10x Your Output With Robust Custom GPTs

Dan Sanchez [00:00:04]:
Welcome back to the AI driven marketer. I'm Dan Sanchez. My friends call me Sanchez, and I'm on a journey to master AI in 2024, and I am so excited. I've learned so freaking much even in the 1st 4 months here. And this is going to be the last part of the AI Fundamentals course that I've been trickling out here on the podcast and on YouTube for the last week or so. This is part 5 where we're going to be covering complex GPTs or how to build them. Because in the last video, we covered how to build a simple custom GPT in chat GPT, and we learned a lot about how to load instructions, how to load knowledge docs in order to get GPT to do more than just one task at a time, how to get it to essentially loading your prompts to load multiple steps into accomplishing a very specific, highly targeted task. So in this video, we're going to be extending that because custom GPTs, I swear, Andre Yee in a previous episode said this, and I I just thought it was like one of the smartest things I've heard on this episode so far is that custom GPTs are like the Excel for AI.

Dan Sanchez [00:01:14]:
Excel was to programming where you could use it to build a whole bunch of different things and prototype different formulas and ways of calculating things in a database. So custom GPTs are for building prototypes or ways of interacting with AI. Right? So if you can master building custom GPT, this is the best foundation you can build for really learning how to leverage AI for almost everything else. This is the foundation of learning how to prompt, learning how to build, not just for one off task, but for systematic tasks and projects. So I've been building to this point because there's so much foundational stuff just to be able to build custom GPTs around super prompts, the step method, and getting in dabbing our toes in the water for simple GPTs. And in this one, I wanna finally open it up, and I'm like, okay. Let's learn about custom GPTs, about what they can do and what they can accomplish. At least for, what is it, May in 2024, they're going to open up and be even more robust in the future, but at least this is a first step in there for now.

Dan Sanchez [00:02:16]:
For starters, we've covered what it was and kind of how to use the basics. So if you're just brand new to this topic, go watch the first step. I'd recommend starting over with part 1 of this 5 part course. But if you wanna just jump into custom GPTs, go watch that one first. It's the primer on custom GPTs. We build a simple three step process one. It leverages a lot of the stuff from the previous videos. But if you want to dive into the deep end, here we go.

Dan Sanchez [00:02:40]:
So custom GPTs, what makes a complex one from a simple one? Right? That's the question. In the last one, we only had a 3 step one, and it was really simple. We put a transcript in, we put text in, it gives us text back in the form of a LinkedIn post. That was the one we built together. And this one, I'm going to be modeling one that I've been working on for months now, and I'm starting to do like a revamp of a 2 point o version of this very fairly sophisticated GPT called My Showrunner. And I wanna talk about what makes it more complicated than the one we had done previously. The cool thing about GPTs is there's a few factors Before, we were just playing with text back and forth, but GPTs Before we were just playing with text back and forth, but GPTs can actually do multiple things. Here, I have a list for you, so I'm just gonna read off from this list.

Dan Sanchez [00:03:36]:
As inputs, things we could put into the GPT as a user, we can of course do text and have prompts in there. We can add code, Yeah. That's a whole thing. We can add documents, text documents, PDF, Word docs, and other types of docs we can add to it. We can add in spreadsheets and it can actually understand the tables in there. We can do web searches with it or have it go and get web an input from from Bing and have it go do some research on the web, that's cool. So that's 5. You can load in images for 6.

Dan Sanchez [00:04:07]:
And audio, you can drag and drop audio files on there, have it transcribe it and move forward. So that's 7 different inputs that I've counted. There there's probably more, but those are the basic ones that I'm using, the most often. Now those are the things you can put into it. And of course, it's AI. You can actually understand it, analyze it, process it, do things with it, manipulate it in order to get different outputs out of it. Some of the outputs you can get of course are text, that's what we were doing before. We can also get code useful if you need little, I use it for little JavaScript snippets around my websites all the time.

Dan Sanchez [00:04:40]:
You can also get tables or spreadsheets back. You can also get images because of the integration it has with DALL E. And soon, Sor will be in there, will be able to get videos out of it which will be really cool. But until that day, we're kinda stuck with these 4 outputs, text code tables, a k or spreadsheets and images. But that's a lot. That's a lot more than we're used to dealing with, and you have to think about how to leverage all these in your work. So before we jump into my showrunner and looking at what makes it complex, not just the amount of inputs and outputs, but also the sheer amount of steps that it has to walk through that makes it a little bit more robust than a simple GPT in order to accomplish many tasks in a larger project. I wanna just take a moment to have you step back a little bit.

