Prompting at Scale: How to Deploy AI Across Your Company

Happy 2026. Hey, how's this for a New Year's resolution? How about you Master AI marketing? The AI Business Society gives you live training with experts who are getting real results. No hype, no fluff, just proven workflows that work. And you've got access to an incredible library of training from many of the incredible guests that you've heard on this show. So secure your AI transformation by visiting socialmediaexaminer.com AI welcome to the AI Explored podcast, helping you put AI to work. 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. Hi, I'm your host, Michael Stelzner, and this is the podcast for marketers and business owners and creators who want to know how to use AI. Today, I'm going to be joined by John Munsell, and we're going to explore how to create a framework for prompting that you could scale across different employees, different divisions of your company, or even for yourself. If you've just been kind of like randomly using ChatGPT and Claude and just you don't have a really good structure and framework. I think you're going to find today's interview absolutely fascinating. And if you're brand new to the show, follow us on whatever app you're listening to because we've got some great content coming your way. Let's transition over to this week's interview with John Munsell, helping you simplify your AI journey. Here is this week's expert guide. Today, I'm very excited to be joined by John Munsell. If you don't know who John is, he's the founder of Bazooka, an AI training consultancy and strategy firm. He's developed the AI Strategy Canvas, a guide to help businesses deploy AI across their companies. His forthcoming book is Ingrained Strategy through the Blueprint to Scale in AI First Culture. John, welcome to the show. How you doing today?

Hey, I'm great, Mike. Thanks for having me on. I've been a big fan of yours for, well, I hate to say decades, because then we'll both look old, but for a while.

I appreciate that. I appreciate that. So today, John and I are going to explore how to scale AI prompting across the company. And I'm really excited to share what you've come up with, but really, before we do that, share a little bit about how in the world you got into AI. I mean, start wherever you want to start.

So I started off in banking and spent 15, 16 years in financial services. But during that time I was using software and I was automating things and I got to be a big fan of marketing back when all you could do was do direct mail. Then in 97 when the web came out, I thought, wow, this is where the future is. So I ejected. At the time I was working with Northwestern Mutual Life and I ejected to start a software company. We started to build web applications and websites and gradually we grew into a digital marketing agency. And we had been using AI, I don't know, for the last three some odd years. And about two and a half years ago I thought, okay, well this is where the new future is. So I sold off the agency side and wanted to kind of contract because my wife and I are this close to being empty nesters. And so we thought, man, maybe we could, we could travel. And as long as we had a light, lean company, I saw like, man, I love AI, I love teaching, I love training. And we developed some very robust tools for scaling AI. And so that's what I did. And off we go, you know, well.

I'm a little curious about. So you kind of decided to sell the agency, which is probably in hindsight, turned out to be a pretty wise move to get into this new world. Tell us a little bit about the journey in the beginning. Like when was that? When did you sell off the agency? What were you starting to do in the AI space? And then maybe just share a little bit of that journey up to what.

You'Re doing now, interestingly enough. So we started our company, like I said, In 97, and then in 99 we developed, developed some IP that became a content management system. And we transferred it into another company called Bazooka, which we grew. And I had a guy working for us in sales, he had a little agency and then he, he wasn't doing too great with it. So he came on board with us and we helped him and he helped us. We, we had a great time. And then he ejected to join somebody else and kind of start his own company again. Gradually his company got acquired. And then I want to say it was literally about like two and a half years ago, he and I were having lunch because we were always, while we were competitors, we were always friends and we decided to go have lunch. And I like, hey, you know, how are you going to grow this year? He said, well I think we're going to go through acquisition. And he's like, how are you going to grow? And I said, I'm not sure I want to Stay in this space, you know. I said, I see a future in AI and it just makes my propeller spin and it's a whole lot of fun. I said, I think I'm. I'm going to go in that direction. He goes, well, would you want to sell the company? And I was like, I don't know that I'd want to sell the company, but would you be interested in buying the portfolio of clients? He's like, yeah. I was like, okay, well, let me know when you're ready and we'll talk. And so, I don't know, six months later or so, he called me up. He goes, okay, we're ready to start acquiring. Are you still interested? I'm like, yeah. So we had lunch and he said, how much do you want? I said, I'm thinking this. He goes, okay, let's do the deal. So off we go.

Wow. But then, like the AI thing, so what did you do after that? Like, I'm curious what came next?

