Alane Boyd (00:03)
On this episode, we are going to explore how to use Custom GPTs as a huge shortcut for consistent output in your own voice.
Micah Johnson (00:12)
So, we see this all the time. And I think we were probably guilty of this for a while too, especially when AI really started becoming mainstream. And that is you go and chat with it and then you feed it a bunch of files, you give it a bunch of context, you write this whole long prompt out and then you get the output. But you keep doing that. Like every single time you need it, you give it those files, you give it that context, you rewrite your prompt. The problem is...
You might forget a file. You change your prompt a little bit. You have all these inconsistencies, and then that creates differences in the you get frustrated. like, no, I mean like this, or use my voice and all this stuff. And the difference is you're using this as like a one-off versus how do you build a system. So let's talk about how to build a system.
Alane Boyd (01:05)
Yeah, and there's a misconception here too on how AI is remembering everything. Like we think it has this huge memory that we've told it everything, that it knows who we are. And it really is not like that. There's limits to the context that it can have for you. And sure, you can call back a previous prompt, you can go back to one of your chats and it can search for it and find stuff, but it is not an all encompassing thing of everything you've ever said to it.
So the Custom GPT route is to feed it specific information that it's always gonna go back to. So it really helps with voice, tone, how you speak, how you want the content to come out, everything that is specific around that one output.
Micah Johnson (01:49)
Yeah, even who your audience is that it should be helping you generate ideas for.
Alane Boyd (01:54)
And, you know, Micah, I think we should probably clarify too that when we say Custom GPTs we're also talking about Gemini Gems or Claude Projects. It's just, we don't want to have to say every single one of those, every single.
Micah Johnson (02:04)
Yeah, this
would be a tedious episode if we had to say ChatGPT or Gemini Gems or Claude Project every single time we are referencing it. So for future note, all of those will work. We're just going to say Custom GPTs.
Alane Boyd (02:19)
And we're using them for ourselves individually and for our company every day. And I love hearing how the clients have started using it too. And they're creating Custom GPTs for their clients. So how, they're creating content for that client, they fed it with their voice and things like that so that it's helping generate the content that they need to, which I think is a great idea.
Micah Johnson (02:43)
Yeah, and if you think about, you know, the different aspects of business. So ultimately we've got a few Custom GPTs where we feed it a topic and it already has all the context and how it should output stuff for us. And we use that as a brainstorming feature. So how would we take this topic like we've done it before? Even for these podcasts, we're now feeding topic ideas.
And then you and I work together manually from some of those results. But we've designed it so that those results are in the exact format, in the exact voice to our audience, y'all that are listening, be able to shape all this so that we have more productive episodes.
Alane Boyd (03:13)
Mm-hmm.
process is so much faster for us now. Like, sure, you can go and use ChatGPT, just chat prompting, without the custom one, but you are doing a lot more work to get to output that you're looking for. Our process got so much faster when we fed it what we were looking for, the output that we were looking for, even the structure, Micah, of how we want the flow to go. And I think that's also when you're having a Custom GPT that you can get that custom
output, how do you want the structure to be? Do you want bullets? Do you want headings? Like you know, those kinds of things can be built into the output
Micah Johnson (04:03)
Yeah. If we think about it, when AI, and we've talked about this with like our RAG episodes and different things like that, when AI doesn't have context, it's still going to give you an answer, but it's going to be from the training data. And it's going to be pretty much the same as the generic answers that everybody else is getting when they type in, "Hey, write me a LinkedIn post about X" or whatever it is. I guess that gets confusing because X is now Twitter. So we can't say.
Write me a LinkedIn post about X because then it's anyway, y'all get what I'm saying, I think.
Alane Boyd (04:38)
That could get confusing and I'm glad that you clarified that.
Micah Johnson (04:41)
Yes, we have a lot of clarifications in this episode, so keep listening.
Alane Boyd (04:44)
Yeah.
So we have kind of a basic framework that we think of when we're building a Custom GPT. One of them is, you know, who is speaking? What's the brand voice, personality principles? My personality is different from Micah's. So I don't want the same voice or output to be for both of us. I want it to be specific to how I say things versus you or what we talk about as a company. What's our company voice?
Micah Johnson (05:10)
Yeah. And for some more clarification, we are not reading from a script, if anybody can actually tell. There is no scripts involved. What we do is bullet point all of this. As we're going through this and as we talk about ideas and what we want to cover in an episode, we're leveraging bullet points and concepts and just the general flow. So that's why some episodes make no sense and some episodes flow really well.
We define that step, like Alane says, that's step one. We're gonna define our voices. We're gonna define what we're looking for so we only have to define that once we don't have to every time we sit down for brainstorming episodes think of "All right. How are we gonna tell chat gpt how we want this again?"
Alane Boyd (05:55)
Yeah, yeah. And the other part of this is who is it speaking to? Who's the audience that's going to be reading it? If it was a content post on LinkedIn, it's not enough to say I want a content post on LinkedIn. Who do you want to read it? Is your audience entrepreneurs? Is your audience healthcare facilities? You know, it needs to know, just like in regular ChatGPT, that
Micah Johnson (06:01)
Mm-hmm.
Alane Boyd (06:21)
it needs to know the audience of "What is this output? Who is it gonna be for? Where is it going?"