Dan Sanchez [00:05:27]:
I recommend grabbing some pen and some paper and taking some time to take inventory of your current job, of the things you're doing all the time. This will be helpful as we're going through this because I need you to understand that the power of this is being able to automate what you're currently doing. And it might not be able to do it as well as you. But as AI gets smarter, it probably will be able to not just do as well as you, but do better than you. The thing you need to learn to do is to break your job into tiny little baby steps so that you can teach AI how to do your job effectively one little step at a time. Because there's one thing AI is not good at doing. It's taking a single prompt that's actually a fairly broad and large project and breaking it up into baby steps itself. It sucks at that right now.

Dan Sanchez [00:06:13]:
I know they're working on it even with GPT 5 coming out around the corner. This is one of the big things that they're claiming they're working on to make it better at doing. But until then, GPT 4 is actually really good at accomplishing the individual tasks if you break it down for it. If you break these projects down into smaller baby steps for it. So take some inventory, pause this video, and take inventory of all the tasks that you do on a daily or weekly basis, maybe even a monthly basis. What are those things that you do that you're like every time you do x, you have to follow-up with y. Every time somebody emails you about this, you have to run this script, this step by step process every single time. And if you don't think you have a step by step process for something you do over and over again, I wanna challenge you.

Dan Sanchez [00:06:59]:
Take a moment to think about what you're doing. If you had to delegate it to an intern who was a genius, but lacked common sense and you had to break down the baby steps of how to do it, think about it. Just pause the video, take a moment and do that now. If you're still listening because you want to get ahead of it, I don't blame you. I do the same thing a lot with videos all the time. I like to preview and then come back and then actually learn the thing. So let's dive into what I've done, and I just want to use my showrunner as an example in order to show you how I've done this before in the past to hopefully give you some inspiration and some insight to how these things can work. Again, to reference something I said in the past video about simple GPTs, there's no right or wrong way to build GPTs.

Dan Sanchez [00:07:41]:
Though after building a few dozen of these things and failing with some and succeeding in others, I have kind of honed it into a specific way. And I found that others are using similar methods in how to craft these fairly sophisticated GPTs to accomplish very specific tasks. So there's other ways to do this. And just like Excel, there is not exactly a right or wrong way to organize the information. Some ways work better than others. Right? It's a generic blank canvas tool that you can do a lot of things with, and that's what's so exciting about it. But hopefully, I can teach you some of the things that I've done and developed myself, and you can pass it on. Or I could pass it on to you to grow and expand even further.

Dan Sanchez [00:08:18]:
So let's dive in. I'm gonna be showing my screen now and diving in, and I'm going to be doing my best to describe along as I go what I'm looking at. So if you're only listening to this episode, you can follow along. But I am jumping to a more visual situation now. So if you're able to watch on Apple Podcasts or on YouTube or on the aidrivenmarketer dotcom website, you can follow along visually. If not, listening should be good too. So I've opened up chatgpt.com and I'm looking at the last custom GPT that we built together called the pod to post builder. It was a simple one that took a transcript from a podcast and turned it into a LinkedIn post, and we're expanding on this today because this one was simple.

Dan Sanchez [00:08:58]:
You see, it's kind of got like a pre prompt. It's got step 1, step 2, step 3, and then it kind of finishes off with some, like, more robust instructions and a and a final continue to make help me make modifications to this post post until it's complete. That's a fairly simple one. Again, it's only using text as an output. It's 3 steps. That's what makes it simple. It just accomplishes this one small task. But what if you wanna do a lot of task around one larger project? Well, that's My Showrunner.

Dan Sanchez [00:09:25]:
It's actually a custom GPT that I co developed with Susan Diaz and I'm here on my showrunner.com now looking at the homepage where if you put in if you go to my showrunner.com and put in your email and subscribe to our email newsletters, you get access to the instructions for free. Now I'm previewing that page where I walk step by step, how to get the instructions for it. This is the 1 point o. I'm hoping to release a 2 point o really soon, so stay tuned for that. But I'm gonna give you a preview of what the instructions look like for the 2 point o because even though this is fairly sophisticated on the 1 point o, I've learned a lot on how to better organize the instructions and make it smoother as well as added way more steps because you see this one has 8 steps. I'm currently on 12 and probably will finish with 15. 15 steps for this GPT to go through before it's all said and done. So let's dive into it.