Yeah, okay, really good question. Because at the time I thought, oh, I'm just going to go become a fractional CMO for people. And I landed a couple of fractional CMO gigs and realized that isn't really what I wanted to do because it was basically trading the job I just got rid of for the same doggone job. So what I did is I started this group called the CxO AI Roundtable. And every Friday I would teach people something about AI. And sometimes it was investigating a new set of tools and evaluating the different tools that were available to solve a specific problem and telling people, okay, here's the one I landed on that I thought was the best. The idea was that I would try to find what problem the market wanted solved, right? So that was the original intent of these Roundtable discussions. To just teach people what I knew, have a good time, and then try to figure it out. And I don't know, about eight or 10 episodes in this other buddy of mine, we kind of got together. He goes, okay, now I think we know what problem we're solving. I'm like, yeah, there are two things. One, there's so much coming at us so fast that people just need somebody out there to kind of distill it and help them understand it. And then the other was, how do we scale the use of AI in our organization? Because right now there's a million people teaching certain aspects of it and they're all taught a different way. And basically everybody's self taught. And so the minute somebody goes, oh, I can ask, chatgpt how to create a prompt. Now all of a sudden I'm a prompt genius. I'll go sell these prompts for $5 and I'll make a ton of. It's just like, oh, that's not how you do it. And we had developed a process because we're trying to market ourselves, right? And but we would do a whole lot of other things because I still had, I retained like our social media clients. I sold, I sold everything else off but the social media clients. So I'm still doing work for all these clients with my team. And it was during that time that we had developed all these processes that really helped us scale. And we, I think it was in October of last year we decided to do some paid training. The roundtable got to be too much to do for free. And so I switched to paid training and I started to teach people this process that we call scalable prompt engineering, which is different from your classic prompt engineering. Scalable prompt engineering is how do you construct a prompt that allows you to duplicate it, hot swap variables in it and excel its adoption throughout the company, no matter what division you're in, whether you're in marketing, sales, finance, whatever. And so we started to teach that and it was a six week course, the first one. After the first two weeks I could tell that the process wasn't quite clicking and about, actually it was about the same time I came across Social Media examiner. We made the Inc 5000 and I went down to an Inc 5000 conference in New Orleans and I saw this guy speak and he had this thing called the, I think the book was called Business Model Generation. And he had a thing called the Business Model Canvas. And I thought, wow, that's.

So now you're going back in time, right? You just got.

I'm going way back in time. This is 2012, right?

Got it, yeah.

And, and I thought that's a brilliant deal. And I would, we would use that canvas to explain to our own team as we onboarded people what our business model was, who our customer segments was, what our revenue streams were, our expenses and all that stuff. So people could all of a sud. They now understood what goes into running a company as, as opposed to just understanding their little bitty job. Right. So when I saw that people weren't quite understanding the concepts, you know, one night, I don't know whether it's God or whatever, just kind of spoke to me. It was like the business model canvas would help take all of that stuff and explain it. And so I like, man, I got up and I. And I took all the concepts I have and I started mapping it out in that same canvas framework. And that's when I was like, okay, let me see if this works. There's a lot of complexity to it, but just suffice it to say, I presented it in week three and the light bulbs just started turning on for everybody. They're like, everything gelled when you showed me that, and I was like, okay, we're on to something now. Now I got it.

You know, and good news, everybody. We're going to talk about that today. We're going to. We're going to dig into this. So thank you so much for sharing that story. I. It kind of helps people understand. And, you know, frankly, a lot of us that have been around for a while and are just emerging into this world can bring structure, can bring experience, can bring frameworks into this space, because right now it is kind of free flow. Wild, wild west. I mean, the truth of the matter is you're looking at a flashing cursor, right? So it's about as basic as you can get. It kind of actually goes way back to the original computers that we had back in the day, you know, where it was just a flashing cursor.

Right, the DOS days.

Yeah, yeah, exactly. So, all right, well, thank you for sharing that story. I want to ask you this question for everyone that's new, that's listening today. Why is it so important that they understand prompting? Let's just kind of explore that a little bit. What do you want to say to those that are maybe struggling with it a little bit? If you get it right, what can it unlock?

What most people do is they. They'll go in and they'll just ask ChatGPT or Claude or Perplexity a question. They'll get an answer and they'll just be blown away how rapidly the thing responds. And then they'll. They'll start to play with it a little bit more and they're like, God, this is just amazing. And eventually the novelty wears off a little bit. Are you familiar with the AI Hype Curve by Gartner?

I'm familiar with the Gartner hype curve. I didn't know they had one just for AI, but yeah, yeah, they have one, yeah. Oh, really? Okay, so AI is probably in the trough of disillusionment right now or something like that.

