Latest / Data Beats Opinion / Why AI won’t fix your marketing (and what will)

Transcript

Why AI won’t fix your marketing (and what will) · Data Beats Opinion

• 0:00
  Speaker A: AI is trained to be the best average it can be.
• 0:04
  Speaker B: It's not an insult.
• 0:05
  Speaker A: That's the design. It's trained on an enormous amount of data, so it can be useful to as many people as possible. Which means average is the default output and average is what most people are shipping right now. My guest today has been working on how to get above that average line since 2022. Brandon Shelton drove a record $28 million year as CMO of Gear Bubble, then took over as CEO and scaled it past 165 million in total revenue. Now he builds AI systems for marketing growth teams. In this episode of Data Beats Opinion, we get into the why I Vibe Coded it in 20 minutes usually means nothing. The boring folder structure that does more for your output than any other prompt and the prompt that Brandon uses to stop AI from agreeing with him all the time.
• 0:54
  Speaker B: Let's start the show. Brandon, super glad to have you on with me. Really excited to talk about this. And yeah, for people who don't know, tell us a little about yourself.
• 1:10
  Speaker C: Yeah, no, for sure. My name is Brandon Shelton. I've been doing marketing since 2007, but full time since 2010. Most of my background has been with software companies, B2C Space and B2B Space as either CMO, a chief growth officer, or a CEO.
• 1:27
  Speaker B: Yeah. And we met a while ago. 2015, 2018, something like that. It's been a while since we've or we've known each other a while at this point. And it always impresses me because you're always at the forefront of everything and as someone who's generally not, I'm always kind of a, a come lately kind of person. Like that always impresses me and I always, I'm always interested in what you're
• 1:52
  Speaker C: doing and I appreciate that.
• 1:55
  Speaker A: Yeah.
• 1:55
  Speaker B: And so let's talk about what you're doing right now. So you are, you're doing Buddyprompt.com right now. And I think like a lot of people were in the fricking AI space at this point and for good or bad, this is the conversation I think we're going to be having for, for a while now. And there's so much hype and I think that everyone is interesting. It's the kind of curve that I see because I was in this and then I saw parts of my team get in this and then even Scott, who's like the huge curmudgeon on my team, he got into this and it all starts with like, no, this is stupid. And then they all hit that hype. Train. And it's like, oh my God, everything's great and then everything kind of levels out. But like, there's all this AI advice, this noise, this garbage. Is it good, is it bad? Like, did you get caught up in this? Like, how did you, like, what's been your kind of like AI roller coaster ride, if any?
• 2:55
  Speaker C: Yeah, no, for sure. I mean, it's. So I actually started taking it seriously in like 2022. And at this time I was a CEO of a software company and we were trying to integrate it into some of our product suites and also integrating it operationally. And during this time it was like early ChatGPT where, you know, it's terrible at, you know, marketing, writing and stuff like that, but it still did a lot of cool stuff where you could see the value in it. And luckily, like, during that time, you know, the tools that we were using, a lot of it had to do with like machine learning, image detection stuff where we will extract text from images and then using ChatGPT to take that text and, you know, kind of rephrase it type of stuff. So that technology happened to be a little bit better at the time, which a lot of people don't realize. Like, that is still AI, you know, because now it's. We kind of look more of it as like, you know, robots or using it for marketing copy or, you know, images, like creating imagery and things like that. So, you know, during that time I ended up seeing the opportunity just from working with it and realized like, you know, like, this is going to be really big, so I need to like kind of dive into this. I actually resigned from being CEO and started headfirst into like doing consulting work, fractional work, leveraging AI and then also creating my own AI tools. And because I already had a background for, although I'm not a programmer, having a background with software development, kind of just understanding how it worked and hiring so many different developers. Like, you know what, let me just like try to see if I can like build some stuff myself. So once I kind of just started vibe coding and then I realized like, okay, like I can see where like the value is. I can see where this is going. And then over time you start seeing all this new technology come out. You see openclaw, you see all these different things. And I think early on it was very easy to get overwhelmed and it still is now because things are moving so fast. But over time I just started developing a system to understand like, okay, you need to stay up to date with knowing what's out there. But being able to, to filter out like what not to go down the rabbit hole on is just as important as like what you actually decide to spend a lot of your time on.
• 5:06
  Speaker B: And this is something that I don't know if anyone, if any of us expected how fast this was going to move. So we started our first AI project in Segmetrics was around the same time, 2022 and we built an internal AI insights. It was built on early OpenAI chatgpt stuff. It was not great. I, I'll be honest, but I saw the, and we were starting to look at our own models, we were looking at rag models, we were looking at all this individual training stuff and we had to put it on, on hold for a little bit before while we were still in development. And I come back, I think three weeks later and the entire tooling set is different. Everything is different. Like none of the stuff that we had done three weeks earlier was the correct way to do any of this anymore because it had just moved so fast. I was like oh my God, like how are we going to ever stay up to date with this? And how do you feel like, do you feel like it was faster two or three years ago or do you feel like it's faster now? I think there's more visibility now. But what do you feel about like the change in the underlying structure of what's going on?