Dan Sanchez [00:10:13]:
I'm looking at a now I'm back into t, chatgpt.com, and I'm looking at my own personal version of my show runner called the AI driven marketer showrunner because I use this for my podcast multiple times a week. Every time I have a guest, I run this this GPT because it does so good in doing the research on the guest, formatting the different angles we can take, then turning it into titles, and then even crafting an email to send to the guest. It's fantastic doing those things. I'm actually extending it now to do not only all the preproduction, but doing all the a lot of the postproduction. Let me show you. I actually use Notion, and I've switched over to that doc where I have my showrunner up here. Like I showed you in the previous one, I actually color code it here in Notion before drag dragging, copying and pasting it over to chat gpt in the instructions area. And this is where I I write it and organize it and keep it and troubleshoot it even if I scroll down to the bottom, show you that there's versions where I keep previous versions of the instructions so I can kind of reference what I've done when and where, in case something breaks and then I have to go back to our former version in order to keep to figure out what went wrong or why was it working before and now it's not working right.

Dan Sanchez [00:11:27]:
I I keep versions of it for that. It's also nice for clients so they can see the progress of what I'm doing in order to get to the the one that I'm sending them. Let's see. And as if you look at these instructions, you could see it's color coded. And I have a very robust opening statement with the premise with my show's premise. And then it goes through the same instructions that it did before. And you could see it goes from, let's see, step 1 all the way down to step 12 at the moment. What makes this complicated is that the more steps you add to a GPT, the tighter the instructions need to be without room for alternative interpretations.

Dan Sanchez [00:12:01]:
Right? If you've ever played the telephone game, you might have run into the thing where, like, one person turns around and gives specific instructions at the beginning and then gives it goes and returns around and gives those same instructions to the next person and then to the next person to the next person. By the end, you get to the end of the the chain, the instructions have changed dramatically, right, because slight reinterpretations of what was said was given over and over again. It's kind of the same thing with AI. The the more vague your instructions are, the harder it is for understand what you want. So the crisp the more clear it is, the better it works. So in this one, I essentially broke down the tasks that I was doing every time I was doing a pre interview and then all the steps that I was doing in a post interview with the guest, which often looked a little bit like this. And let me see if I can find the exact break point here. Here at step 9, instead of finishing with the guest email that would go to the guest, I was like, well, what are all the things that I do in post production? Well, one of the first things I do since I'm creating the title in pre production, that's already done.

Dan Sanchez [00:13:04]:
I'm actually creating the show notes. So I then extended this custom GPT by asking it like or having it prompt me because, again, I like building GPTs that proactively light lead me or the user through the process. And it says, hey. I have it. I instructed to say the very last part, then ask me. When you're ready, paste in the transcript of the episode below to begin the postproduction process before proceeding to the next step. Step 9, once the transcript is received, take a deep breath, carefully analyze what is said, and produce the following bit. The episode description and timestamps using the template and example in the in the doc titled episode show notes example dot pdf found in your knowledge area to guide your writing.

Dan Sanchez [00:13:46]:
You could see it's all color coded. Every time I have a knowledge document reference, I code it in brown here just to keep straight of when that's happening because it happens. You can see there's lots of brown areas where that's happening a lot. And it says write the intro and outro exactly exactly as it is in the template with the guest details filled in. Ask me if these look good before proceeding to the next step. You could see I'm using what I call the step method over and over and over again just like we built that simple GPT to craft these. I'm now taking that simple GPT and turning it into a single step over and over and over again because I've been putting the transcript once, and then we can use the same logic that we learned all the way through this course in order not to just come up with one piece of content, which is the show notes and the timestamps here, but to come up with every single piece of content that I could possibly get out of this one episode. So the next thing it does is, again, analyzing the transcript and the title to generate 5 visuals because, again, it's another output that a GPT can give is actually using DALL E to create visuals.

Dan Sanchez [00:14:47]:
I actually usually use mid journeys. I find it creates better visuals. But I have it actually write out descriptions for the images and I have it use it in a specific language rather than abstract language, which is better for Midjourney, but, DALL E can still execute as well. And then if I it'll give me 5 different options pre written and then I'll go and then say, hey. Yeah. Turn 23 into an image and then DALL E will actually execute it for me. And then that's what makes it complex but also more useful because now not only am I getting text content back for the episode, but I'm getting useful images I can use to promote the episode. Now these images won't be perfect because the AI is not I don't know, it's still up and coming, right? It's still got a lot of fine tuning it needs in order to be able to create full designs because that's what these little episode images are.