For most people, it is headed into the trough of disillusionment. Most people get into it and they're at the peak of inflated expectations. Right, right. And so by Gartner standards, everybody's pretty far down that curve, and they're going into the bottom of it, which means just that they've started to play around with it, and they're like, okay, this thing keeps saying delve, you know, skyrocket, goofy stuff like that. It's not really giving me the results. The other thing they start to figure out is I got to go back and forth and back and forth and back and forth and back and forth to finally get it somewhere. And I still have to edit, right? You know, it always depends on what you're doing. The models get better and better and better, and they get faster and faster and faster, and they have more and more knowledge in it, and they're pretty amazing. But if you still have to go back and forth and back and forth and back and forth, even though you've. You've compressed your time from like eight hours into one or three weeks into a full day, you can compress it even more. And it's funny, once you get used to speed, it gets frustrating when it's not fast enough, right? You get spoiled. And so what, What? The reason you need to know how to do it right is because I've seen so many people's different training on how they teach you to do prompt engineering, and they're basically teaching you how to build giant paragraphs of requests, and that's the wrong way to do it. Everybody thinks that's the only way. And I've seen training after training after training, and what you have to do is you have to break that paradigm, and you have to then break it into chunks and then organize the chunks. Once you understand the principles of it, it's. You write a prompt faster, and what you get out of it, the first try is significantly better. I mean, like, infinitely better. They were like, for instance, we just ran 10 LSU professors through our training, and one of them, actually two of them, do a whole lot of grant writing, right? So they apply for grants for research. And this one guy even came on like our Mastermind call last week just to tell us, because he was so excited. He built it as part of their Capstone project. They build a custom GPT and he built one to apply for grants. And this guy, English is not his native language, so he has a rarely a very thick Spanish accent. He's like, I just had to tell you, first of all, I get what I want with very little edits right out of the gates, because now I know how to structure a prompt. Instead of me going back and forth for hours and hours and hours. I'm at 90, 95% on the first go round. And he said, I just got to share it with this. And he pops up an email from somebody that he applied for a research grant. And the email comes back and it says, I just wanted to compliment you on your application. It was so well written, so well organized, and you're so eloquent. I just wanted to call that out. Okay. They don't ever get emails like that. And it happened to be from a professor at a competing university, which was even, you know, they don't complement each other. Right.

So what I'm really hearing you say, John, is that if you can get this right, and we're going to talk about this today, you can increase the likelihood that you get what you want on the first go round instead of on the fifth go round.

Right.

And when that happens, that allows you to essentially get a lot more out of AI because then it frees up a lot more time for you to get creative with it and all sorts of benefits and. And we're going to get into a lot of that. I would love to. To ask you. All right, where do we start when it comes to your method for better prompting?

Okay, great question. So I want to back up just a second, because you not only get better results out of the first try, but think about that at scale throughout your organization. If everybody knows how to do it the same way, then everybody's communicating the same way. And anybody can come in and see your prompt and know exactly what it does and say, if I just swap out this for this, I can get it to do something completely different. Right. So that's what makes it scale really quickly throughout an organization. The way to get started, obviously you can go through our training, but on our website, what I'll do, Mike, just for you group, I'll come up with a URL. At the end, I got to think of what it might be and I'll build it out for them. But basically, if they go to that URL, I'll let them request a copy of the AI Strategy Canvas so they'll actually see the campus. They'll have a copy of it and they can work with it and they can start to make sense of it.

Well, the. The best URL would be something along the lines of your website.com SME. So I'll just bring that up when we get to the end.

Okay. Yeah.

Because SME just stands for Social Media Examiner. Okay, so where do we start? With your system, you have a method for better Prompting. Right. So how do we develop a strategy, if you will, for thinking about prompting better? Right, that's my next question, which is really like, okay, like, we understand the benefits are amazing now, like, where do we actually begin? Like, I want people to actually. Because there's a lot of people who are doing what you said earlier. They're just, they're just going round and round with the, with the prompting system. Right. And what are the elements to developing, you know, if essentially a better prompt?

So if we talk about the canvas, there are nine blocks in the canvas. The reason that most people get off track is they just start in with what we would call the request. They just start asking the AI, here's what I want you to do. All right? There are eight steps you need to take before that. They're not necessarily steps, but there are context blocks that you need to think through ahead of time. And that's what we train people to do. So the first one is what we would call the target audience. Sometimes that's the. Also known as the avatar or the Persona. In other words, who is the beneficiary of what we're trying to get out of the AI? Who are we writing to? The next one is the company. What does the AI need to know about my company? How much do they already have? What kind of detail do I need to provide it? Sometimes it's just a brief description of the company. But if you are going to be doing research grants and stuff, and those applications are constantly asking for name, address, background, testimonials, history, et cetera, all that is part of your company context and you're going to stick that elsewhere so it'll call on it and you don't have to keep retyping it. The products and services are block number three. What do we have to do to tell the AI? How much information about our products and services are we talking about? One, are we talking about them all? You know, what kind of detail do we need to provide there on the products and services?

