Marketer's Guide to Advanced Prompt Engineering

Dan Sanchez [00:00:05]:
Welcome back to the AI Driven Marketer. I'm Dan Sanchez. Friends, call me Danchez. And I am still on this journey to master AI in 2024. If you remember from the first episode, my goal was to really learn how to even put together breadcrumbs towards mastery. And now that I am over a quarter, 4 months into this journey, I have actually learned a lot about what I'm now calling the AI fundamentals that all marketers should learn if they're getting into artificial intelligence or how to use artificial intelligence to do better marketing. So with this episode, I wanted to do a I guess for the podcast, it's a 5 part series. For everybody else who's not listening to this via podcast or YouTube, this is video 1 of a course that I'm putting on, around the AI fundamentals.

Dan Sanchez [00:00:53]:
Like, everybody's gotta go through it before you can even get to the advanced stuff that you'd wanna do for marketing and setting up automations and making new personalizations in your email campaigns or any of that kind of stuff that's really fun, you have to have the fundamentals. And I wanted to start with video 1 here around how to prompt better. Yes. Oh my gosh. It sounds so boring, and this would have sounded boring to me. But I promise, this is the foundational skill. If you don't master this one skill, you can't move on. It's like it's the learning how to dribble with a soccer ball before you can really learn how to play soccer well.

Dan Sanchez [00:01:27]:
Right? It's those things, like learning how to dribble basketball. Right? Pro basketball players practice the fundamentals over and over and over again for hours every morning even when they're become pros, because we know that drilling the fundamentals is what sets up for the higher level game. And the higher you go, the more your fundamentals need to be solid. So I wanted to start with this because it it's like the thing that I keep coming back to over and over, the more advanced that I get, the more I'm like, man, I need to go back to the fundamentals again and learn something that I haven't understood before because this is where this is where it all starts. This is why we call it prompt engineering and not just prompting. But I wanted to come back to lesson 1. We're gonna spend a few lessons on prompting. You'll see how it starts to stack up in further lessons.

Dan Sanchez [00:02:12]:
If you skip this lesson, lesson 2 in this little 5 part course is going to be even harder because it relies on the lessons from part 1. So let's talk about this. When it comes to prompting, most people go to something like chat gpt and ask it a very simple command and expect greatness back. And while AI is extremely intelligent, even with what we have, at the date of this recording, which is chat gpt 4, for turbo or whatever, it's still limited. You still have to help it along, just like you would have to help a human along. You can't just go to a marketing manager and be like, hey. Write me a marketing strategy. He'll be like, okay.

Dan Sanchez [00:02:50]:
And they'll probably give you a plain vanilla marketing strategy. AI would do the same thing. It'll give you a plain vanilla marketing strategy. It's kinda like, okay, it's kinda looks like something I copied and pasted off the net. Learning how to prompt better is learning how to give more specific instructions so that AI can give you a more specific result. Okay? Because what I like to call AI is my genius intern. It is really good at figuring out all kinds of things, but it's like an intern. You gotta you gotta hand hold it through the process.

Dan Sanchez [00:03:17]:
And so what I wanna do is help you understand what

Dan Sanchez [00:03:17]:
that process looks like, or kind of give you a AI community a super prompt, Right? How not to just ask it a simple thing, but how to give it a very beefy set of instructions so that you can get a very beefy result out of it. Right? I wanna walk you through it step by step where I level it up very simply from a lame prompt all the way up to a super prompt, right? And there's a framework that I've been using ever since I got exposed to this idea of super prompts for almost a year now. I actually discovered let's jump to this slide, and if you're only listening to this, I'll spell it out for you. This is a screenshot of a chat GPT cheat sheet published by Element 451 when I was working there as a marketing director. You I'm gonna source it here. It's just at element5451.com/blogchatgptcheat sheet. You can find this full cheat sheet. This is literally a screenshot of just the bottom section of this cheat sheet.