Dan Sanchez [00:15:36]:
I still have to bring it into photoshop, add the guest image, add a title to it, maybe even clean it up a little bit. But it's able to illustrate and create assets for me that I didn't have access to before. It was hard to find stock photos to create these kinds of things and I love that about it. In fact if you want to go in deeper I created a whole episode on how to get how to do graphic design and images with AI in a in a previous episode I'll drop in the show notes. But I'm using it to come up with the descriptions that I can then use with DALL E in the next step or then copy and paste over to Midjourney to create episode images for my for this particular episode. Now I'm still working on this, but I'm gonna add step 13, 14, 15 to come up with the LinkedIn post, which I'm just gonna take what we already did in the pod to post builder. I'm gonna take those instructions and put it put it in there. I'm also going to build a blog post writer, because even if you ask to turn a transcript into a blog post, it usually doesn't do well.

Dan Sanchez [00:16:31]:
But I've been honing for a couple of weeks now. I've been transcripts I can make out of transcripts. So I've been honing in that prompt for a long time. I've put hours into it. So by the time it makes its way into my my showrunner, this GPT, it's going to get pretty dang good blog post out of podcast transcripts. Unfortunately, it's limited on what kinds of blog posts you can make, but that's just the nature of taking an interview and making blog posts out of it. But it is it can create some blog posts particularly well. And and the last note I'd like to make is this knowledge area I have down here.

Dan Sanchez [00:17:07]:
You can see I'm using the step method over and over again, which is why I have lots of templates to go off of. Every time I wanted to create content I don't want it to guess at what I want that content to look like. I create a template and an example for it to go off of whether it's the guest email, the podcast type, the episode titles, the show notes, even the script I have for me to read as the host myself. I want to template everything so that it has very clear and precise instructions of what I'm hoping for as the user when I'm asking it to do these things. So I've just reviewed what my showrunner looks like, all the steps involved, how I'm using super prompts, the step method, and chaining these together in order to build to take what is a fairly robust process. Turning doing all the pre production and helping me speed up all the post production is huge for me because I do so many podcast episodes, not only for clients but for myself, for the show. Being able to take what used to take hours or even at the Sweet Fish, the agency I used to work for, I used to take 2 weeks to knock out all the stuff can now be done in less than 60 minutes by using a GPT like this in order to execute almost all the work that it took to turn it into all these assets. The only thing that's left is all the work of copying and pasting it in all the places it needs to go, which I'm currently working on a solution for.

Dan Sanchez [00:18:29]:
So stay tuned for that because as we've now taken all the work of creating content for it, it's created a backlog of then posting it all. So that's a solution that needs to be in the market that I'm hoping to work on and launch to this show soon. But in the meantime, hopefully you took a moment to think about and take inventory of all the things that you're doing over and over and over again, Pre production and post production for podcast episodes was something that I was doing over and over again. If you're doing it over and over again, just steal this GPT and use this to speed up your workflow. But if you're doing other types of marketing, you're doing customer interviews over and over again, What's the pre production and post production of those that you can build with the custom GPT in order to accelerate maybe the case study creation of those interviews? What are other tasks that you're doing over and over again? Are you having to analyze things, organize things, to look at CSVs and find a few bits and pieces of of data in them in order to come up with insights for that become content for a PowerPoint. Custom GPTs can help accelerate a ton of that work and this is where a lot of the real learning from AI happens and building these and testing them and going through modifications of them over and over, which is why I have versions of these things. Before I release this, I will have run it dozens of times in order to make sure that the outputs are actually consistent from what I was hope to what I was hoping for and expected it to produce. If you're not spending time creating it and then revising it over and over and over again, you're missing out because this is where the real learnings and how to prompt AI properly, come from.

Dan Sanchez [00:20:09]:
It's in those revisions of the prompts, the revisions of the instructions of custom GPTs in order to get it right. So start small and then build your way up. Start to add in different inputs. Start to see if you can get different outputs out of it. And as you do, you'll start to find that you become an AI driven marketer. It'll turn you from a team of 1 into a team of 10 or even a 100 as AI becomes more sophisticated and your ability to leverage it becomes greater and greater.