How deep do we go? Because I think there's a lot of depth potentially behind some people's products and services.

Yeah. So everything really depends on what you're trying to get out of the AI. Right. Like, for instance, I have. Just so your audience knows, you can ask ChatGPT or Claude to give you some information, but you can also build what's called a custom GPT in chatgpt or a CLAUDE project. That's where you can upload a lot of data, which becomes the AI's knowledge base. And so it'll constantly call on that data when it's executing something. So I have what we would call context blocks for the products and services. And it might just be a description. Like we have a product called the AI Skills Builder. It might just be a brief description of what the Skills Builder is to a company. And we built that out. So it's kind of promotional and somewhat descriptive. On the other end, I have a really deep description of it with all of the things that it teaches people, all of the information about prompt engineering, all of the different skills tracks, and I have that uploaded as a knowledge doc inside of a GPT and cloud project. So that way if I'm writing something that needs a lot of detail, I'm going to use the project. If I'm just simply writing something that just needs to mention it, I'm just going to use what we would call a prompt stack and I'm going to throw that in there.

Okay, cool. I totally understand what you're saying. And just to clarify for everybody, so far we've talked about your target audience, which is your avatar, Persona, customer. You probably should have a good paragraph or a couple paragraphs on who that audience is. Your company is probably just. You could probably go to your about page and copy and paste that in there, I would imagine. But on the products and services, what I'm hearing you say is that for these context blocks, you probably should have a very basic product slash service description. But if you're going to do something that's much deeper, like develop marketing materials for that product, then you might have ancillary documents, is what I'm hearing you say. Is that correct?

Yeah.

Okay, correct.

And you can actually use AI to produce these documents, which is what we do. Like for instance, on the Persona, you can make the Persona as brief as we're talking to CEOs of small businesses in the healthcare space.

Right.

Or you can get very detailed about the psychological profile, their, their needs, their desires, their goals, their, their fears, et cetera.

Right.

We, we do both. Right. And then we also can build a really deep, detailed Persona document that would be become a document that we put inside a custom GPT. Right. So each block has different variables or different variations of how deep or shallow you want to go.

Okay, cool. Quick question. How many podcast episodes have you recorded? How many webinars have you hosted? How many client calls or team meetings have happened that are now just sitting there in your files? Here's the big unlock. Those aren't just old recordings. They're full of insights about Your audience, your brand voice and your customers actual objections. You just need to know how to extract them. That's exactly what AI educator Erica Stanley is teaching AI Business Society members on January 8th. By the end of this training, you'll have a repeatable workflow that turns every past recording into a content and data powerhouse. One that keeps paying dividends long after the mic's been turned off. This training is exclusive to AI Business Society members. But here's the good news. You can join us right now by going to socialmediaexaminer.com AI and start transforming your old content into new marketing gold. SocialMediaExaminer.com AI to join right now. So far we know the first three, right? The audience, the company and the products and services.

So the fourth one is kind of a generic block that we called context. And context means what's the additional information you need to give the AI so that it can do its job better? So sometimes like if you're writing a blog post, in a blog post, you've given it who your target audience is, you've given it a little briefing about who your company is and maybe the products or services that you might want to append at the end of the blog post. But the context would be something like what are your thoughts on the subject? How do you want to guide the AI so that it doesn't sound like it's all AI, but you actually have opinions on it. So it could be your thoughts and opinions, it could be new ideas you have, it could be testimonials, right? It could be anything like that, event, details, etc. So those are additional context things that are going to vary so much. They, they didn't deserve each like company products and services and Persona, they deserve their own block. Right? Context is just going to vary all the time.

It's just going to be like a catch all for whatever other stuff.

Correct. And, and you can label it whatever you, you wish. But it's just when you're having a discussion you need to stop and say what other context do I need to give this to get it to do its job? Right, Got it. Fifth block is the role. What are we asking the AI to do? What, what have we hired it to be? Right? Is it going to be an analyst? Is it going to be a world class expert at something? Is it going to be a copywriter? And then we get into the style brand, voice, which is box six and that is how do we want it to write? Do we want it to write in a very detailed manner? What kind of voice or tone do we want to give it? What kind of storyline do we want it to follow? What kind of narrative do we want to do? Is it first person, second person plural, you know, et cetera. So all of those are in the style of brand voice. Now you're going to have a different style for an email you send out to a friend than you would to like a research document you're sending to another firm. Right, right. So each one of those, you want to create several of these so that depending upon what you're writing, it can assume that voice. But the thing that people don't realize is that, which is why we tell people to write scalable prompts instead of just word vomit prompts, as we call them, you will experience what we call prompt conflict. So if you give it too much attention to the role you've told it to take on, that role may supersede any style prompts that you have where you're telling it to write in a friendly voice, but you've told it it's a chemical engineer.

