Loading summary
Transcript103 lines
- [00:01]
A
Melanie Miller walked into our conference with a successful marketing business. She walked out realizing her entire business model was about to become obsolete. And she caught it just in time. Her words quote, the AI teaching was mind blowing. I'm so far ahead of so many other people. Unquote. Now, here's what you need to know. A lot of marketers are already transforming their business with AI, while others are still wondering if they should pay attention. I know that's not you because you're listening to this podcast, but I want to share with you that the gap six months from now is going to be absolutely staggering between these two groups. At our upcoming conference, AI World 2026, we're going to bring together the world's leading AI marketing practitioners to show you what's actually working right now. Today we've got a sale going on where you can save $300 or you can get $800 off the all access ticket to the full social media marketing world experience. Now this is important. This special deal expires on Friday, December 5th. Get your tickets today at a BusinessWorld Do Live. Welcome to the AI Explored podcast, helping.
- [01:12]
B
You put AI to work.
- [01:14]
A
And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators and business owners who want to know how to put AI to work. Sometimes the best AI tools aren't the ones everyone else is using. People say that hands free AI automation is the key for creating great content. But is this really true? In today's episode of the AI Explored podcast, we'll explore streamlining content workflows with Claude Projects. My special guest is an AI coach and trainer who helps marketers and Entrepreneurs thrive with AI. His YouTube channel is @ blazing zebra. His course is AI power user Casey Meehan. Welcome to the show. How you doing.
- [02:06]
B
Today? I'm doing great. It's great to be here.
- [02:08]
A
Michael. Super excited to have you. So let's just start with a little bit of your journey. How the heck did you get into.
- [02:13]
B
AI? Well, I was running a marketing agency. I worked at marketing agencies for a while. I started my own a very long time ago, 15 years ago or so, and just had been passionate about robots since I was a young age. The only toys I wanted to play with were robots. I was always trying to ask my dad if he could build me a robot. And then in 2018, I started to get really interested in Machine learning, just as a, as a side project, started to go to a lot of different meetups, started to learn how to build my own neural nets and was just something that I did kind of on the side during COVID I was able to hire some really fantastic machine learning experts to help me with some, some of my marketing analytics for my agency. And then when ChatGPT rolled out, I realized I had to pivot pretty hard. Our agency specialized in nothing other than long form blog posts and all sorts of written content. So it hit us really square in our strength and it was something that I was passionate about. So within months of ChatGPT rolling out, I had my first AI coaching clients and really pivoted my agency. Now I'm just helping as many people as I can try to make that.
- [03:26]
A
Pivot. Love that story. I can relate. As someone who is a long form writer myself and have so many friends in the copywriting world, it was a for sure like kind of smack in the side of the head. And a lot of marketers right now are feeling that same smack in the side of the head because they're like, oh, this thing's actually pretty.
- [03:41]
B
Good.
- [03:41]
A
Yeah. So tell us a little bit about who you help today and kind of what you help them.
- [03:44]
B
Do. Well, I've been traditionally been helping marketers, but that now with the YouTube channel has really spread into all sorts of small teams. Primarily my marketing agency, we worked with some pretty large software companies, but now I'm coaching these small teams and we do group coaching. So like you said, My YouTube channel, Blazing Zebra, I put out all sorts of research every week, new practical tutorials that you can take and run with. And then I do have some group coaching sessions in Patreon as well, you know, for power users, to people who are building applications with it, to people who want to become AI coaches themselves. So that's the gist of what I'm up.
- [04:26]
A
To. Excellent. So what do you see as the biggest misconception? We're here to really talk about Claude projects today. So what do you see as one of the biggest misconceptions when we think about automation and we think about.
- [04:37]
B
Claude? Yeah, so Claude, you know, I'm just wrapping up a big training like you mentioned that power user training. And over the course of it I started to realize that CLAUDE was really my preferred tool. You know, but going into it, I do use all of the tools pretty much every day, but I kept coming back to, you know, hearing myself saying Claude over and over, Claude projects. And I think the stepping stone is everybody's probably listening to this, has chatted with a large language model. But if you go one shade up and start to look at these project, ChatGPT has projects, Claude has projects, Gemini has something similar, I believe. Maybe Gemini doesn't have it quite yet, perplexity does. But anyway, there's a lot that can be done in these Claude projects. These are really valuable workspaces and they can be used to, you know, gather up information about projects and information regarding clients and things like that. But what people don't really realize is they can be used for these different automation workflows. There's, you know, a lot of buzz around these no code automation tools out there where you're basically chaining different APIs together. And those can be problematic for a lot of different reasons because as you know, with these LLMs, they're wrong 20% of the time, they're right about 80% of the time. So if you chain a bun of those together, what comes out the other side is often not very good. And if anybody's messed with those, especially when it comes to a content workflow, they know they can be really challenging. But with these Claude projects, you know, the misconceptions are that CLAUDE is, you know, sort of second tier, maybe that it's not quite as good as some of the other models. I think people that are really in the know, I've heard like in Silicon Valley a lot of people use Claude, but most regular folks that I speak to. Claude is a fairly new tool, but it happens to be the best writer, I think, of all, all of the frontier large language models. And it has been right along, you know, if I'm using it for any sort of writing, it knocks it out of the park, especially coming up with taglines and even marketing strategies. So yeah, the misconception is that it's, you know, not quite as good as some of these other tools, but it can really excel, especially in the marketing area, especially when it comes to these different kind of step by step automations that we're going to, I think, talk about some examples here in a.