• 6:12
  Speaker C: Yeah, I think the new technology is probably still just as fast but I think the, the overall foundation of it is starting to stabilize a little bit because I think that you know, early on obviously open AI was so dominant and as these other players started coming in, I think they're starting to understand which sections they want to focus on. So you can, you can very clearly more now see like OpenAI is really trying to be a consumer based platform. That's why they ads, they're really trying to go after that market. Google is trying to go after that same market but it's just they have a different business because they already have the Google behemoth behind them. And then Anthropic is very focused on B2B. So like now you can see those major companies verticalizing and then you see all the other players like lovable and gamma and stuff that are specific and specific verticals where I think now there's kind of becoming that layer of foundational things that's kind of in place and they're building on top of that.
• 7:09
  Speaker B: Yeah, and I think it's been interesting watching the niching down of different places like, like you're saying OpenAI is really trying to go broad as a broad to consumer. It has everything in it. I don't think any single thing it does is specifically better than a competitor that's more niche down. But because it has your entire tool set right there, that's a huge value for B2C, I think, in a lot of places. Where do you see kind of the AI being used in market? Because you're, you work in the marketing space, you are, you're in the sales, you're in this B2C world. Where do you see AI coming in and being valuable versus people essentially just bringing out slop? Because there is so much. And I think we're, we're seeing a growing pushback against AI slop, but some people are better at seeing it, detecting it. But just before you answer, just a quick story. My, my wife is watching this YouTube video that's AI and I only know that it's AI because I know the person in the video and that's not his channel. Right. So someone had stolen him and used him as an avatar to do whatever. Right. My daughter just walks by, glances at it and says, mom, why are you watching AI? Like, that's the, that's. And my wife had no idea. Right. So that's the. It felt like there were like three different levels of understanding the AI. So, like getting back to the question, like, how do you see this affecting marketing when some people are very tuned into what is AI, some people aren't. And kind of that cognitive dissonance.
• 8:48
  Speaker C: Yeah, it's interesting, I think, because it's so early on, there's going to be that wide discrepancy and a lot of it has to do with like the audience that you're targeting. Like, if you're going after more B2B, if you have a business that you're like selling anything in the B2B market, you're probably going to see more resistance because they're probably going to be a little bit more educated on the AI as a whole. But for B2C stuff, you know, if you have like some of the obvious things like a ton of EM dashes in your copy or just like super weird things happening in videos, like people are like, now they'll be more skeptical. But overall, like, I think for B2C, especially if you're doing marketing to that, to that segment, it's probably the bigger opportunity right now in terms of the sophistication part. But like, it's, it's interesting. Like even the people who I See, might have like big personal brands in the B2B space, stuff like that. A lot of them are using AI for writing and it's obvious like people know like you can tell from, well, people who understand, like how, like, because everyone's using Claude for writing, you can kind of tell the structure of the sentences and like it makes a lot of people start sounding the same. But like you'll still see some people with huge audiences just because they either got it early or they've understood the type of content to use with AI. So I think a lot of it is people just understanding like obviously how to get better with AI in terms of injecting like your own personal styles of things. Like a lot of customization and just having a nice workflow will help. I think some people just don't take the time to, for example, like using writing as an example. If you want to have AI write more like you, most people will either take some of the things that they've written and kind of feed it to AI and that's kind of like where people stop or they might have AI ask them some questions and that and that's where they stop. But like, if you really wanted to understand like how you write and how you talk and how you just communicate overall, you probably should take some videos that you've done and get transcripts made. You probably should take different types of conversations, like personal conversations on your social media, that's not business based. And then personal emails, maybe grab some texts and then also take, you know, professional emails where you're talking to colleagues. So you have a variety of different types of sources of communication that you've had and feed all of that to AI and then have AI bucket it. So that way it's like if you're doing a more, you know, a personal email to someone, then you can use AI for that and they'll know you like your tone for that. If you're doing an actual, you know, video sales letter script where you're going to be talking on camera. Oh no, like this is how you talk when you're on videos. So I think a lot of it is just understanding how to just go a little bit more in depth to get that customization so you can sound as little as AI as possible. I mean, sometimes it's going to be hard, but that's really what it boils down to.
• 11:35
  Speaker B: Yeah, I think that. I fully agree. I think that the amount of examples and real examples, not just like I talk like this, but having those actual examples helps a lot. But I think that there's, I think you're exactly right where there is maybe a lack of care in making sure that your voice comes through. But I think that there's also a lack of understanding about how much needs to be in there and what type of stuff needs to be in there. Because I've done the thing where I've taken all my social posts I've ever made pre AI, stuck them into AI and like, hey, talk like me kind of thing and it still doesn't get it right. And so it's like I've written a book. Should I put the book in there? Like, I think this is where a lot of people, especially like if you're following people on Twitter or like what's going on on LinkedIn and stuff like that, there's a lot of hype around what is possible. But when you get onto those one to one conversations with people, you realize that no one has, I won't say no one, but a lot of people have not gotten the flywheel working yet. Like, what is that process? And I think that's where, I think that's where a lot of people struggle. You know what I mean?