Right.

It isn't going to be too friendly. It'll be very clinical almost, you know. Right, right.

Love that.

Devoid of personality. Not that there's anything wrong with being a chemical engineer. I'm just saying it'll take on that kind of role.

Yeah, yeah.

Block seven is resources, which is sometimes that's incorporated into the prompt itself. Like a resource could be a transcript that you're going to attach and analyze. Other times a resource could be an external link where I'm telling it to go analyze this link and come back and get something. Other times the resource might be an API call where I'm pulling in data from some other data source. Number eight is the rules. Now if I'm in a strategic discussion, the rules are, am I coming under any kind of governance internally or externally? Like in the eu, they have this EU AI act that goes into effect in March and then it's kind of stratifies through August and it's a very punitive act. Right. You better know what that is. If you're employing any kind of AI systems or you're even selling any AI into that space.

Is it only legal stuff in rules, or could it also be certain kinds of do's and don'ts?

Oh, no, no. In the prompt side, that was at the strategic side when you're trying to figure out what you want it to execute. But when you get down to the prompt side, like we have rules for, I don't want it to say certain things Like, I have a block that I call exclusions. That means I don't want it to say delve. I don't want it to say skyrocket. I don't want it to say, I hope this email finds you well. So I've given it a bunch of stuff that I'm like, I'm so sick of hearing this. And I can instantly spot AI now, because, you know, nobody has tried to write it to where it doesn't say that stuff. So those are the things that you put in the rules block. All right.

Okay.

The final block, which is the ninth block, which is the one that everybody does first, which they really should do last, that's the request. That's where you're actually telling the AI what you want it to do, how you want it to execute. Right. Having all this other context ahead of the game is what makes everything so much more effective and efficient. So when we make the request, we're telling it. I want you to look at target audience, and I want you to speak in this style, and I want you to incorporate my thoughts. Now go analyze this URL, which is a. I don't know, some sort of blog post, and then incorporate my thoughts and rewrite another one. Right. That's. That's all you got to do.

Okay, so just so we're clear, We've got these 10 context blocks.

9, 9, 9.

Target audience, company, product, services, context, role, style, slash, brand, voice, resources, rules, and request. We're putting this data traditionally into a custom GPT or CLAUDE project. Is that what I'm hearing you say? Or is this more of just a framework that you could use for any kind of thread that you're doing? Any kind of inquiry inside?

Yes to all is the answer. Right. So not every prompt is going to have all of this.

Got it.

But if you're going to be doing something over and over again, you want to start with the framework and start assembling all these pieces. Now, what we do is we have a. We use notion and we keep a library of all of our prompts, but we also keep a library of our blocks. So I might have five different Persona blocks, but I also have a big file folder in Dropbox with the big detailed documents of the Personas and the companies and the products, right? So depending upon what I'm doing, like, if I'm doing something over and over again, I'm typically going to do a GPT that does it, right? So it's just. It's got all the data, it already knows the style, the brand, voice, etc. And I'm just querying against all of that.

And just for beginners, when you say GPT, you mean cloud project or custom GPT?

Yes, Correct. Yeah, but if I'm just doing like a lot of one offs, just a random something, just. Yeah, you just got a request. I mean, like, here's the thing, a lot of people, which is why the book is about creating an AI first culture in your organization, a lot of people just think, oh, I got to go and write a blog, and that's all they're going to use it for. But you can use it to brainstorm with yourself, right? You can use it to role play. If you're in sales, you can use it for so many different things.

Things.

But you have to be thinking, how would I get this information out of AI rather than my boss? You know, how would I get it out of AI rather than my competitor? How would I learn more about my competitor using AI? All of those things are what we want people to be thinking. But they're. When you have new uses, you don't necessarily have to create all these fancy prompts. It's just that you're going to start asking questions. But it's so much easier if you think, okay, should I put all this information in a block and then ask ChatGPT or Claude to execute on the block, or should I just start writing a bunch of stuff?

And just so we're clear, when you call this a block, like if we put it in a Google Doc, would we just have the word target audience colon and then my company colon or something along those lines? Is that really what we're talking about? Or something more sophisticated?

So when you're constructing a prompt, you can have these things called delimiters. And a delimiter creates what I would call a block block, which is I want to tell it. Here's where this particular set of information starts and here's where it ends.

Got it?

And then everything in the middle is the information I want you to digest.

And how do you indicate the end of it?