Dan Sanchez [00:04:21]:
It's a great cheat sheet. This was the most valuable part of it, and it shows you how to deconstruct a a super prompt. And I use this all the time, as as a good cheat sheet for me even crafting custom GPTs and better prompts. So I come back to this all the time. It's a really helpful tool, And it breaks down a super prompt into 5 different sections of role, result, context, intent, and constraint. You could see it gives some examples of how a super prompt might be crafted to craft a tweet. So your role, it gives an example of role, act as a social media expert. Result, create a Twitter thread with no more than 5 tweets.

Dan Sanchez [00:05:01]:
The post informs followers about the new university research on global warming. Intent, the goal is to increase attendance on the upcoming lecture. Constraint, keep each tweet 280 characters and use relevant emojis and hashtags. Right? So I'm gonna show you what this looks like. And a piece that I've done recently where I started with the super prompt, I actually reverse engineered it down to its most minimal normal prompt that most people use, and show and progress it back up to show you, like, the difference it makes to break a prompt down into these 5 sections. So but first, let me talk a little bit about why we need to define all this stuff. I'm gonna use this little image to illustrate it. If you can't see it, it's just a I don't know.

Dan Sanchez [00:05:40]:
I think this is a stock photo, and it's just a cluster of nodes, and they all look it looks complicated, like integrated. But at the end of the day, these large language models are essentially clusters, almost like these, like, super sophisticated mind maps of relationships between words and the probability that they belong together, so it can more accurately predict which words it needs to give you based on your prompt. It's more of an art than it is an I mean, it is a science, but prompting and prompt engineering is more of an art. It's not like coding where you're saying, hey, insert variable here and then deal with it in this way. It's more loose than that. And I'm gonna kind of explain that as we go through, because we're trying to navigate this loosey goosey maze of associations that these AIs have learned in order to get the better results out of it, which is why we engineer our prompts to be able to try to get it to be more consistent. That's why we engineer it. So let's dive into an example of a chat that I did.

Dan Sanchez [00:06:42]:
And I'm just in OpenAI's Chat GPT. I highly recommend paying for the Chat GPT Plus if you haven't already. This is where I spend most of my time and all the AI stuff. And eventually, while you can get away doing what I'm doing right now with the free version of ChatGPT, if you wanna get to the videos 45, you're gonna have to pay for the plus. I'm sorry. I'm making this course free, but you're gonna have to at least pay for 1 month in order to experiment with building some custom GPTs. It's one of the best things on the market right now to really learn and hone your AI skills. And paying for chat GPT Plus is one of the best.

Dan Sanchez [00:07:16]:
I'm not sponsored by them. I don't get any kickback. I'm just saying this is the best AI tool to pay for. If you're only paying for 1, it's this one. So let's break down my really simple prompt. And I'll be going back and forth between this and the image that I had out because remember, super prompts have 5 things: role, result, context, intent, and constraint. Now in this in a very basic prompt, the one thing that everybody kind of does is what we start with in the super prompt is the result. What do we want? Right? Because if you didn't give it anything and you just kinda said, hi, it would it would just say hi back because it doesn't know what you want.

Dan Sanchez [00:07:53]:
So most people are pretty good at understanding you have to ask for something. So that's what I've started with. Hey, write me a LinkedIn post about sales. We all know this post is probably gonna be garbage, but AI will always try to fulfill the thing. So it's gonna try here, and it wrote me a LinkedIn post about sales. And it's got rocket ships on it, unlocking sales success, the power of adaptability and learning, and it goes and outlines continuous learning, customer centric selling, and leveraging technology. Lots of it asks a question at the bottom and has a bunch of hashtags, and it's it's just kind of a plain vanilla post, not a very good one. It looks like a chat GPG generated post.

