough score on it, again, you could just pick 5 for all of them. You could choose to ace the quiz, right? It's not like a test. It is you identifying. Uh, but if you picked 5 for all of them, uh, or something, then you get into the questions where it's like, okay, cool. Tell me about the biggest project you've ever made. Tell me about the thing you're most proud of that you've ever built. And we dig into like a lot more of the depth of what makes a good founding engineer. But these were my fundamentals personally. One or two of these probably you can give, like, you know, take it or leave it. But most of them really helped me sift through applications.

How do you interview them? If they self-score high enough and then you find them interesting and then you meet them on a Zoom call, how do you interview them?

I had a trial task I would give, which is a kind of annoying frontend to build that was also just a bit of full stack. And then I added some bonus tasks in. That was after interview. If they did well on an interview, I would send them that. But actually it was 50/50. I changed my process a lot over the years, but generally I would just interview them and I would just try and understand who they were.

Okay.

Because you can prove the programming capability. So are you fun to work with? Are you truly, when you say I code 60 hours a week, Do you mean it? Do you want to fight? Do you have that grit and journey in you? Because if you don't, again, it's like you're just not going to be happy. And I don't want to give anybody a job they're going to hate. And so it was much more of a, tell me, I would ask people for their hopes and dreams. Maybe it's a personal thing, but I told everybody I wanted this to be the best job they ever had. And if anything gave me the sense it would not be for them, then I of course would not hire them. Because that's not— they're just not gonna have fun. And so now what that ended up with is like, you know, not to use the frickin buzzword, but it feels like we have, you know, a bit of a family, right? And like, that's what like the new people who have joined the team, I asked them even, I was like, hey, how do you feel about working at Flightcast versus like all the other companies? And they said like, it feels like more of a family. And it's just like, we're all grinding our asses off, but that could be really fun or it could be completely terrible. And I think that difference came because of like those kinds of questions that I asked in the interview. So I'd always be like, where do you want to be? What's the next step in your career that you want to take? Like, what's your next like financial aspiration? Is it like buy a house? Is it like get out of debt or something? And like making sure that the numbers I could offer would get them there. So I wasn't like undercutting them or being like, oh, sorry, I can't pay you enough to get out of debt, you know, or something like every little thing that I could ask, I did ask. Yeah. And then if they felt like they'd be great to work with, I would send them like a basic programming test. And even with all this AI, people will be like, oh, coding tests, not as good anymore. And I'm like, just make a really difficult frontend. AI can't do that at all. And if it can, great, they figured it out. I've been sent so many terrible tests, so many terrible trial tasks that are like, you know, ask somebody to build a pixel-perfect UI. That's the trick. A pixel-perfect UI that is a difficult frontend component. So we do like a timeline for like video editing. Uh, and that's like hard. You have to do the zoom in, the zoom out. You have to be able to drag things in, drag things out. It has to adjust. You have to be able to like all of these little things that if done poorly will run at 1 FPS. Like it'll just suck. So that basically made it so that I could do the interview as a very human thing. And then the coding test as the, okay, cool. You seem cool enough. I trust that you've built things. Now prove it.

Now build it.

Give me a day or two. Yeah. And if you could build it, then you're in pretty much.

And it's almost, it's almost something not vibe codable. Or, or if they figure out something that can actually vibe code it, that's fine too, right?

Yeah. If they figured out how to vibe code it, great. You're that good at vibe coding. You can build a really complicated thing. That's just you. That's just coding, right? At that point, like you built a complicated thing. It just happens to be with AI, but there are people who will ship me dog water. Like one time I gave a guy feedback and he replied to the email 5 minutes later. Saying, okay, I fixed it. But I had given him like half an hour of feedback on a Loom on his trial. And he replied 5 minutes later and I was like, cool. You took the transcript, you fed it into AI and you hit deploy. You didn't look at it and you sent me an email and you fixed maybe 1/3 of the things that I said and 2/3 of the things got worse. So cool. Not that guy. Right?