- [06:59]
A
Second. Love it. So Claude and content and projects, what is the benefit? What is the upside if people pay attention to what we're going to talk about, like what does it unlock that maybe you're not going to be able to unlock.
- [07:15]
B
Otherwise? For me, it has helped me do probably five times the amount of work that I could have done previous to using these type of tools. And basically it really shifts from just chatting with it and kind of starting with zero to having, you know, true little assistants that are really good at little micro mini tasks. And then these can be manually chained together where you're manually kind of copy and pasting and working with each of these little assistants to go through your content creation process. For me, it's a weekly process. I create a YouTube video every week, so I have different CLAUDE projects that I use each day of the week that sort of march me along from having no idea what I'm going to create all the way to having a full finished video with an ebook that complements it and all the metadata and everything exactly how I want it, with enough engagement that it, you know, really comes from me and the AI working together, I guess, rather than putting it into some system and just kind of hoping something good comes out the other.
- [08:22]
A
Side. Love it. Okay, let's talk about what the heck is a cloud project. First, let's define what that is at a very basic level. Just because people might not fully understand it, it might not be using projects within open AIs, ChatG, GPT and such. So let's talk about it and any advantages that you want to talk about. I have my thoughts I'll throw in there as.
- [08:42]
B
Well. So you can think of a CLAUDE project as basically just a workspace and like I said, it can be used for client work and for these processes that we'll get into. And it has a chat interface that you're probably very familiar with, just the regular chat back and forth, but then it has off to the right hand side a place where you can upload different files. And this is what's really powerful. That's your knowledge, your knowledge base, your context. And it also has a little box for instructions as well, right above those files that you put in there. So, you know, when you're starting up a new chat, you often have to give it a bunch of context about what you're talking about. You have to say, I'm working with this client, you know, here's a transcript of our last call and they want to do something like this. But so you have to upload a bunch of stuff into the chat, you know, if you're just using the chat, but if you're using a project, you can upload all of that and just it stays in that context, it stays in that knowledge base. So you don't have to get the LLM, you don't have to get clawed up to speed each time with what you're doing. It has all that information right there. And then, like I said, it has Those instructions as well, where you can tell it, we're going to be focused on marketing today. We're going to be focused on creating a blog post today based on these examples, or we're going to be focused on, you're helping me with XYZ client and here's all the information about the client. So the instructions are just kind of a high level of what that project, what that tool should be focused on. And then the knowledge files are all the complimentary material that go along with that instruction that get accessed in every single chat that you fire up inside of that cloud.
- [10:28]
A
Project. What I love about cloud projects is a couple things. Number one, I have cloud projects and custom GPTs are kind of effectively the same thing as our Google Gems. But what I really love about Claude projects is, first of all, they are. Claude overall is a much better writer. I find that the way it handles the human language is much better. I also like how Claude interacts with me. I absolutely love the way that it interfaces with me. It feels more approachable. It's not super hype y and it's not super dry. But it's also consultative and very creative, which I absolutely love. In addition, with custom GPTs, every time you use it, it's kind of like it doesn't know who you are, like it has amnesia. But with a cloud project, it's not the case. Right. Claude projects have their own project memory, is my understanding, which they've recently rolled out, which means all the other chats that are inside that project are all inter exchange into the knowledge base. So you can have hundreds of chats inside of a project. Any other thoughts popping into your brain about what you love about.
- [11:34]
B
Projects? That consultative approach is really a major advantage and I've noticed that I'm doing a lot of coding with these tools, which we're not going to get into today. But I've noticed it in that world as well, where Claude is always thinking one step ahead and ChatGPT kind of tries to do this a little more, but Claude is just so much better at thinking through what that next step should be. So, yeah, I totally agree with that. Consultative approach being a major.
- [12:04]
A
Advantage. Yeah. For example, sometimes I'll find it will go ahead and without me even prompting it, telling me, yeah, that email is an 8.5 out of 10. It's just teasing me like, all right, well how do I make it a 10? And then it'll be like, you know, where a normal person would have to say, what am I missing? Or how do I make this Better. In addition, I find that at least with my projects, I'll ask for multiple variations and sometimes even without any prompting, it will tell me which one it thinks is the best and why, based on what it knows about what I want to do. I just don't see that happening inside of the other projects. Okay, so CLAUDE projects require a $20 a month subscription. That's important. It's not available with the free plan. So let's explore how you are actually utilizing Claude projects to speed your content creation. Because I can't wait to learn more about.