• 12:53
  Speaker C: Like, yeah, no, I agree. It's like, it's people who share like different tidbits on how to do certain things with AI. It's always presented as if like this is the framework and once you do that, you're done. But that's not how I operates. Everything you do that you're trying to, even, even if you're creating a workflow framework, if you're trying to teach it how to write like you, it's going to be a living thing and the more you give to it, you're still going to have to have conversation, have it write stuff and still add to add tweaks to that ongoing. So like that's just like the base foundation. Even like, like I have a voice DNA document and it's like, even with me, I have to, I'll have it write something and then I have to say like, oh, oh, these words. I wouldn't use that. Add that to like a do not use these words list.
• 13:42
  Speaker A: Yeah.
• 13:42
  Speaker C: You know, so it's like it's not something where you can just stop and let it go. And sometimes I'll even write things manually and then use my voice DNA document to edit my writing because then that way I know for sure it's going to sound like me because I did the first draft, you know, so it's like there's all these like little different ways. But people, it's not as sexy to say that because it's like, oh, I, I have this special way you do these 10 steps and Claude's gonna write just like you. And you can press a button. It's gonna do all your social media posts and you're gonna get all these people gonna sound just like you. It's like you might get traction from it, but it is probably not gonna sound like you if that's the way you're doing it.
• 14:19
  Speaker B: Yeah. I, I have also found like there are those magical moments when you can one shot something and you're like, wow, that is the best script. Or wow, we just put that, that software together and it just worked. And it's like, but those are very rare. It always feels like this is a first draft of something. And I mean we're, we're redesigning our top page right now and I'm using Claude to help figure out new ideas for what we can do with it. And it's, we've been going on it for like two weeks. It's, it's an iterative process. We can't just go in and say, Claude, make me a, make me a website. And it doesn't. Right. It's this. You have to have taste to it.
• 15:01
  Speaker C: Yep.
• 15:02
  Speaker B: And I think that none of the AI has real taste, but it has a lot of good ideas. It's just understanding which ideas are the, the right ones for the, for what you're trying to do.
• 15:13
  Speaker C: Yeah. Especially like when talking about building things. So it's like even like when I created Buddy prime, but it was more out of necessity, but it was like a tool where like when people say like I vibe coded something in two hours or I vibe coded something in 20 minutes, whatever that time. Like, like that is a relative statement. Right. Because saying that I vibe coded something and it works, that's different than saying like I created a software that has a database that has security, that's actually live on the Internet published and people can really log in and use it. Like those are two different things. So like sometimes like people may hear someone say like, oh yeah, I did this in 20 minutes. And it's like up and going. It's like, yeah, when I tried to do that, it didn't work that way. It's like, well, it's because it was relative. Like it might, maybe it worked locally on their desktop and they could use it, the tool that way. But if you're trying to build something to sell, there's a whole nother set of steps you know, that you need to do to be able to actually launch, you know, an actual software.
• 16:13
  Speaker B: And I think, especially with Vibe coding, not to get too technical with it, but I think that there is a bias towards solved problems. Right. Because most of people probably listening to this are not developers. But I remember back in the day, we had frameworks. Frameworks would come out and people would be like, I rebuilt Twitter in a weekend. And it's like, yeah, I mean, it's not too difficult, right? Like, you're a developer. And now everyone's like, I Vibe coded this in 20 minutes. The software, like the base feature of, hey, I can tweet something and it shows up on a box or a to do list is not the challenging part. It's like what you're saying, it's security, it's marketing the damn thing, it's building all this stuff around it that is difficult. I was on this call and they were talking about AI, and one of them said, yeah, what you should do is you're just going to be amazed. You go in and you say, build me Tetris. And it will just build you Tetris right there. I'm like, yeah, there's like 800 Tetris clones on GitHub. I can build it in 12 seconds by git clone. Tetris solved problems.
• 17:22
  Speaker A: It.
• 17:22
  Speaker C: It's.
• 17:22
  Speaker B: And it goes back to the writing, right? Because the solved problems, the. The bare minimum that genericness is so easy for AI to do.
• 17:32
  Speaker A: It's.
• 17:33
  Speaker B: How do you uplevel from there?
• 17:35
  Speaker C: Yep. Yeah. No, for sure. Like, it's. I feel like a lot of the time that I wasted early on was not like, trying because I didn't fully. I was first really diving into AI, didn't fully understand that AI was not as good at that. So it's like, you spend so much time prompt, like, why isn't it not like elevating this? Like, right then you. Over time you understand, like, no, it. It's literally trained to be the best average that it can be. Like, that's what AI is trained on. Like, it's trained to be able to put it in the hands of a lot of different people. And it's trained on large amounts of data. It's. That's how it's trained.
• 18:14
  Speaker B: Yeah.
• 18:16
  Speaker C: It's supposed to operate this way so a lot of people can get use out of it.
• 18:19
  Speaker B: Yeah. What is. If you're making a sentence, what is the average next word that comes after this sentence for this context? And this is, I think, exactly where I think People fall into the trap of where is AI creating leverage versus it's just kind of busy work and spinning right. And I've seen a lot of people get stuck in kind of this. I'm going to get the AI to just work and do it and then it's going to be all great. But I'm working on the AI more than I'm ever going to spend actually writing an article or anything. And then everything changes two weeks later. So where do you kind of see AI bringing leverage to marketers and to businesses rather than just like navel gazing AI for AI sake?