Dan Sanchez [00:08:31]:
Right? If you've ever seen these if you ever ever seen AI come up with social contents, like, yeah, this and it honestly looks more like a small blog post than it does a social post. But so let's go into how do we get away from writing lame prompts like this one, how to level it up into a super prompt so that we can get something so much more specific and helpful and actually useful as a marketer. Right. And I'm using this one as the illustration, though. This plays across anything you'd ever wanna ask GPT. So you have to think about how does this apply to what you wanna get out of AI? How can you level it up into a super prompt? Maybe you start small like this one, and you continue to add on to it like I did in this illustration, but just know it can do more than write social media posts. This is just an easy example on a place we've all probably have used AI before if we played with it. We're gonna show you how to get the most out of it to at least do this one action, but it applies to everything, everything.

Dan Sanchez [00:09:22]:
So let's dive in. What can I do to go next? Well, if we look back at our graphic, what the next one in line is the role. You'd be surprised that defining the role actually can change the post quite a bit. Let's jump into the first role I assigned to it. You are a famous LinkedIn creator and sales expert. Write a LinkedIn post about sales. Okay? So I stuck with the same result that I wanted, write a LinkedIn post about sales, but I gave it a role. And if you look at this, it says unlocking this here's the result that it generated for me, unlocking the secrets to sales success, a blend of art and science.

Dan Sanchez [00:09:57]:
Okay? So it gave me a kind of the similar format. It's almost more of a blog post than it is a social post. It's got the hashtags. It finishes with It's hard for me to judge if the content's actually better. It seems like it's a little bit better, but this role didn't really produce that much of a difference, in my opinion, as a result as it did from the first one. Right? Maybe it's a little bit better, but not that much better. But hold on. Before we throw away this role thing, let's change the role again and see if that changes our result.

Dan Sanchez [00:10:29]:
I tried another one, you are a VP of sales at a fortune 500 company with a passion for mentoring sales directors. Same result. Write a post about LinkedIn. Write a LinkedIn post about sales. Let's see what it did. Now you the intro of this one is much different. It gave it a title with wrapped with rocket ship emojis, elevating sales, a strategic approach for a sustainable growth. As a VP of sales at a fortune 500 company, I've had the privilege of navigating the highs and lows of the sales word world, learning that success in sales is as much about strategy as it is about execution.

Dan Sanchez [00:11:02]:
Here are 3 core strategies that have consistently driven our teams to achieve outstanding results. Now this post is still not great because it's just making up stuff. Right? But can you see the difference that it's changing here? It even even its outro is changing. As leaders, our role is not just it's writing as the role of a VP of sales at a fortune 5 member with a passion for mentoring sales, and that's reflected in this post. So think about the role you want to have these things coming as. Are you having it right as you? Maybe you need to describe your role. Are you having it right as a guest? Are you wanting it to behave in a certain way? Do you want it to behave as an expert, as a layperson? What do you want it to do? Right? Who do you want it to behave as? That's why defining the role is important because you see, even in this generic write a post as about LinkedIn post about sales. It's a generic result.

Dan Sanchez [00:11:58]:
You could see how the role has changed this specific very specific role has changed the post. So it's good to define the role. Let's move on to the next one in our super prompt pieces here. I think I'm gonna move on to beefing up the result a little bit. No. I'm gonna try a different role real quick. I forgot I said one more. You are a CMO of a series b startup, and you could see this role also impacted it.

Dan Sanchez [00:12:22]:
Even though it's kind of a weird role to be writing about sales, a CMO of a series b tech start up, it actually does reflect here too. So let's go on to the next one. This one, I changed the context. So that was the next one in the super prompt formula here is we have the role, we have the result, which is write the post. Here, I'm giving it more context, a way more context to make this post way better. Context is literally like if you if you could just give it one thing, give it context, and it will write way better for you. So here I gave it a ton of context. You'll see how.

Dan Sanchez [00:12:53]:
First, let's go back to I get I changed the role very specifically to a very to a specific person. You are an expert sales coach and former VP of sales, Matt Brownlee. Now this Matt Brownlee is someone that I had the privilege of producing an episode for on the podcast Close Mode, which I am the producer for that show. And Brian Dietmeier was the host. Brian Matt Brownlee was the guest. And now I'm giving it con so that's the role. The role I'm giving it is Matt Brownlee who is, in real life, an expert sales coach and v and former VP of sales. The context based on the transcript from the episode result, some more specific result to find one solid idea in the episode and turn it and and turn that idea into an actionable post for LinkedIn.