- [12:52]
B
This. So the technique that really started to accelerate my workflow was this few shot training. So in the machine learning space, there's. I don't know where these terms came up with, but there's like the. You'll hear people talking about one shot. All that means is you didn't give the, you didn't give the LLM much background at all about what you were trying to do. Few shot means you're giving it a few examples. So you don't really need to worry about that as much as just remembering that if you can put a few different examples of what you're trying to create in that knowledge base, you're going to have a very powerful tool on your hands here. And the instructions for that CLAUDE project can be very, very simple where you can just say the user will input some ideas for a YouTube video, and in your knowledge base, you have a bunch of examples of YouTube titles that has worked for this particular creator or user, and you will generate a bunch of YouTube title ideas based on the files in your knowledge base. So sort of what I call this IPO process, this input process output is what my term for IPO stands for. And if you can clearly define, you know, what's going in and especially what you want out of the model, you're going to have some, some really great results. And I see a lot of prompt engine creating these very elaborate prompts that describe your voice and tone exactly, you know, with all these crazy different words. And sometimes that can work great and then sometimes, you know, just depending on the day, it might go completely sideways and not work very well. But I realized if you use the old writing adage, which is you show don't tell, right? That's what all writers, you know, a lot of writers talk about that in their writing. You kind of show the action rather than telling people about it. And if you do the same for these large language models by giving them just clear examples of what the output should be, they can really Nail it and become very, very useful very quickly. So that is, is my process and I've broken everything down. Rather than trying to do too much with just one prompt or, you know, with one session, breaking it down into these smaller chunks where I start, you know, my week with. I have some ideas of what I might want to do and I, I put those into my title bot here, my title Generating Claude project which has access to titles of my videos that have performed well. It takes my notes, my chicken scratching thoughts about what I want to create and it formats them in ways that align with my previous successful videos. So that's a lot different than just putting that right into the, to the raw chat because oftentimes it's, it's then relying on its training data to come up with these titles. And I'm sure you've had this experience, Michael, where they're super hypey or they're just not at all right for your audience. Because in its training data it has information about, you know, what YouTube titles were working five years ago, you know, so it's kind of out of date. But if you give it exactly what you want in that output examples of those, you're going to have a lot more success in generating those ideas. Does that, Did I explain that clearly.
- [16:11]
A
Or. Yes. And we're going to break down this IPO process in detail here in just a couple of minutes. But I want to tell you what I'm hearing you say, what I'm hearing you say is this concept of few shots means you are giving it a very simple instruction which is to effectively go out and look at the examples and to model the examples. And the reason why this works is because this concept you introduced called show don't tell and the idea is, which I think you would probably agree with, is that AI is really smart at looking at examples and probably behind the scenes turning it into math because you have more machine learning knowledge than I do. But just, just from your machine learning experience, can you kind of explain why this actually works? Just so people can wrap their head around.
- [17:01]
B
It?
- [17:02]
A
Yeah. So I think you don't have to get super technical. You can just kind of conceptually, yeah, no.
- [17:06]
B
Problem. But Sam Altman said something many years ago now, I guess it was when these things first came out and everybody was kind of comparing them to Google, you know, like a search engine. And he says these aren't search engines, but they're reasoning engines. So what's tricky about them is that they do have a lot of facts in their training data. But they may be off. You know, that training data might be old, it might not be up to date. And there's, there's been a lot of advancements since then. The beauty of these things is that they can take text and really kind of mangle it and reason over it and kind of think through it. And so if you give these models all of what they need to be successful, then it's much easier for them to say, oh, this, this is where this person's coming from. You're giving them, you know, enough context to be successful and you're not relying so much on that training data, which is, the training data is confusing because when you're just using Chat GPT out of the box or any of these models out of the box, it's basically they scraped the whole entire Internet, right? And that created a whole bunch of information that they know, but it's also kind of how they think as well. So if you can kind of give them, here's all the information I want you to use to solve this problem, and you just use what they're really good at, which is their reasoning ability. You're going to have much better outputs than when you're relying just on that training data there. And that's where the few shot example really shines is because you're giving it enough context and you're not kind of leaving much up for debate about what, what you're looking for in the.
- [18:47]
A
End. Okay, we're going to break down your process in just a minute, but I have a slight technical question for you, which some people might be curious about. When you are inside of a chat, inside of a cloud project, you have the option to choose a different model, right? You've got like in Claude, you have Haiku, Opus and Sonnet, and as of this recording, Sonnet is their best model. Do you think it matters which model you pick depending on the project, or do you just leave it on the default whenever you go in there? Because, you know, some of these models are more deeper think models and some of these models are.