• 19:07
  Speaker C: Yeah, for sure. So I think for me a lot of that changed once I started using a local folder structure. Like some people like get really complicated and have, you know, LLM wikis and you know, second brains and all this type of stuff which people can build out. And I have similar things as well. But like really you don't necessarily need that to get the most leverage from it. Especially as a marketer. What you really want to just have is a clean folder structure where from first, like you have clear naming conventions. So AI can find things fast, faster than it would be for you to click through on your computer and find these things and have the context it needs for the specific tasks that you need. So like, you know, as a marketer, most people will have either frameworks that they use themselves or frameworks that they learned in the course, or frameworks that you know, someone taught them and that they adapted or whatever that may be. So like the first thing you want to do is like what? Make sure AI has access to those things. So you either create skills or if you want to be even simpler, just create a markdown document, put that on your computer you like and you can keep it that simple. So it's like now you have these frameworks and let's say you're an email marketer, you know, so whatever your email frameworks are, you have that there. And then now it's like, well, if you were doing emailing, like what's the next context you would need, you would need to know like the audience, like who you're speaking to. You would want to know like what's the goal of the email? Are you trying to sell something? Are you trying to create content? So you kind of just go through the whole process of what you would normally do. And that same information that you have in your head that you know that you would need as a marketer, you want to have that in your folder structure. So like, like that's that's really the, the way to get the most leverage. Because it's like you can have no your company or a company you're working for, or if you have clients, you know, you have that company client folder. And then now there's a subfolder that has, you know, marketing information. There's a subfolder that has general company information, like the company name, the company URL, the company's mission statement, what their business model is, maybe the name of their products. You know, just basic things that you would know as you're going along doing stuff. And you just want to be able to have that in an organized way. And that's how you can get the leverage. Because like now when you start doing stuff, it's like, you don't have to go find all these different things. It's like, hey, I want to create, you know, email marketing strategy for XYZ product. And we're trying to hit, you know, $50,000 in the next 30 days. So then it own, it'll be able to go ahead and grab all the context. And it's like, okay, now I need you to help me plan, you know, how many emails should we send in the 30 days, how frequently? What are the angles and stuff? What are the angles and the hooks or that you think that we should maybe incorporate, you know, inside of this email strategy for these next 30 days. And then it can help basically use it. You're using as like a thinking partner. I feel like that's always the best way because you're going to have your ideas anyways. But if you're leveraging AI to help you think about things in a different way or contribute other ideas, then you're able to come up with better strategies and also execute on the writing itself even faster.
• 22:17
  Speaker B: Yeah, I feel similarly, where I feel AI is a better rubber duck or a conversation partner than it is like a go, like a go do it type of thing. Right. But one thing that you're mentioning there, I think is really important is that especially for creators, coaches, people who have been doing this for a while, and especially agencies, we have a huge library of SOPs. We have a huge library of how we do things because most of us have employees or we have contractors. Contractors, or we have someone that we are delegating these things to. And good companies traditionally as well have SOPs. They have processes, and AI is really good at understanding those. So if you can organize for your teammates and for your employees and your contractors and for people on your team, then it's really the same skill set to organize it for AI. And, but I think one of the challenges that people get to is that okay, now I have all this content, but I still have to organize my prompts. I still have to organize like how I'm talking to AI and especially, okay, now I'm in a team or whatever, like, and I think this is what kind of spurred on Buddy prompt, right? Like, how do you organize and keep all this stuff available for AI? Because it's, it is, it's, you know, I often talk about we don't have a content quantity problem anymore, we have a content quality problem. And AI can create infinite amount of content. So then what the heck? How do you organize and structure this, you know, 100%.
• 23:58
  Speaker C: So yeah, that, that's definitely the reason why I built Buddy Prop. Because as I started building out these folders, I realized like, okay, well, first I had like a rigid way of like, this is how I'm going to set these folders. You know, it's going to be like a company level folder. And then I'd have the strategy folder, the marketing folder, research folder, and then subfolders within that. And I realized like, well, AI can just organize this for me. So what I started doing was I would have the company folder and then I always would know, like just the raw company information like that I would always start with. So like, you have to just give like mission statement, business model information, strategic goals, that type of stuff. You know, you want AI to understand how, like if there's a product suite, like how the product suites fit into the business school, like that type of stuff, and I would create that document and then the rest of the stuff I would have, I would just throw it all in a root folder and have AI analyze it and say, hey, what's the best folder structure? Like, I want to use this to basically be my thinking partner to do, you know, XYZ marketing work or XYZ Growth growth work. And it would just tell you like, hey, this is the best folder structure. And then you, if you disagree with a few things, you can kind of tweak it. And then it's like, all right, this sounds good, go ahead and put it in the folders for me. So it'll go ahead and it would create you all the folders. And then what's really good is that AI will create these markdown documents inside of each of the folders and in the root directory to tell itself how to navigate things. So that way if you have a lot of files, every folder it goes into will have like a read me that says like, hey, inside of this readme. Like, this is how you use it. So it can find things even, even cleaner. And then from there it's like, okay, well, now that I have all these things set up and I started creating skills for specific frameworks and stuff, I'm like, there's certain prompts that I use for processes, like, where do I put this stuff? So I had it all, like in Google Docs and I'm like, copy and pasting all the time. I'm like, this is like too much. Like every time I need to do research, I have like a set four step prompt. And like, it was just too much work to go copy and paste it. So that's why I create a buddy problem. Like, I need to be able to put all of the prompts that I have in one place. And then if I'm in a browser, I can access it in the extension. If I'm in, you know, cloud desktop, I can access it through the MCP where I can just say like, hey, grab my three step prompt for research. And it could just pull in all three prompts and execute it. And it's already going to have the context because it's on my desktop. It has. And it has access to like my folder structure. So it saved me so much time. So I'm like, I. All these things I spent so much time optimizing, like, I don't have to worry about this anymore. And it's like, if I do want to improve the prompt inside of my cloud desktop, I could just say like, hey, for prompt number three, add this line. And it would just connect the buddy prompt and update it.