Dan Sanchez [00:13:42]:
That's a much beefier. And then I'm loading the whole transcript from the episode right in the prompt itself. So that is such that is a much more specific request. I'm giving it all the content to pull from, and I'm giving it almost 2 steps here. Right? Based on the tramp skips, find 1 solid idea, step 1, and then turn it at that the idea into an actionable post for LinkedIn, step 2. So I'm actually giving it steps in here, and let's see what it created. Let me scroll down to the bottom of our page here, and it even gives us some feedback based on the rich insights shared by Matt Brownlee in the transcript, an excellent idea to transform it to an actionable LinkedIn blog post. So here's the blog post.

Dan Sanchez [00:14:24]:
Post title, mastering sales with the Discipline of an Athlete because that's what his post is about. It was like correlating focus and discipline from his triathlete, and marathon running times into sales. So and you can see it's he'll may may read a little bit more. Whether you're towing the starting line at the Boston Marathon or preparing for a critical sales meeting, the same principle apply same principles of discipline and focus apply. Prep and then, he outlines 4 4 parts. Preparation is key. Coaching makes a difference. Team drives performance, results define success.

Dan Sanchez [00:14:58]:
And, of course, each one of those has a sentence or 2 expanding on it. And then he the post kinda wraps it up. Again, it's still kinda long. It's still kinda like a mini blog post more than a a LinkedIn post, but dang, this content is now original because I fed it a whole

Dan Sanchez [00:15:14]:
freaking transcript with the original content. It's no longer plain

Dan Sanchez [00:15:14]:
vanilla content. It's based content for, if you ever wanna make social content using AI, it always always works best if you feed it something to start with. If you feed it a blog post, or maybe you feed it at least a tweet to inform your LinkedIn post. It's got something to go off of, and it has original content. You wanna think of AI for content marketing more of like a repurposing machine more than a content creation machine, and it will do much better. But I also honed in, the context and the result as well. I didn't just say, hey, based on the transcript, write a LinkedIn post. I said, hey, first find one solid idea in the episode, and then turn that idea into one actionable post for LinkedIn.

Dan Sanchez [00:16:04]:
Because that's what I would do as the producer normally is to find one good nugget. This one still didn't follow through on the directions because AI sometimes has a hard time doing multiple steps in one go, but sometimes it does. In this result, I am pretty happy, and it's made a huge step forward from our last post where I was just giving it a more specific role. So now I've layered on a highly specific role, a highly specific instructions and tons of tons of more context. But we can go farther. Let's go farther. Come on. Let's see.

Dan Sanchez [00:16:38]:
Next, we have intent. So what happens if we add intent to the post? Let's see. In this one, I did the same thing with the trans got the transcript, you are the expert sales coach and former VP of Sales, Matt Brownlee, and then the same instructions. But this time, I changed and added one more sentence. I said, finish with the line, and I put in quotes, you can find my full episode at close mode dot media because the goal is to entice oh, I left that quote open. I wonder why it screwed it up at the beginning. Because the goal is to entice readers to listen to the episode. Right? I gave it the intent.

Dan Sanchez [00:17:11]:
What what are we creating this for? Just like when you delegate to humans, you need to give them the end state. What is this supposed to actually accomplish? It usually leads to better results, and that's true for AI. So let's see what AI did. Let's see. It wrote the post, and the post honestly, this is now totally a blog post more than a LinkedIn post. So this is where the engineering comes in. You have to, like, tweak it and tweak it in order to get better. But you could see at the end here, you can find my full interview, close mode dot media, and it's hyperlinked where we delve deeper into these insights and discuss how to apply them to achieve sales excellence.