- [19:21]
B
Faster. Yeah, this is interesting because it feel like it's, it's really user to user. Some of the people that I know that know these models best will have a different opinion on this. For me, that opus, which is not the highest level, is still the best writer. For me, when I go at the highest level for anything writing, it tends to overthink. I have the same experience with the Chat GPT thinking, which is one of the highest tiers technically, not the highest.
- [19:47]
A
Tier. OPUS is the higher tier, believe it or not. So as of today, it's true, sonet 4.5 is a higher number, but the opus 1 is the deep thinking one. So that's the one you find gets better output for you.
- [19:58]
B
Right. Just not to correct me if.
- [20:01]
A
I'm wrong, I love.
- [20:01]
B
It. I know exactly why you're saying that, because when they all rolled out, Opus was. Was the highest one. But that Sonnet 4.5 is actually one of the highest performing frontier models across the board. Okay, all right, so it is. It is very confusing. And that's the one that I use for coding, and that's the default.
- [20:19]
A
Model. When you go in there, when.
- [20:20]
B
I'm doing anything with code, that sonnet 4.5 just crushes it. But you're right, you know, it is very confusing because Opus, when it first rolled out, was the highest one. And there may be more parameters in that one, but as far as performance benchmarks go, that sonnet 4.5 is the beast of the Claude world these.
- [20:42]
A
Days. But for writing, you're saying you get better results when you choose Opus? That's good to.
- [20:46]
B
Know.
- [20:47]
A
Correct. Okay, so that's really helpful. And obviously that could change. So you probably recommend experimenting with these.
- [20:53]
B
Right? Right. Yeah. When a new model comes out, I always give it a try. But sometimes, you know, if you got something working, this is what I learned the hard way. If you got something working, just there's no need to optimize it. You know, it's like I come from the SEO world, and if you're in the number one position, if you got a keyword in the number one position, don't do anything, don't do more optimization, because you can only go down. I found the same thing with these models. It's like if I got something working, even if it's the older model. Yeah, I'll dabble, maybe. See? But it's like if it's working for my workflow, it's like experiment on something that's not working, I guess, rather than something that does.
- [21:31]
A
Work. Hey, before we continue, I wanted to tell you about an upcoming live training that could be a game changer if you've been experimenting with custom GPTs. What if custom GPTs could actually scale your business? I invited Wendy Breakstone, known as the Custom GPT Gal, to lead a special workshop for AI Business Society members. In this training, you'll discover the three types of profit ready GPTs every scalable offer needs so you can stop guessing and start building with purpose. How to identify the exact friction Points in your business where custom GPTs can deliver results faster while freeing your time and which custom GPTs to build first because they're not all equal. And some will move the needle a lot faster than others. For sure. This training is happening live for AI Business Society members on December 11th 7th, and it will be recorded. If you're ready to scale your business with custom GPTs, head over to socialmediaexaminer.com AI to join us. Trust me, this might be the strategic framework you've been missing. Head over to socialmediaexaminer.com AI to sign up. Now, here's something that might surprise you. Having a smaller team is about to become your biggest competitive advantage. I know it sounds totally backwards, but AI just flipped the equation. Having fewer resources means you're more agile, more decisive, less burdened by the coordination challenges faced by bigger teams. The marketers winning with AI right now are the solo operators and the small teams who have figured out AI. And the ones that are struggling are the ones that are overwhelmed by too many tool choices and stuck in experimental mode rather than actually implementing workflows that are proven to work. That's why we've built AI Business World. Our AI Business World conference is where you receive two days of pure implementation training and where you discover AI workflows that actually work. Right now, you can save $300 on your tickets to AI Business World, or you can get a $800 discount on the all access ticket to the full Social Media Marketing World experience. This sale ends Friday, December 5th. Get your tickets right now at AIbusinessworld live. Your size isn't your limitation, it's your advantage. You just need the right systems to leverage it. Visit AIbusinessWorld live to get your tickets. Okay, so we're gonna get into building a cloud project right now. So you mentioned your ipo. Process input, process output. Explain that a little bit more and let's drill into, like, how that is important and why that's important. Specific for developing out the cloud.
- [24:17]
B
Project. Awesome. Yeah, so I just come back to this time and time again, but it's thinking about what job are you trying to get the AI to complete and what are you going to give to the AI to start that project and what are you looking for from the AI as the result? And I find that people don't often have these very clearly defined. They say, well, I want to chat with it, I want efficiencies, I want this and that. Well, great, but you got to get a little more granular than that. And this can Be really helpful if you're doing client work for people as well. But saying, okay, well, I have an input that maybe is a call transcript that I had with a client and from that I want to get, you know, a strategy document out or even just a series of next steps or something like that. So you're clearly defining what's coming out of the project. Once you have those two things clearly defined, it becomes a lot easier for the LLM to use its reasoning power. Like we talked about, these are reasoning engines. That's what it's really good for, saying, I'm going to be giving you this, I want this. In the end, then the prompting becomes very, very simple where it's just like all the prompt really is, is, you know, telling it what you're going to give it and to create something like what it has in its knowledge base, which is the files that you upload into the CLAUDE project. So I can get into more detail, but I'll pause.