• 26:37
  Speaker B: For me, this has been the superpower that I've seen with Claude and stuff like that is that Claude and all AI can improve itself, right? Just like you're saying. It's like, oh, this prompt keeps doing this thing that really annoys me. You just ask Claude, you say, hey, why do you keep doing this thing? And it'll say, well, you didn't say not to or whatever it is. And I say, okay, put that in the instructions, put that in the prompt so you never do that again. I forget what it was. We were doing some code and it was always doing this really weird thing. I was like, why the heck are you doing that? It told me and I said, make sure we never do that again. And it did it. And it's like, this is great. And it goes back to the iterativeness of this. Everything that AI is bringing out, you need to consider a draft and it's just. Okay, how do you polish? How do you polish each time?
• 27:25
  Speaker C: Yeah, no, for sure. And it's like, not even. I mean, this will get a little bit more into the weeds. But like, even the way that, like how AI works on a fundamental level, like, because some people will say like, oh, like just if you just put that in the folder, like if you tell AI, like, hey, in the instructions, don't not do this, then it should just not do it. But it's like, well, if you like just from a chat perspective on how AI works, like, AI by nature cannot remember everything. Like, that's like, that's the way it's built. It can't. It's just, it would be too much data and it will make it hallucinate. So, like, as you're chatting, AI will read the instructions and we'll try to do it as best that it can, but the longer the chat gets, it's. It's taking summaries and it's passing it down to itself. So like earlier in the chat, you might have given it specific things to do, but it's not going to remember that. So like, that's where prompts still are very important. And that's why, like, sometimes if you. People who like, oh, I'm using autonomous agents for this, and they'll get frustrated because it's like, I put this in instructions to not do that. It's like, well, that's just how, that's just the nature of how AI works, right? So you're gonna have to have like these clear prompts to sometimes just re. Inject that context back in or retell it. Specific things you don't want it to do, or the specific things you want it to do, that's just like, currently, that's how AI is, right?
• 28:45
  Speaker B: Yeah, it just doesn't have the context window. I think it was something. Even though CLAUDE will go up to a million tokens or whatever, after about 100,000, it starts getting real dumb real quick. And they've done practices like, yes, you have this context window, but it, it doesn't. It's like remembering a number that's a thousand digits long, right? Like, you might be able to do it, but like, or you might be able to look at that number, but you're never going to be able to
• 29:12
  Speaker C: Remember at all 100%. Like, even when you're chatting, like when it, when you see Claude specifically, if you use a cloud, when it says it's compacting the conversation, it's Literally writing a summary so it can remember at least as much as it can of that part of the conversation before you keep chatting. That's literally what it's doing.
• 29:31
  Speaker B: Yeah. The first line of all my Claude MD files is call me like Ishmael or something. I have a code name, right? And I'm like, always refer to me as that name. And so I know when it stops referring to me by that name that it's forgotten that part. And I'm like, okay, we need to review and do this again.
• 29:51
  Speaker C: That's pretty smart. So what do you.
• 29:54
  Speaker B: I think agentic stuff is something that is on everyone's mind. I think it's like the holy grail for especially marketers. Right. And I see a lot of people dipping into it. I see a lot of people wanting to get there. There have been very few people I've followed that have been like, hey, yes, I have agentic stuff working in a repeatable way. And let me show you, like, you know what I mean? Like, how do you do you do. First of all, do you do much stuff agentically?
• 30:26
  Speaker C: Yeah, I do. I don't use a ton of autonomous agents just because I like to stay in the loop for approval for a lot of different things, but things that are like super task based, where it's like grabbing data, like that type of stuff, like it's fine, I'll just let autonomous agents run it or even simpler sometimes just have like an API running in the background to grab information. But yeah, I do use agents quite a bit.
• 30:50
  Speaker B: Is that something that you kick off or is it like okay, every day at 2, go do your stuff kind of thing. Like how are you kind of setting that up and using it?