Micah Johnson (06:26)
Yeah, and a quick pro tip here is if you give it resources, a good example would be examples of LinkedIn posts that were your best LinkedIn posts. Give it that and ask it to replicate that style. That can be your style that worked really well. And that way, it's not off on these tangents. It's all focused on specifically what is working. It's almost like.
It's learning as it goes.
Alane Boyd (06:55)
Mm-hmm. And one of the things that I like too about this is like I can feed it with some of my content that I've written because I absolutely want it to sound like me, be like me for the output. But I also might want there might be a couple of posts that I like that I've seen before from other people that I can use pieces of to say, hey, this is my style and I'd like to incorporate a little bit of this. I want to be a little more edgy or a little more funny. And here's an example of that.
Micah Johnson (07:24)
Yeah, and what you're talking about specifically is tone and approach and maybe even narrative flow, not use these exact words and plagiarize somebody else's crap. It is, you know, the concepts around.
Alane Boyd (07:27)
Mm-hmm.
Absolutely, because I still want to sound like me. I don't want to sound like somebody else, but maybe I want to just adjust my tone a little bit. I like being edgy.
I'm not, so I gotta practice. I need help, I need some coaching.
Micah Johnson (07:48)
Okay, all right, that's fair. The
thought of being edgy. All right, so that's step one. Fill it full of that. Step two is what, Alane?
Alane Boyd (07:58)
Well, you need to think about what kind of do's and don'ts or constraints that you want it to have. Because if you leave it wide open on just what you want it to do, but you don't tell it any of the don'ts, then it can still get a little broad in it's output.
Micah Johnson (08:11)
Yeah. So an example would be if you're creating SOPs, tell it you want it to make an SOP, right?
Alane Boyd (08:18)
Or
Or, even a framework. For us, sometimes it's a structure where we want, okay, what's an idea of a hook? What's an idea for the story that we wanna tell? We've got a flow to how we want things to happen. So if we tell it that framework, then it knows, hey, every time it has an output, then we want it in framework.
Micah Johnson (08:36)
Yeah. And again, we're only defining all of this once. And then we're going to chat with this whole definition to get the outputs and to get what we're looking for. And that kind of leads us to step three, which is actually defining the output. Like, I want this topic, in this way. And I can just go back to that Custom GPT and
I don't even have to think about my prompt. My prompt can literally be topic, like that's it.
Alane Boyd (09:04)
You don't have to
write anything else. just feed it the topic and it's going to spit out everything else.
Micah Johnson (09:09)
Yep. Yep. And, you know, there's some cool tools in Gemini and ChatGPT as well where you can say, "hey, put this in a canvas." And now when you're brainstorming, like, heck, you and I did this recently, Alane, we're brainstorming on something. We've got ideas that were generated in our voice using the Custom GPT. We put it in a canvas. And when we get fresh ideas from just you and I dialoguing, then
Alane Boyd (09:35)
Mm-hmm.
Micah Johnson (09:36)
We go back and say, hey, let's add a section about this. Let's add a section about this. And instead of us having to manually type it all out and take the time, you and I flow about the ideas. And then we're very rapid. Like, while the AI is editing the output, we're already on the next idea. And when it's done, we are like, OK, well, let's add this topic or let's change the hook to this. We're not manually in there typing all of this. We're coaching it. And then at the output,
Our final process always is take the output. It is not verbatim from the AI. We're then refining and shaping it like humans, to say, what do we want to do? Where do we want to cut? How do we condense this down? And just get down to the bullet points and the topics that we want to cover.
Alane Boyd (10:22)
Building these first two are the perfect foundational step to teaching an AI agent what you want it to do. So it's great if you're thinking or if you haven't really been dabbling in agents or even if you have, starting off with a Custom GPT and getting this worked out is going to make the agent part, not the actual connections of everything, but the instructions for the agent flow faster because you've been working this out.
Micah Johnson (10:30)
Yes.
Alane Boyd (10:51)
with the AI already.
Micah Johnson (10:53)
Yeah, so what you're talking about is if we can perfect it in, say a Custom GPT, how do we operationalize that in a automated workflow? And we're already going to know the context, the prompt, everything else to give it so that we can literally take that, put it in an automated workflow, have a trigger, and have the output consistent every single time because we worked on it from the Custom GPT side.
Alane Boyd (11:03)
Mm-hmm.
That's exactly what I'm saying. And you said it so well, but it really is, you know, if you're trying to take some baby steps, when you go to build the agent, if you're going straight to build part and you haven't kind of worked some of this out, it can be challenging because you're trying to build the agent, the whole workflow automation, along with what do I actually want the agent to do and the output that I want it to do and have the right instructions. So this can save you some of that heartburn.
by doing this first and really getting to an output that you're like, man, I nailed it, or the AI nailed it, I guess, in this case.
Micah Johnson (11:54)
And I mean, the heartburn does kind of come in that testing phase. So by testing it in a Custom GPT with the prompts and the context that you want to feed it to, and in different scenarios, different topics, you can iterate so much faster.
Alane Boyd (11:58)
Mm-hmm.
Absolutely.
Micah Johnson (12:09)
If you're constantly redoing everything with your chats with AI, the whole point of this episode is try a Gemini Gem, Claude Project, or Custom GPT to load that stuff in once and reuse that over and over again to save yourself hours.