- [25:45]
A
The. Right there. I love this. Okay, thank you for pausing. So the input is what is the data that you're going to give it? And in some cases it might just be a prompt, you know what I mean? You might not have any. Or is the input what goes into the cloud project? That's really what we're talking about. Is that correct when you say.
- [26:00]
B
Input? Yeah, input is just what are you going to give? So like you want a blog post out of it, what are you going to give it to, to create that blog post? You're not going to just say create me a blog post. You're going to say, I want to create a blog post about X, Y, Z. You know, or ideally it's maybe you have a convers or a, a call transcript or just some sort of idea of what you want that blog post to.
- [26:23]
A
Be. Ah, okay, got it. So you have a, like a basic outline, for example, could be an.
- [26:27]
B
Image. Right, exactly. And that's. And maybe when we get into sort of my exact process, this will make more sense of how these sort of chain.
- [26:34]
A
Together. Yeah, perfect. Okay, good. So let's start with the system instruction for this. We're talking about this few shot concept, right? So what is like the system instruction? Just state it again. And can we use this system instruction for any project or is it mostly for the kinds of projects we're talking about.
- [26:50]
B
Today? So pretty much every CLAUDE project, the idea is that you would give it some sort of system instruction. I've found that the less elaborate that is and the more examples you give it in the knowledge base, the better it's going to perform. So there's a lot of people that do these very fancy JSON system instructions and they can work, work, but that is really just where in this few shot example that we're talking about where you're just saying the user will give you X, your job is to create Y. You have three examples of Y in your knowledge.
- [27:27]
A
Base. That's simple. Okay, cool. That is that simple. All right, good. So let's talk about the knowledge base then, because obviously this is the critical piece of the whole thing.
- [27:36]
B
Right? One quick thing is that it's simple by design. It's not going to get better the more words that you put in there. I think that that is a big misunderstanding with these models and the way that people use them is they think that the more words that they give it in in that instruction, the better it's going to be. Well, these things are going to look at every single word. They're not like a human where the more you talk about something, the human's just going to pick up on the gist of it. It's going to look at every single one of those words and it might, you know, sometimes get it right, sometimes one of those words it might take too seriously and it may throw it off. So it's important when you're doing your prompts or your system instructions that you kind of look at every word and make sure that it needs to be in there. You know, Michael, you're a writer. You know the. No extra words or whatever, no.
- [28:25]
A
Useless words, last words is hard, man. So getting it down. That's why these examples are so important, right? Because if the, if the system prompt or system instruction is that simple, then I think giving it good examples is absolutely critical. So let's talk about.
- [28:38]
B
That. So depending on what you're creating with it, you want to give it a handful of.
- [28:42]
A
Examples. You can go with your example about the thumbnails if you want. Let's go ahead and go with.
- [28:46]
B
That. So my process each week starts with kind of thinking through the title and thumbnail. So this is where the, you know, the input has the least format. It's just like, hey, I have some ideas on doing a video about something. And in the system instructions we will say user will put in a couple ideas about what, what they're thinking of creating a video on. And you will generate a bunch of title and thumbnail ideas based on your knowledge in your knowledge base there. And in that project, I have just one file that has 10 different of my best performing video titles and then a little description of the thumbnail that went with each of those. So it's matching it exactly to not just my style, but proven success stories from. From my style. So there I'm just able to generate some ideas and then I have. It'll. It'll come back with, you know, 10 and I'll have a conversation with it, which you can't do with these. No code tools, you know, because I'll say, okay, I like, you know, numbers 3, 7, and 8. Create 10 more that are in line with that. And pretty quickly I'm able to find a couple that, that feel right and I'm able to move then that onto my next cloud project, which is my next step in my.
- [30:01]
A
Process. This document that you're talking about that is the knowledge base for your cloud project. It simply has 10 titles for your top performing videos on your YouTube channel. And then descriptions of the thumbnail. How did you describe the thumbnail? Did you use AI to assist you in that process? Like, how descriptive do you need to get on the thumbnails? Now, you and I were talking about this. You said your thumbnails are super basic, but not everybody has super basic thumbnails. So, like, any tips on this kind of.