This goes back to earlier when you were talking about don't outsource all your thinking to AI. This telephone game of passing AI output from this AI to that AI and then just looping among them without reviewing any of them.

Yeah, or looping it to me, right? Just like you put it into the AI and then you just ship it to me and you haven't looked at it. It's like, come on, man, I got stuff to do. Like, what are we, what are we doing?

Right, yeah. Do you have one, one advice for me to become a better founder? Like, I'm building a media company, right? Podcaster, building products around it. Yeah.

I would study the market in terms of like the attention economy that you're in is a term I've always really liked. What does a TwoSet viewer see on their homepage? What are the thumbnails you're competing with? What are the topics you're competing with? And I would try and center the content that you make around the topics that are of interest now to help you get further off the ground. Generally, I'm not a big fan of like trend chasing, but there's trend chasing and then there's just being topical. And with this AI stuff especially, it's so topical, like week to week, month to month, you can get really, really deep in it and you don't have to be that deep in it. But if everybody's talking about how to do marketing with AI, like finding a guy who has done amazing marketing playbooks with AI, thinking of the thumbnail and title before you ever talk to him. And then recording the episode, knowing what's going to be said almost to some extent, at least the highest, highest level of it. And then shipping from there. I think that that will heavily improve your workflow and give you a lot more confidence that the outcomes have viral potential, that the episodes that you create have viral potential when it's like, okay, well, I saw 4 good YouTube videos about this topic in the past month that all got half a million views or something. I will make this podcast like I would make a YouTube video about these topics, but from an expert lens. And I will ask very pointed questions to get to similar talking points or meaningful value. So it is your take and your expression of a topic people are interested in, as opposed to, I think what happens a lot with podcasts is speaking generally about general topics to the point where there isn't a lot of takeaway. And so know your audience very specifically, know what they're interested in, find what's working in this attention market for that topic, and then talk about it. That would be my biggest suggestion.

Thank you so much, Rox. Really chatting with you today. I've heard of many people talking about building your thumbnails first before you actually make the video. You understand that it's a really hard thing to do.

It's why you have to do it.

It's almost like a different way of thinking, but I get it. It feels like, have you heard of the Silicon Valley, you know, entrepreneurs talking about sell first before you build?

Yeah. Yeah. I did that for so long. That was, I learned from Peter Levels. You know, I used to sell things I hadn't built all the time. So many of my projects were landing pages with a payment, with a payment button that had no functionality just to see if people click the payment button. Everything that we could sell always, always, always. But it's also, you can tell if the video is going to be good, right? You ask yourself, is this thumbnail and title good? You don't even need the world to like it. But you could show it to some people who would watch your videos. You could show it to people like me and say, hey, is this interesting? And do that 100 times until I say yes. And 5 other people also said yes. And they're like, great, how do I make this episode? How do I make this super interesting episode that people told me they would click on, that people I know and respect that I think are in the audience segment would click on. And then you're starting from such a position of power and confidence in a way that you were not before. It doesn't even have to start the episode yet. And then you can go and find the guest who could talk best about that topic, whatever that topic is, right? And you can be much more picky and everything. And that's kind of like, I don't know, some percentage of what Steve does, but also some percentage of just like what all YouTubers do. Outlier theory. Find what worked.

Mm-hmm.

Try and do your own version or angle on what worked. In a special way that's still you. And that combined with all the other strategies, now you're playing hardball. Now you're playing the game at the next level. And it sucks and it's difficult and that's why you should do it.

It sucks. I have to say it's hard. It's hard to be an entrepreneur. And I feel being a creator entrepreneur, it's almost the same thing.

It's so hard. It's only getting more and more difficult. The meta continues to evolve. Your job is so hard.

Thank you so much for all your advice. I really like it. And I'm thinking, of course, at least I can— there are multiple action points I can take home. This, maybe I can just test this with my YouTube community and just post it, post a fake thumbnail on it and just ask, what do you think of this one?

Post a community post and say, which of these would you prefer? And you can have them vote A or B of the thumbnails that you post and see what they say.

Yeah.

Thank you so much, Roxette.