Yeah, so there are several ways. We use a very basic and simple way, but the AI is typically going to know what you're doing. Some people I've seen use like a whole bunch of backslashes and then a word and then a whole bunch of backslashes and then the word again. That's a whole lot of characters. The reason I don't like that is because first of all, every character is going to consume a token. It can get expensive tokens. Are, you know, the rates are going down, but still it's, it's. It's time consuming and it's bothersome and it's hard for somebody else to actually visualize the prompt. We use a very simple thing. I know Claude likes people to use the equivalent of HTML, where you have these little brackets and stuff, where you open a tag and close the tag. You don't even have to do that. We just use all caps and a colon, which would signify the beginning of a container. And then we use a slash and the same word at the end to signify the end of the container.

Like the slash on your keyboard with the question mark, right?

Yeah, that slash. Yeah, the forward slash, not the backslash.

So, like audience colon, all caps. And then at the end it would be that slash and then the audience, all caps again. Right? That's, that's the indica. It's smart enough to know that's the beginning and the end. So, okay, we've got these nine blocks, right? This is where I want to get into the scaling side of it. Right. So once we begin, for lack of better words, codifying this procedure, maybe it starts with you. Maybe you're the CEO or founder, or maybe you're in the marketing department, whatever, and you come up with this. How do we get to scaling? That's where I'm curious.

It starts. I mean, the way we always do it is we start by teaching people how to think through the canvas first. From a strategic level, how do we define what's an AI initiative? You know, if you're in a marketing department and hadn't really started the C level down, but you're in the marketing department, you're like, man, I need to come up with a way to simplify something. You would get that canvas out and you'd start thinking it through. But if you're the CEO of a company and you want to get your team together and, you know, we, you say, okay, we need to execute. We need to figure out how to scale AI. When you have this canvas and you use it to start the meeting and you, you say, okay, we need to identify some areas in the company that we can scale by integrating AI. The canvas helps you have those discussions and identify all the moving parts that are going to have to go into the initiative, then you have to break down into a discussion with who are the team members that have to actually build out that initiative. It could be different departments. The canvas guides that discussion as well. And then when the engineers or, you know, even if it's just a copywriter or whoever it might be in your organization. You're executive assistant needs to start building out a prompt. They use the strategy canvas once again to understand how to build that prompt. So what that does is it allows everybody to speak the same language. They all know everything that needs to go into operating the AI the right way. So instead of everybody doing it differently on their own, where nobody is doing a prompt the same way and everybody's doing it differently this way, everybody can share their results, everybody can understand what somebody else is doing and anybody can come in and troubleshoot somebody else's prompt as opposed to the other ones. Like I said, the word vomit prompts, it's really hard for somebody to go in and say, oh, I know what you did wrong.

So this is where it gets intriguing, right? Because what I'm hearing you say is you're using notion. Other people might be using spreadsheets or Google Docs or Airtable or who knows what, right. There's a bazillion different things that you could be using. You mentioned earlier the concept of hot swapping and stuff like that. And that's. That gets my wheel spinning. Like once people understand this, this canvas of these nine different concepts we've been talking about the context blocks, I think that's what the canvas is, right?

Yeah.

How do they actually hot swap stuff? Like help me understand how that begins to work.

Interestingly enough, this morning I was on a call with one of our students and she got stuck on the context block, couldn't quite understand what it was and I had to explain it to her. And so it was a one hour meeting and I said, look, here's what I'm going to do just so that you can see how all this comes to play. And I showed her a bunch of different prompts so she could understand how it came in. But I took the transcript because, you know, we use AI to record all the meetings. So we used fathom. I took the transcript and then I ran it through Claude using six, I believe, six different prompts to analyze a transcript so that it would get six different outputs, right? So one of them would be to write an email that that tells everybody what was discussed and what are the action items. One of them was for us to analyze the meeting and say what were the problems that we helped on that support call and what were the aha moments where the student really got it so that we can now teach it better the next time? And then there was one that was just a summary that we would put into our CRM to make notes on what was on the call. So all these different prompts we used to analyze the exact same bucket of information. The only thing that we were swapping out, in some cases, we were swapping out the style so that it would sound differently. But most of the time we were just swapping out the request portion because the style and the role and the context was all the same. You know, the context in this one was just a block that we called transcript. And so you're just literally swapping that out. But anytime somebody wants to, let's say, just analyze a transcript and send out the to do list, typically they're going to be embedded in your AI tool anyway. But if you want your own custom one, you build your own custom one. And the only thing that you have to swap out is the transcript itself. So you literally just cut and paste, or you just build a GPT that just says wait for somebody to paste in a transcript and spit it out where you've got the instructions. Already been into the GPT.

Okay, I love this example, right? You had a call with a client, with a customer, and you recorded it and you had it do different things with that data set. Now, had you already built out your cloud projects or your custom GPTs to do these kinds of things and all you had to do was just essentially provide the context or was it something else? I'm curious.