- [30:30]
B
Stuff? Stuff, yeah. So you can use now any of the models to reverse engineer what is in the thumbnail. So you could upload your 10 best thumbnails or just. I would do it one at a time. So you're getting the, you know, you're getting a good amount of compute for each one. But you could, in cloud or Gemini, any of the tools you can upload your best performing thumbnail and say, hey, describe this thumbnail. I would say, you know, in, in a concise manner, basically, and, and then just kind of work with it so that you can get, you know, a description of that thumbnail. For me, you know, I just. My thumbnails are very simple. It's really. And I think a lot of people too, it's the, the words that appear on the thumbnail, they need to be just like four or five words. Most people are. If you're in a B2B marketing space, you know, those are going to be really important. You know, if you're Mr. Beast or you're doing something like that, it's going to be different. But if you're doing any B2B stuff, it's really coming up with those four or five words that are, that are going to show up in that thumbnail that are, that are going to be the most important. And that's what I have mine trained on. What are those four or five words that's really like the headline and then the, the title is a little bit more like a subhead that pulls people.
- [31:46]
A
In. Well, and I would imagine for folks that aren't super active on YouTube, you can go to any YouTube channel and you can sort by popular and then you can just go ahead and probably take a screenshot of the top five and then the next five and then you could probably drag it into Chat GPT and, or Claude and say extract the title titles and describe the thumbnail in such a way that an AI model can make a descriptive recommendation of how, how the thumbnail looks or something along those.
- [32:14]
B
Lines. Right, exactly. And with those, again, I've done a lot of that with those. You want, you know, less is more with those words because it may return like a very long.
- [32:23]
A
Paragraph. Something as simple as Casey on the right side of the image, Casey on the left side of the image. Pointing, not pointing, text on screen, background fuzzed out. That kind of of.
- [32:32]
B
Thing. That kind of thing. For my style of videos, you know, it's a very big topic depending on what type of video you're doing. But yeah, mine I'm really just looking for what those big words should be. That I've done a million tests and that's what really moves the needle is if I can get that the words that appear on the thumbnail. Right. The rest is pretty.
- [32:52]
A
Easy. Okay, perfect. So this is just one of many different projects that you use in your process. And we've spent a lot of time going through this first one just to help everybody understand. Like this is the baseline, very simple system instruction and in this case also a very simple text or document that is attached and I would imagine you update it every once in a while if you have some.
- [33:16]
B
Outperformers. Yeah, Context engineering, that's you know, quarterly. Probably want to review it.
- [33:21]
A
Quarterly. Okay, so let's talk about some of the other projects that you're using in Claude for in this few shot.
- [33:28]
B
Concept. So once I have a title and thumbnail that I like, I can take that to my next cloud project, which is the hook creator. So inside of that one this time I have three different documents and each one of those has the hook, which is the script of the first 30 seconds to maybe a minute of, of the videos. So all that needs to line up directly with the title and.
- [33:54]
A
Thumbnail. The top performing.
- [33:56]
B
Videos. Exactly. Sorry, I didn't say that. But this is of my top performing video, ideally that are. That are somewhat different. You know, you want to get like three that are. That are fairly.
- [34:05]
A
Different. Do you include the title in there so they can connect the dots, or do you not worry about.
- [34:10]
B
That? I don't think that I do, but it would probably be a good idea. It would probably be a really good.
- [34:16]
A
Idea. Okay, so you've got the hook. You said there are three different docs. Keep.
- [34:20]
B
Going. So this then will take my. I'll copy and paste whatever the title and thumbnail idea from my first Claude project that we just went over into this hook creator, and it will then generate. Generate a script for that first 30 seconds to a minute. I'll work with it back and forth. So I'm happy with it. Then I have that. I can take all of that now, copy it into a third cloud project which will create an outline of the video. And this one, I typically edit more than the others because I have some.
- [34:51]
A
Ideas. Sorry, before you go on to the outline, going back to the hook thing, you said there are three different documents in there in your knowledge base. And I think I only caught one, which.
- [34:58]
B
Was. They're just three examples, samples of different.
- [35:00]
A
Hooks. Oh, is that what it is? Okay. And each one's a separate document. Is that effectively how you're doing.
- [35:04]
B
It? Yeah, I don't think it really makes much difference if they're all in one versus if they're in three. As the pieces get longer, which they can get way longer, which we'll get to kind of at the end of my process, then they do need to break apart. But at this point, I mean, that all would fit on like one to two pages. So it could be one document. Doesn't matter that.
- [35:23]
A
Much. Do you find that you try to give it a variety, like, let's say you've had. You know, how sometimes we tend to kind of follow a routine. And do you give it, like, if they're all kind of the same, do you try to pick a couple that are outliers to give it a little more creative freedom? What's your thoughts on.
