Everyday AI Podcast – An AI and ChatGPT Podcast - Ep 563: ChatGPT's New Custom GPT's: Advanced techniques to win back time

Episode Date: July 9, 2025

Think you know ChatGPT's custom GPTs? 🤔Probably not. Last week, we tackled the basics and what's new with OpenAI's refreshed GPTs. For this AI Working Wednesdays episode, we're getting into some advanced techniques to hep you win back time. ↳ using the crazy powerful o3 model to your GPT's advantage↳ context stacking↳ custom actions to connect to third party sitesYeah.... don't sleep on this one shorties. Ep 563: ChatGPT's New Custom GPT's: Advanced techniques to win back timeNewsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on this? Join the convo and connect with other AI leaders on LinkedIn.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:Custom GPTs: Advanced Techniques OverviewOpenAI's Context Stacking StrategyO3 Model's Thinking CapabilitiesBuilding Efficient Custom GPTsCustom Actions and API IntegrationZapier Integration for Dynamic DataChatGPT's Context Window ManagementCreating Evergreen Podcast ContentTimestamps:00:00 Custom GPTs: Evolution and Insights03:23 "Mastering GPT Context Stacking"09:31 "Context Stacking in Chat GPT"11:20 GPT Context Switching Advantage15:33 Customizable GPT Usage Explained19:51 Evergreen Episode Update Strategy21:44 Optimizing AI for Continued Learning23:48 "O-Series Models: Advanced AI Capabilities"28:41 Building GPTs for Episode Research30:03 GPT Model Customization and Sharing33:18 Securing API Keys in GPTs36:55 Zapier Enhances GPT Email Capabilities42:12 "Use Chrome Extensions for Tokens"43:48 "AI at Work Wednesdays Survey"Keywords:OpenAI's custom GPTs, advanced techniques, save time, context stacking, o three model, ChatGPT updates, logic and reasoning, plan ahead capabilities, agentic tools, custom actions, third party data, API, building GPTs, leveraging AI, context window, transformer model, generative AI, organization usage, fine tuning performance, productivity enhancement, AI agents, AI tools integration, custom configuration, everyday applications, tech strategies, new rendition, midweek break, AI experts, smarter AI usage, AI-powered planning, AI transformations.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live and Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. OpenAI's custom GPTs have gotten some huge under the hood updates.

And unless you're listening to this show every day or maybe reading our free daily newsletter every day, there's a good chance that you could have missed it. So today, we're actually going on part two of our two-part series, both investigative, what's new inside of these updated and refreshed GPTs. But today, going over some advanced techniques inside there, that can help you win back time. All right, I'm excited for this new rendition of our weekly series, AI at Work on Wednesdays. So this is kind of your midweek break where I encourage you to follow along with me live as I use AI and put AI to work for me in my business. And I hope you can apply it and grow your business too.

All right. Let's get it started. Welcome to Everyday AI. What's going on, y'all? My name's Jordan Wilson and this thing, it's for you. Every Day AI is a daily live stream podcast and free deal in the newsletter, helping everyday people not just learn what's happening in the world of AI, but how we can leverage it to grow our companies and our careers. So if you're a business leader trying to do exactly that, you're in the right place. It starts here with the unedited, unscripted, live stream and podcast, but if you want to be the smartest person in AI at your company, you got to go to your everyday AI.com. Literally like the best free generative AI university on the planet.

You can go learn from hundreds of experts that we've interviewed over the years on our free website and get that daily newsletter. We send it out every single day with the highlights of each episode that we do, as well as keeping you up today with everything else in the world of AI news. So if you want the daily news, make sure to go check out in the newsletter. And before we get started, have to mention really exciting episode coming tomorrow. So make sure that you tune in.

We're going to have Dr. Ben Gorsal, who's literally coined the term AGI. He's one of the OGs in the AI world. So it's going to be an exciting conversation. You don't want to miss it. So make sure that you tune in tomorrow or listen to the podcast. All right. Let's get straight into it. Let's talk about what's new inside of these custom GPTs.