• 30:58
  Speaker C: Yeah, so like a good example would be like for like meta ads, for example, I know like I'm going to, you know, look at data daily, look at data on a seven day basis, 14 day basis, 30 day basis, 90 day basis. So like in that situation I'll set up an agent or sometimes I'll use Claude to create like an API or CLI connection to grab that data on an interval for me so I don't have to remember. So that's like a situation where it's fine to use an agent or just any automated running task inside a cloud in the background, but when it comes to like analyzing the data, so I'll have, I normally will. For me I like to be in the loop for that stuff. So that is like the data will already be there so I don't have to do worry about Grabbing it and it normally will be from multiple places. So it's saving me more time because it's like if you're running Meta, you're probably using like a tracking solution like Segmetrics to track it. You probably have, maybe you're using like Google Tag Manager as well. Like you probably just. Or you might have a backup tracker like with Google Analytics on top of Segmetric. So like all these different things, I'll have all of that being pulled or if you're using like Microsoft Clarity for heat maps. So it's like I, I'll use that agent to go grab from all the platforms under the interval and bucket it. So like I'll have a data folder and in my data folder it'll have like, okay, this is the Google Analytics, this is Segmetrics. This is, you know, Microsoft Clarity. So I can just grab and put all the files there and then within those subfolders are, I'll make sure that it's dated and make sure that like I have it like per month. So this way again kind of goes back to you want to help AI find whatever you need as fast as possible. And it's like I always say, like the analogy of like, it's like your desktop computer. Like imagine taking the 50,000 files on your computer and just putting it all on a desktop and then you personally trying to find it. It's going to be.
• 32:43
  Speaker B: So you've seen my mom's desktop as well.
• 32:49
  Speaker C: That, that, honestly that is like the, that was definitely an old school problem because my, my mom, she was similar. Like a ton of things on the desk. I'm just like, it's even, even if it's you, like you're going to have a trouble finding it. So imagine trying to tell a robot, like, go find this. That's like, that's essentially what you're doing if you don't have a clean structure. And again like you can have AI create it for you but like, you know, you want to just have it nice and granular and set that up. So yeah, I'll use the agents to grab that information, put it in the right folders, make sure that the naming convention is good. And then once I'm ready to evaluate things, then I'll have like different skills and different agents helping me. Like I'll normally always start with my own hypothesis. I just don't give it to AI because I know it's going to steer the AI to, because it's, it's also trained to be agreeable so like, I, I don't, I normally will start with it giving me its opinion and then I will give my opinion and then I have a prompt called my brutally honest prompt where I'll basically have it like, make sure that it's giving me like, it's honest feedback, make me earn its agreeability instead of it just agreeing with what I said. And then that's normally how I'll start determining, like, okay, based on all this, like, this is what I'm going to do.
• 33:55
  Speaker B: Yeah, yeah. If you, if you start off with like, hey, this is what I'm. This is how I'm thinking about it, then it poisons the entire conversation. I always start with what I want to do and any limitations that there, that exist and that's all I give it. It's like I want to create a video to put on my homepage. I cannot do X, Y and Z. Right. I don't want generated. I don't even know at this point. But like, I'll give it those limitations and then it tries to figure out how to fit within those limitations instead of me saying like, I really want to use recut for this or like, whatever. And then it's saying, well, here's how you'd use recut, even if recut was not the tool to be used in that situation.
• 34:39
  Speaker C: Yeah, no, 100%. It's a subtle thing and it's very important, especially if you're doing marketing and growth work. Like if you're working with a client or working on your own company and it's like, oh, yeah, we need to do a paid media strategy. But if you go to any type of AI and you're like, hey, yeah, so like, what's the best way to use Meta and Google? Well, now you're telling AI indirectly that you think that it should use Meta and Google, but maybe that those weren't the best platform. Maybe you should be using TikTok, maybe you should be using, you know, native advertisers or something like that. So like, even like the framing of that question, you would want to start with something more generic where it's like, we need, you know, a performance marketing strategy. Please tell me, based on the business information that you have in this folder, what are the best channels to use and what would be the budgets needed for those channels type of thing. So you're starting from a more holistical standpoint and you're not steering it at all.
• 35:29
  Speaker B: Yeah, yeah. And you can. And in that conversation you can always drill down and say, well, I Kind of want to look at these things, but you've at least given it that initial idea of starting with everything, right? And then you are steering the conversation instead of starting out and never having those, those other options available to you 100%. So what do you, you know, we've talked about a lot of different ways to use AI, a lot of different towards content generation, towards analysis, towards vibe coding, towards all these things. I think a lot of people get overwhelmed by the number of things that there are out there, the number of choices that they have for smaller marketing teams. So people who don't have necessarily the budget, who, who really want to get the most value out of AI in a small team, like, where would you recommend that they start? Like, what would be the thing that you would say, what you really need to do is this.