- [35:41]
B
That? That's ideal, if you can, is, you know, I try to figure out, you know, there's the one, right, the one that you know is the banger the best. And then you figure out, okay, what's the best one that's like completely opposite of this almost. It's really good, but as different as possible. And those are sort of your north, south. And then it's like, can you find one that's like Way different than those two. That would be ideal. But it's not always possible. If you could have three, like pretty different ones. But this can be helpful if you have a client, you know, that you're building these for, because this can be a really lucrative thing to build for people. And they give you like a thousand examples or what, a whole bunch. Then you can use LLM to sort through it and figure out, you know, what are sort of the three. Three different, kind of different ones that might make the best. Rather than giving it a thousand, you want maybe three to ten different.
- [36:28]
A
Ones. Okay, perfect. So now you have another project set up with outlines. So talk to me about what makes this project unique as far as the way you set that.
- [36:37]
B
Up. Well, it's basically the same thing. I just have examples of outlines, examples of that output, and I'm going to be giving it everything I have so far. So the hook and title, and then I have the, the title thumbnail hook will be the input to this. And it has in its knowledge base, you know, several different outlines that might have different, you know, styles and different flows that it can then generate, you know, the outline of the video. And with that, I'm pretty much ready to.
- [37:07]
A
Record. Love it. Okay, cool. So at this point you do. Have you talked about this critique one? Is this something that happens before? After you record? After. Oh, okay, so, okay, record. Keep.
- [37:20]
B
Going. Yes. So then I'll record. And I love Descript. Probably some folks listening will be familiar with Descript and that will automatically transcribe everything and that's how you edit in there. So then the next one is after I record, I'm able to take the transcript of the video and pull that into another Claude project that has transcripts of some of my most successful videos. And it'll say, hey, what from this transcript script? How does this one not match up with these others? What am I missing here? Am I missing a call to action? Am I going on and on about something in one section or, you know, how can it be improved? And it will critique my performance in a way that I can go back in and edit inside of descript or maybe add some voiceover, some commentary to get that up to snuff there. So I've turned it into that video critique.
- [38:11]
A
Bot. I love that. I mean, this is kind of a big unlock for a lot of people that are listening right now. The idea that that most people stop at the outline and then they just go create or they stop at the output that they've co collaborated with the tool like Claude. And then they move on. Talk to me a little bit about why you decided to come up with a critic, because I feel like that's kind of cool. I mean, like, did you have that from the beginning or what's your thoughts on.
- [38:36]
B
That? I've been doing it for a while. Yeah. And it just, I, I guess probably like a year ago I came up with it. I think it might have come. Come up from when I was, I had to give a big talk at a conference. Maybe this might have been when I started doing it. And you know, I was doing these rehearsals and of course I was recording the rehearsal so I could view it back and I was like, why don't I just toss this in to the LLM and see if it. And it can critique you without even having those examples. You know, that's an interesting route too. But having those examples where it knows, you know, here's what really worked. Yeah, that can be.
- [39:10]
A
Helpful. Okay, cool. So what's your experience with it being a critique? Is it generally pretty accurate? Do you take everything it says as gospel or do you kind of have to filter a little bit in your human.
- [39:20]
B
Brain? Yeah, definitely have to filter it in my human brain. It's amazing how off these things can be sometimes. And again, that's where the challenge. I think that's why the chat user interface like blew up. Because these things were around, you know, there was GPT 2 GPT, 3 was before there was chat GPT and nobody cared. But as soon as it turned into this chat, chat interface where if it was off, you could be like, no, that's not what I'm talking about. Do it this way. That's when these things became really valuable. And so you definitely have to, have to work with it. And that's where a lot of people struggle when they're trying to create these fancy automations. And like we said at the top of the call, you know, if it goes sideways one out of five times, like, that's just not that it's not robust enough to really put into these different no code tools and expect them to, to be, you know.
- [40:11]
A
Worthwhile. One of the cool features of Claude in general is this thing called artifacts, which kind of makes it look like a little Google Doc on the screen and you can select text and improve it or modify it. Do you find yourself using artifacts a lot in the output or do you find yourself just, just taking the text as it appears on the.
- [40:28]
B
Screen? I actually, I, I do think artifacts are, are great, but for a while I don't know if it's doing it anymore, but there was a moment where it was trying to create an artifact fact, you know, I'd say hello and be like, here's an article, here's a, here's a JSON file app as a response. And I was like, whoa, no, let's chill out there. These things are always trying to go to that next level, which is, can be, can be problematic. So I actually have those turned off and I might turn them on again to see if it's gotten.
- [40:57]
A
Better. But it makes it so much easier to read, especially if you're dealing with output that's a lot of.
- [41:01]
B
Text. Definitely does. Like, if you're definitely work, just working on, on a piece of text and you can. There's a lot of advantages too, where you can highlight a certain area and say, just recreat, create this passage and that type of thing. But I found that it was kind of launching into that way too often. What about you, Michael? Do you, do you use those a.