Dan Sanchez [00:17:43]:
Join the conversation. Let's evaluate our sales game together. Hey. Do you think this post is now if if it wasn't a total blog post up here, but if we ended with this, you think they would be more likely to accomplish accomplish the result of getting more people to listen for the full episode. Of course, because there was nowhere even in the previous post that was mentioning the episode. It had no idea that the reason why I wanted to write a LinkedIn post was to get more people to listen to the episode. Now that I've given it a very specific, like, hey. Include this at the end.

Dan Sanchez [00:18:14]:
Here's why. It's actually included exactly what I wanted to at the end and given a a persuasive little sentence to have them go join us by listening to the episode. So that's why we add the intent. Now this is the fun part. This is where you can really start honing this sucker in, and you'll see when we show the results of this next one, what happens when you start to add in constraints. Right? Because even this last one, it kind of, it's so long, and it's it's still better than the other ones because it's based on original content. It's based on that podcast transcript, and still coming from Matt Brownlee's perspective. So that's good, but it could be so, so much better.

Dan Sanchez [00:18:53]:
Let's layer in some constraints and see what happens. So, again, I have that same post going on up here. And now I still have the transcript at the bottom, but let's layer in this one more thing. I put in other instructions, colon, and then I just gave it a bulleted list of things that the AI should know before writing this post. So I give it the instructions right in the first person from the perspective of the guest. It was already doing that pretty well, but just in case, I included that because it'll often because it changes sometimes. Write in short, choppy sentences with harsh line breaks. And I put the word harsh because that's kinda what LinkedIn people do.

Dan Sanchez [00:19:29]:
They write in that style where they have, like, one word sentences on a whole line or just a very short sentence. The next one is do not use hashtags or emojis. They did it in the last one, but it often still does. I don't know why, but chat gpt thinks that emojis and hashtags have to be in every LinkedIn post often. And then last instruction, do not include a title, but do start with a short attention grabbing first line. Right? So I've layered in, these are instructions, but they're actually constraints for the post to kind of narrow down the scope of what we want. You can see how this post is a very long prompt. Of course, the transcript's really long, but just look at the almost two paragraphs of information we've given AI now to go off of.

Dan Sanchez [00:20:10]:
Because, again, the more specific you can be in your request to AI, the more specific the result, which is why super prompts are so important. Let's look at the results. Here it is. Your network is your net worth. In sales like marathons, the right preparation and the right team can make all the difference. Line break. Imagine starting a race already knowing the track, the pitfalls, and having trained exactly for those conditions in sales. This translates to understanding your client's needs deeply and crafting solutions tailored just for them.

Dan Sanchez [00:20:42]:
Now pair this preparation with a strong team. And he goes on, you can see if you just take a quick look at the video if you're an Apple Podcast or on YouTube, this has all the line breaks onto it. It looks much, much closer to something you would find on LinkedIn. And it finished. You can find my full interview at close mode media dot, close mode dot media right at the end of the post. So this is actually starting to look something like I would actually post to LinkedIn. Why? Because a super prompt made it more specific. So take this lesson about super prompts, about defining the role, the result, the context, the intent, and the constraint, and go and practice it.

Dan Sanchez [00:21:21]:
Layer it in. Change what type of role you're speaking for, and just play with it. When people say play with it, this is how you play with it. You experiment. I guess more than play is experiment to see

Dan Sanchez [00:21:32]:
what different roles produce. And then change the result a few different ways and

Dan Sanchez [00:21:32]:
open up new chat produce, and then change the result a few different ways and open up new chat gpt windows to see if you can hone it down further. Play with the context. Play with the intent. See if you can narrow down the restraints and get it to act in a certain way. It's about experimenting your way to success. But expect a lot of failures on your way there, because once you can get better and better and better at crafting these prompts, it's going to level you up not just in your ability to get something quick out of GPTs, but to make even much larger projects, which we're gonna cover in the next video. How to level up even what we're doing in SuperPumps with what I call the step method in order to get even more consistency out of AI. So stay tuned for that course video coming next.