• 36:24
  Speaker C: Yeah, that's a good question. I think it always should start with like, what the actual, what your actual goal is first. I think that's very important. And really honing in on like what, what you're really honing in on, like what the clear end goal of what you're, you're trying to do. Because sometimes you'll see like, hey, we want to grow sales and we think, you know, we want to do paid acquisition. But it's like, did you, did you really bulletproof bullet? Did you really stress test, I should say, did you really stress test understanding? Like, are those the viable channels for you to even, you know, work that? So I think you should always start there. You want to kind of treat it more like a growth and first principles mindset where you want to really think deeply on what you're trying to accomplish. And is the strategy you're thinking about doing, does that actually really fit, like your resources and, you know, where you are once, you know, have gone through that, you can also use AI for this. I actually have like a growth engineer that helps me with these type of things where it's like, hey, this is, these are my constraints, these are the budget constraints. This is what we have team wise, yada, yada, yada, like, we're trying to hit this goal first. Is this realistic? You know, like, and if it is realistic, then how can we get there? So then once you have that, then you can take that and determine, like, what are the best tools to use with AI. And I would always say, you know, always ask your question, like, can we do this manually? And if we're going to use AI, do we need to use it autonomously or do we just use it, use a chat base like you kind of want to start with really basic principles. A lot of times like I'll kind of look at it from, I'll write out a workflow and that will consist of the steps that I need to do. So let's just say like I determine like, okay, I'm going to be doing like a paid a Google search campaign and that like that's the best medium to do. So then I'll be writing out the steps of what we need to do as far as, okay, well we need to do keyword research. We need to, if we're going to use like Google Display with that as well, we'll need some images. Then we'll need to create landing pages, you know, kind of just the steps of, you know, that, that we need workflow wise. And then from there you can determine like what AI tools you can use to, to actually accomplish these things. So like, you know, when it comes to any type of marketing writing, definitely cloud is the best for image generation. You can use the new Chat, GPT or Gemini for that for anything that's like analyzing data. Most of them now are pretty good at it. So you can kind of choose between cloud, Gemini or ChatGPT. I prefer cloud, but any one of those things. So like now that you know, okay, well for each of these steps this is what we can use AI for. Then you can determine like whether any of that can be automated or not. And it doesn't have to be automated with AI. It can be using Zapier, it can be using make.com, whatever that may be. Like is any part of this workflow that I've mapped out manually, like can this be automated? And then if maybe certain parts can, maybe certain parts can't. And then the last, the last question you should be asking yourself is like, should I be using an autonomous agent for end of this automation parts like that literally should. You shouldn't be starting there. That should be the last thing. Because if you're new or you haven't really incorporated AI a lot, you could end up spending so much time setting up the agents like you, like you kind of talked about before, like the actual goal of what you're trying to do was, you know, get up a paid no Google search campaign to, you know, 3x your sales or whatever that may be. You're going to waste so much time just getting started on new technology that you haven't used yet. It doesn't make sense to start there. Just start, you know, start, start the other way around. And as the campaign's running, then you can go back and, and look at, and learn more about autonomous agents and see if you can, you know, implement it to, to speed things up.
• 40:17
  Speaker B: Yeah, and it's tempting because it's a form of procrastination really, to get the, oh, I'm going to make this all automated and agentized. I've done this because I'm trying to get better at social media right now and I have built, built so many tools to help me schedule and like manage my social media. And Claude, at one point I was building on one of these tools and says, Keith, are you just procrastinating writing these posts? I'm like, yes, yes, I am. But I want it to be a system. And this is, I mean, this is one reason I love talking to you about these types of things is because like you said, you've been doing this since 2022. You have this systematic view of AI and the tools. Like everyone's on the AI hype train right now and talking to you is like, it's a breath of fresh air because you're not on this hype train. You're like, let's look at this systematically. Here's what AI is good at, here's what it's not good at. Let's see where we can actually bring it in. I find that to be a really good strategy. How do you feel like that has played out for you and are there times when you're like, oh, I wish I was on the hype train a little bit more or like, how do you feel about that?
• 41:31
  Speaker C: Yeah, no, I think for me it's worked out really well. Like I. Because I've worked at a lot of companies doing fractional marketing, fractional growth work. I'm actually working with a company now to actually help them build like an AI marketing tool while also helping them like iron out some of their, the client stuff that they're doing as well. So it's like, for me, it served me really well because I always balance between, between staying up to date with what the latest and greatest is. So I would know and like a good tip. So like, if any listeners use Perplexity, they actually have. I don't know if they still have, but they used to have like a free trial a whole year that you can get it for free link. I don't know if they still have that, but it's worth the money even if you don't do the free trial. And they have like a tool called Tasks where I have three sets of AI news prompt prompts and what it does is every single day, it grabs, you know, AI news for me across different things, like stock information announcements, white papers from, like, you know, the more PhD type people in the AI space, like, just different categories of stuff. And it, it sends it to me every morning. And then I have a weekly roundup prompt that will, you know, round out the week. And then I have a monthly, like, deep dive one that kind of gives me like a full on, like, research paper for the full month type of thing. And that alone helps me because it's like I can just pop over there. I'll kind of look to see if there was anything interesting or anything related to what I'm doing as well. So that way I can filter out things that is not directly related to what I may need, you know, to learn. Right now, if I see something interesting, I'll read an article or two and then I'll determine, like, okay, do I need this right now? And I always, like, try to put that framing in my head where it's like, oh, wait, this is like, when Open Clock came, I was like, wow, this is really cool. But I just set up this folder structure that's working really well. Like, do I take the next two or three days, like, trying to learn this tool that at the time was super unsecure? And I'm just like, no, it probably doesn't make sense, you know, so it's like, you know, I'll watch a. Watch an interview with the owner, you know, passively while I'm, you know, doing driving in my car, stuff like that. I'll read some articles, like, fully understand the technology, but then I'll still stick with what I'm doing. And then at some point, once it is at a point where it makes sense for me to do, then I'll go back and like, all right, well, now I already understand it, so I know, like, all right, now I can kind of dive in and spend that time. So for me, it's worked really well. Because honestly, I feel like when you jump from tool to tool and you spend so much time going down the rabbit hole, it can be really hard to be productive. Unless, like, your job is literally like being like an AI news guy where you're just making YouTube videos on the latest and greatest of everything. Like, if that's your profession, that's a little different. But if you're actually doing work and you're trying to get leverage from AI itself, you don't want to spend all your time going around rabbit holes and consuming so much information. For things that are not specifically towards what you're actually trying to accomplish.