So Open AI originally rolled these out almost two years ago. And I think originally there was maybe more hype than byte with these GPs, mainly, because they couldn't take advantage of all of these new chat GPT updates. And if you are brand new and you're like, what the heck is a GPT anyways, I like to tell people it's a custom version of chat GPD that you can use, not everywhere, but essentially everywhere inside of chat GPT. So it's a way to make chat GPT kind of smaller, smarter, and more specific for your needs and then to be able to reuse it at any time without wasting a bunch of time, right?

Sharing files, going back and forth, trying to get chat GPT to respond exactly how you want it to. You can just create a custom GPT for that. So in today's episode, here's what we're going to go over. We're going to learn about what I call context stacking and doing that with GPTs. It's an advanced technique, but it's actually really simple and it's going to save you a ton of time. We're going to talk about how to build GPTs with O3 in mind, the new O3 model from a chat GPT that can think, reason, logic, plan ahead, and it can agentically use the different tools at its disposal. And then last but not least, we're going to go over how to use custom actions inside of custom GPTs to bring

in third party data via an API. So I'm going to show you the back end, how to do that and give you some different examples. And so even if you're a non-technical person, it doesn't matter, right? It's very simple. It's very easy as long as you understand what you're doing. And that's what this whole show is about. I'm excited for today's show. I don't know, live stream audience.

Are you guys excited? Good morning to Jay and Big Bogeyface there. Kathleen saying, Hello from New York. Hello, Kathleen. Robert, join in here from Raleigh, North Carolina. Come on and raise up. Cecilia in my hometown, Chicago, saying, happy hump day. It's good to see y'all.

Monica joined from Chicago as well. Kyle's super excited. You know, I'm curious, live stream audience, drop your favorite custom GPT that you've ever made in the comments. You know, I'm sure we can all help each other. But also, if you have any specific questions that you want me to tackle when it comes to GPTs and use. these new 03 models. Make sure to get it in the comments now. I'll look. It always helps if you

put like question there because, you know, there's usually a lot of different comments coming in from different platforms and, you know, I'm doing this all unedited, unscripted. So if you do have a question, just type in question, get your question in and I'm going to try to answer them either as I go or at the end. So again, jam-packed episode today, but let's first go over what is actually new. So this is from OpenAI's website. And we're going to be be doing some live demos and everything like that. So podcast audience, I'm going to always do my best to describe what's going on on screen. But we're going to be doing a lot of live demos, looking at some custom configurations and all of those things. So this might be one of those ones. Check your show notes.

You know, we always in our, like I said, in our free daily newsletter, we link back to the website. So you can go and check that out or it's, you know, on our, you know, YouTube channel, everything like that. All right. So here's what's new with GPTs. And this is from. OpenAI's website. So they said creators can now choose the full set of chat GPT models, GPT 40, 03, 04 mini, and more when building custom GPs, making it easier to fine-tune performance for different tasks, industries, and workflows. Creators can also set a recommended model to guide users. So some key details, they say GPTs without custom actions can use the model picker to select from all models if you do use custom actions. So we're going to be showing an example of

you can only use GvT4 or 4.1. And GPTs to build them, you have to be on a paid plan. To use them, you can be on a free plan. All right. So I know a lot of people, you know, if they don't want to spend the $20 a month, but y'all, let's be honest, whether we're talking about Chad GPT or Google Gemini, even clawed as much as I, you know, kind of drag clawed through the mud for if you look at it in the like the wrong way, you hit your rate limits.

Microsoft co-pilot, whatever, the 20, 25, $30 a month, whatever, it is a literal bargain. All right. So, you know, I get it. You might not be able to scourge up $20, you know, every single month, but you can literally do it for a few months, build a bunch of GPTs. And then if you don't want to keep paying, you can still use those GPTs. You just can't modify them anymore if you're on a free plan. All right.