- [41:18]
A
Lot? I use it all the time, but I specifically say use artifacts for creative output. So that way it's not just everything is an artifact. Does that make sense to like, when it's actually creating.
- [41:28]
B
Something? Yeah.
- [41:28]
A
Definitely. Okay, so once you've critiqued the video and you're done with the video, when we were prepping for this, you came up with a really creative way to have like an offer to people that are watching the videos. And I would love you to talk a little bit about this thing that, that we were, you know, that you, you know, you know what it is. So go ahead and explain what you.
- [41:48]
B
Do. Yes. So with each video, I create a complimentary ebook which is over 20 pages long, which really extends everything in the video. So my videos are around, I don't know, 10 to 15 minutes long. And then these ebooks have, you know, all the prompts that I go over in each. A lot more resources, a lot of things so people don't have to go back and forth and take screenshots and all of this. And you know, if it wasn't for cloud projects, I would not be able to do this. I mean, that document, doing that manually before AI would have taken multiple people, people a whole week to.
- [42:27]
A
Work. And it's a huge thing because it's a great way for you to grow an email list if you want to. I mean, does it work for.
- [42:33]
B
You?
- [42:33]
A
Yeah.
- [42:33]
B
No. So I charge for those. That's a low tier charge in my Patreon. That's six, seven bucks a month. For access to 155 of them now 158. So yeah, that, that is huge. And so that's one of my last steps. There is I, from that final transcript, I say, hey, let's create an ebook from this. And it has example. This is the one where it's critical since these are such big files, I have separate files in my knowledge base. And this is where this few shot comes in super handy because. Because it would have no idea how to create this ebook the way I want it without these examples. And it will go and crunch through. It doesn't create the whole 20 pages, but it'll probably come up with the first 10 to 12.
- [43:18]
A
Pages. That would be perfect for an artifact, by the.
- [43:21]
B
Way. Right. Because I do end up heavily editing that and I do spend hours on those. So it's. I hope people don't think that I just crank them out and don't think about it. But, but those, I do spend a lot of time editing and improving those. So they are very valuable. And it's really. That becomes my library. I'm always referring back to, to all of those. And how did I do this, you know, three weeks ago or.
- [43:43]
A
Whatever? Well, what I love about what you've done here is you've got really an end to end process that you've created here with a bunch of different cloud projects that are super specialists. And what's really cool is these ebooks could be foundation material for something that you might want to do down the road, like a course or like an actual physical book. And just, and by the way folks, whether you're creating the exact stuff that, that Casey has talked about, like I would imagine you've got projects that you're using for other kinds of things as well. This is just an example that everybody can wrap their head around.
- [44:16]
B
Right? Yes. Yeah. I think this is one of the highest leverage use cases for AI because it's not. Anybody can do this. This isn't rocket science, but it's super reliable. That's the thing is, you know, you're still in that chat format. So when it's a little bit off, you can say, yeah, just redo this. You know, I forgot to tell you that my audience doesn't care about that type of thing or something like that. So it's very collaborative with the AI and it's really grounded in things that have been successful for you all along. So I do have other projects for specific projects, you know, for specific client work or specific, you know, software that I'm building Something like that. Those are a little bit different where they're kind of collecting everything. But this is what. I don't know what allows me to get through my week. So I have like a notion page that has my whole week. And it's just these different links to these different Claude projects. And, you know, no matter what mindset I'm in, if I'm really tired or whatever, it's just like, well, here's the next.
- [45:15]
A
Thing. Well, that's what's great, is Claude never gets.
- [45:17]
B
Tired.
- [45:17]
A
Exactly. So at Blazing Zebra is where you can find your YouTube channel. Casey, see if people are interested in connecting with you outside of YouTube or possibly working with you. Do you just want them to go to your YouTube channel? Is there anywhere else you want them to.
- [45:32]
B
Go? The YouTube channel is a great place to start. You know, pretty much everything I do is up there. And then inside of my Patreon, I have, you know, the different group coaching groups there broken down into whatever interest you're in, whether it's this type of thing, which is my power user group. Then there's a coding group, there's an AI coaching coaches group as well. You can get to all of that stuff from Blazing Zebra. AI is my website there, which has a newsletter. But that's awesome. That's basically.
- [45:59]
A
It. Casey, thank you so much for sharing your process with us.
- [46:02]
B
Today. My pleasure, Michael. Thank you so much for having me. I really appreciate it.
- [46:05]
A
Man. Hey, if you missed anything, we took all the notes for you over@social mediaexaminer.com A82. Be sure to follow this show on your favorite podcast. And if you've been a listener for a while, we'd love a review. And also let your friends know about this show. I'm active on Facebook, LinkedIn and X. And do check out our other shows, the Social Media Marketing Podcast and the Social Media Marketing talk show. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Expression podcast is a production of Social Media Examiner. Get your tickets to AI business world right now by visiting AIbusinessworld live.