• 44:27
  Speaker A: Yeah.
• 44:27
  Speaker B: And I think the benefit that you get there is that you're not, you're out of that day to day, like, oh, my God, this is trending. And like everyone's doing this and then it's gone two days later. Right. Like, I think social media, especially in the AI space, changes so quickly. Josh Pigford from Bare Metrics just posted and he was like, you remember a couple of weeks ago or a month ago when everyone was really into building markdown editors for 10 days? And it's like it was. And I remember that everyone was like, I have this new markdown editor. Everyone was building markdown editors. And then nothing.
• 45:00
  Speaker A: Right.
• 45:00
  Speaker B: And it's like if I had gotten on that, that hype train, it's like, oh, I'm doing this. And it's like, this is going to change everything. It's just, yeah, it's just a waste of time on that. Just running in that hamster wheel.
• 45:11
  Speaker C: It feels like, you know, for sure. And it's like kind of, you know, play around with some tools and then figure out the ecosystem that you like. Because at this point, there is pretty much three major players between OpenAI, Google and Google with Gemini, and then Anthropology dropping with Cloud. So, like, yes, tomorrow ChatGPT might launch something that's better than all of them, but all of them are iterating so fast that if you jump from tool to tool in a month or so, the tool you just left probably just released something else that's better than the thing you just jumped to. So it's like you also, like, have to keep that in context. So it's like you don't really want to jump. It's not that you shouldn't test new stuff, but it's really more about just understanding, like, what each things are good at. And you're leveraging it for what it's actually good at as opposed to like, oh, well, you know, I heard Chat GPT launched this cool thing. Let me go try, let me try their image editor. Because people are saying it's better than nano banana 2. And it's like, yeah, you can test it out, but if Nano 2 was working perfectly fine for you and you had no issues, and if you have an elaborate setup around that, it probably doesn't make sense to leave it to go set it up for ChatGPT because they're probably going to come up with Nana Banana three in a couple months and then you're going to be like, Dang why did I move all the stuff over?
• 46:30
  Speaker B: It's really funny because between this and the vibe coding, I can build it in a day. I feel very seen right now because this is the exact process that a bunch of us devs went through for years. It's like there's a new database system coming out. Okay, we're all going to move over to MongoDB. Oh, there's a new, there's a new rendering engine for JavaScript now. Oh, we're going to move on to Grunt. Oh, no one's using Grunt. It's now onto Pikeman. Oh, no one's using Pikeman. It's onto webflow. Oh, no, it's not onto webpack, it's onto Vite. And it's this constant thing. And now everyone else in the world is feeling like we. And it is wonderful that you all know my pain.
• 47:07
  Speaker C: Now listen, I remember I tried to learn development, this is years back, probably 10, 15 years ago, and I started learning Rails, but this is when Rails was super hot. And it's like if you try to learn Rails now, it's like everyone's on Node and all these other languages that's like more popular. It's like, it's. That's just the nature of technology, especially something this popular, like, and when there's this much money behind it, governments involved, like, this technology is going to move fast. So settle in, just learn, know what, what each tool is good at and just figure out the ecosystem that you're comfortable with and just settle in. As long as it's one of those big three companies, you're probably going to be okay.
• 47:49
  Speaker B: And eventually it'll be the local models you have on your own machine, which will be awesome. So you don't have to pay for tokens anymore. Brandon, thank you so much for joining us. This has been absolutely amazing, as always. I love talking to you. Where can people find you on the Internet?
• 48:01
  Speaker C: Yeah, so. Well, most of the time I do a lot of LinkedIn posting on my LinkedIn at Brandon. Well, slash Brandon Shelton one. And then also on Facebook I actually do some business posting and different things like that. And then obviously, you know, Buddyprompt.com awesome.
• 48:18
  Speaker B: And we'll link to all those things in the show notes. Brandon, thank you so much for joining us and have a great day.
• 48:23
  Speaker C: Thank you, man. Pleasure.
• 48:24
  Speaker A: Cheers.