Yeah, so in this particular example, I had six different transcript prompts in our database. And so I just popped them all in there and I uploaded it into our community so that everybody could see how these things differ. Because I figure, hey, if it helped one person, I might as well share it with everybody else. So now everybody's got access to all six of these prompts and they can see, oh, but what will happen is like, for instance, the one at the end where I said, I want you to analyze this support call and tell me where the aha moment was and what we could do to better teach this concept in the future. Think about all the different companies that could look at that particular prompt and go, oh, you know what? If I just modify this, it'll apply to this scenario over here.

How are you triggering this, by the way? Because I don't think the tools are smart enough to actually have variables. Right. This is mostly a copy and paste job. Or how are you actually getting these variables, if you will? Because we're kind of talking about a concept of variables here, right?

Yeah.

How are you getting those into the various AI systems?

Well, there's Several ways. One, you know, we just tell it, here's the variable. And a variable might, might be. I'll give you a couple of examples for like, in style, we would have a variable that says variation in sentence length. And we would have a calibration there that where it would be really long sentences or really short sentences or a variation thereof. We have another one called humor. You know, and instead of just saying I want you to be slightly humorous, I got a gauge from 0 to 10, you know, I want you to be 10 is. Is I want you to be ridiculously funny. And then a zero or one would be, let's just cut out the humor. Right. Same with words like, you know, jargon, technical jargon. I want you to really dial back the technical jargon. So those are all parameters and variables.

Did you have to train it or did you just tell it, presuming it was smart?

So I've identified and the list keeps growing, but you'd be surprised how smart AI actually is. But to extract a variable from AI is not as difficult as you would think. Like, I've identified at least 56 different style parameters that I can use when writing reading. There's no way I would use all 56 because it would just be a chocolate mess. But I noticed a few. I'm like, man, how do I. I literally would just ask it. I'm like, how do I get you to quit saying this or to be a little more brief, or how do I get you to quit writing these 11 word sentences where every freaking sentence in the paragraph is 11 to 14 words? You know, and it'll come back and tell me. But it would typically come back in what the kind of prompting that everybody else does, which is a paragraph. It'll say, here's how it describes how you would prompt it in the future. And I was like, that takes too many words. What's a variable I can use and a numeric I can adjust to fine tune it?

Ah. And we'll tell you. Okay, and are you finding that's the same for the different models you're using? Like, if you figure it out in one model, it works in the other model.

90% of them work in all models. But those occasions where the model requires you to give a description as well as a numeric. So in other words, instead of saying, I want a humor level of five, it might say humor equals moderate humor, six out of ten.

Okay.

You know what I mean?

Yeah. So you can get to these variables by asking it. Okay, so this is really interesting just to Kind of help summarize this. What I'm hearing you say is the strategy canvas. These nine blocks are things that you have complete control over. You can put in there whatever the heck you want. There's these other variables that you can layer in there, which are really more variables about the way you want the output to sound. Right. Do you want it to be more. Shorter paragraphs, longer paragraphs, humorous sentence variability, all that kind of stuff. And you can get at that through experimentation. But what I don't think you can do is you can't say, like, you can't just put in a variable for my company and it's going to know what the heck it is. Unless you're inside of a custom GPT and you've told it, this is what it is. Right?

You'd be surprised.

Really?

Yeah. I can actually go in and let's just say if I know your URL and I want to know, who is your target audience, who is your icp, where is your primary location? All of those are variables, are they not?

Presumably, yes.

Who's your leadership team? All that stuff. All I have to do is give it your URL, I have to spell out what those variables are, and I tell it to go solve for those variables. Variables, huh? Wow. And now it comes back and it gives me all.

And then in the future, you can just reference those variables and like the CEO of Social Media Examiner.

Yeah, yeah. You just. You just word them up as variables now? Correct.

Fascinating. All right, tell us a little bit about how you're using this to write your book, just because I think that'd be really interesting to people.

Oh, yeah, Very meta. Using. Using AI to write a book about AI. So, first of all, I've been teaching people this for almost a year now. Well, actually two years, but for a fee for one. But so I have a process, right? And that's the process that I want to teach people. So what I first did is, is I gave it all the information about our process. What's scalable prompt engineering. I gave it examples of the prompts I gave it. This is the crazy thing. I gave it a copy and I keep pointing over here to it because I'm looking at it and you're not, but I gave it a copy of the AI strategy canvas just to see if it could analyze it and tell me what it was, and it. Freaking dead. I mean, it just explained it to me and I was like, yeah, you nailed it. Okay, cool. So that becomes a source document. So then I just went into it and I said, like, look, here's all the documentation. I want to write a book on this explained to people. And I created a prompt that would basically say, give me an outline for the book. Gave me the outline for the book. I modified the outline and then I ran that outline against another framework for writing business books, and it modified it still. Then I would go in and I would start writing it chapter by chapter. And I would always insert my own thoughts into the prompt. So I'm using scalable prompt engineering to prompt a GPT that has my stuff in it to actually write a book about prompt engineering. So that's kind of how I do it. And I would write a chapter and then I would take that manuscript that may only have one chapter and a table of contents about what's coming. I would then upload that into the GPT and then write chapter two. And then I would replace it with chapters one and two and the table of contents and I'd write chapter three. And I would do the same thing over and over, just rinse and repeat to where now I had the full manuscript in there, and then I had it analyze it against the standards for writing a book. Give me more feedback, it would give me more feedback, and then I would modify the chapters. Then I uploaded a bunch of testimonials and I said, now I want you to go analyze the book and tell me where these testimonials would be be most appropriately placed. Boom. It comes back and it tells me where to push those testimonials. You know, so anyway, that, that's pretty much how I've been right now.

I love this. And you know what's really cool about this, John? Once you get this thing done, you can just go ahead and change who your Persona is and say, hey, write me a version of the book for small businesses and then write me one for brands. You know what I mean? And then you could essentially take the exact same framework and you could publish a variation of the book for small businesses and for brands, and you could do another one for consultants or agency owners. And I would imagine you could quite literally take your message and have it substitute out areas where it's going to have to come up with different examples and stories. And now all of a sudden you've got a framework that you could pump books out all the time if you wanted to. And then you could begin having it prepare decks for you too, right, for when you go speaking.

Oh, yeah, look, I've already done that a million times. So the, the GPT or I have it Both I have a GPT and a Claude project because I like, like the way Claude writes copy better.

Right.

And I like the way ChatGPT analyzes better. So I'm literally going back and forth with each one. But I had to do a webinar last week, actually. Was it last week or was it. Yeah, it was two days ago. I have. I have three of these. One we're doing in Southeast Asia and Australia, one we're doing in the EU next week. Yep. And then one we're doing for North America the week after that. So I had to create this webinar. So I just go into the same GPT that I'm using to write the book. I tell it to write the webinar outline, I modify the outline, and then I try to polish it up again. And I have a whole series of prompts that actually creates the slide deck. Because there's a methodology to how I like slides to go. I don't want it. You know, if you just say, hey, go create a slide deck and all this junk, it'll create just absolute garbage. But what you want to do is tell it to. Don't put any more than four bullet points. Don't put more than two slides in a row that have four bullet points. Make some slides just have one primary concept in a word block that you can talk around, write the speaker's notes for it, you know, and it basically created it all.

It's amazing. Now, John, first of all, thank you so much for sharing your insights with us. Tell everybody what your preferred social platform is, if they want to connect with you on the socials, and then if they want to learn more about your business, where to send them, and if you want to have a/SME URL on the end of it to get your thing, you can do that as well. So where do you want to send everybody?

Well, most people can reach me on LinkedIn. That's where I spend most of my time. I keep threatening to get back on X and I just.

There's a lot of AI stuff on there.

Yeah, no, I need to be. I'm just not that prolific, you know, I get it. So, yeah, on Bazooka, I believe I'm JW Munsel on LinkedIn, but. Or you can go to Bazooka.com B I Z Z uka.com and what I'd like to do, for your viewers only, Mike, is if they want, I would be happy to give them a free copy of the book when it's published. This will give me the incentive to write the doggone thing. Get it out there. If I got a lot of people out there who want it, do you.

Want to send them to bazooka.com SME would that work? You think you can?

Here's where I would like for them to go. So if you'll go to bazook bizzzuka.com Ingrain I N G R A I N and in the comments box you just tell me that you heard about it on the SME podcast. Then I'm going to put you in line for a free copy of the book as soon as it's published.

Perfect. John, thank you so much for sharing your insights with us today.

Hey Mike, thanks for having me on. It's been a blast. I've always been a fan of your work and I just hope that I added some value to your group.

Hey, if you missed anything and you want the notes, simply visit socialmediaexaminer.com a27 Be sure to follow us on whatever app you're listening to so you don't miss any of our future episodes. And if you like the show, would you let your friends know about it? If you want to tag me, I'm Telzner on Facebook, Stelzner on LinkedIn, and iKestelsner on X. And do check out our other shows, the Social Media Marketing Podcast and the Social Media Marketing 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 Explored podcast is a production of Social Media Examiner, AI Business World early bird pricing ends on January 16th. You can save $300 on your tickets. Register at AIBusinessWorld live today.