So, and like I said, this is part two. We're going over some more advanced techniques today. But if you do want to go back and listen, last. Last Wednesday, we did part one of this, you know, our new, you know, AI at Work Wednesdays, I think is what we're calling it, which are you guys, are you guys liking these Wednesday shows? So anyways, last Wednesday, go listen to it if you haven't already. Part one is episode 559. So we went over how to automate like meeting action items, how to turn spreadsheets into story driven slides, how to generate custom learning plans, how to generate custom learning plans, how to get instant market research dashboards and how to create real time company financial reports.

I mean, my gosh, I literally built these GBT's for you all. And we demoed them all live. And I did share, you know, I told people, hey, if you repost the episode, I'll share the GVTs with you. So if you did share it, I did, or if you did repost it on LinkedIn, I just sent you guys all the, all the list of the GPT. So if you shared, repost this episode as well, I'll send you a list. I think they were actually pretty good.

You know, not perfect, but great ways to automate your day-to-day work and save some time. All right. Here's the fun part, y'all. We're going live. Hopefully no demo demons, but for sure, there's going to be some demo demons. All right. You know, what can go wrong when trying to, you know, share like 13 different tabs in a browser using generative AI and just really breaking the thing?

All right. So, live stream audience, do me a favor. do me a favor. If you could be my helping hand here, let me know that you can see my screen because I don't want to keep going. And if something bounces off screen, please let me know because otherwise I'm going to keep talking and at least for our live stream audience or for those listening

or watching later on YouTube, you're going to be a little confused or on our website. All right. So let's first talk about context stacking. So if you Google it, you're probably not going to find anything except for your every day. AI. Context stacking is a term that we've kind of, I guess, coined and have taught people as well over the years. So what this is, ChadGBT actually has a huge advantage, and I'm actually very

surprised that the other AI labs haven't caught on, right? Although this may change in the future as Open AI any week, month now, kind of goes to this GPT5 architecture, where the system will kind of choose on its own what model to use right now you can go between different models but what happens usually uh when you're using other platforms google gemini claude etc you can't always flip between different models and keep the context of your chat going so if you are brand new to a i and you're like okay like what the heck does this mean when i'm talking with chat gbt all right um i'm on a on a pro plan So the context window is 128,000 tokens, which is about 96,000 word. If you're on a paid chat chitpt plan, so chat chpt plus, you're working with a 32,000 token context window, which is about 28,000 words.

What that means is when I share a bunch of information, if I copy and paste, you know, 10 pages and chat chad chp t goes through and it responds a bunch of words, that eats into that context window. And it's only going to remember if you're on the chat chp t plus plan, like so many people, are, it's only going to remember the most recent 28,000 words. So if you end up having between you and chat GBT a 56,000 word conversation, it's going to forget the first 28,000. So it only has a running memory of the last 28,000 words of context. That's very important. But you can take that to your advantage.

Because like I said, you can start in these different modes or models in chat GBT and then switch and keep the context. That is huge, especially when you talk about bringing in GBT's to the mix. I mean, you can't even switch models in Claude, right? So if you're using Claude for Sonnet and you're like, oh, man, I'd really love to use Quad for Opus right now and you go to switch and it's like start new chat. So what that means is then it has no idea, no context and you essentially have to start over, right?

And what if that's a chat that you've been using every day? You've gone through our, you know, prime, prop polish process. And you have a lot of good information and you're getting great outputs from chat, gpt or you know from whatever large language model if it doesn't have context stacking or model switching built in you're screwed you got to start over right so that's why this model or context stacking is huge so what that means is you can use different modes different gpts in the same context window and it takes in all of that information so let me give you a quick example all right so uh do we see do we see the screen yeah okay i think

So I saw one person say yes. So thanks, Muhammad. You said yes. So hopefully that means you can see my screen. So what I already did, and I didn't want to make you guys wait 12 minutes as I did this. I know I always do these live. I'm going to do the second part live. So I already ran a deep research query in chat GBT.

Okay. If you don't know what deep research is, come on, y'all. Like it's like whether you're using, I think chat GVT. and Google Gemini have the best deep research, and it's a pretty big drop-off between those and everyone else, right? Grock has deep research, perplexity has deep research. You know, everyone has deep research. But open AIs and Google Geminiis are in a league of their own.

But essentially, it will go through and agentically research something for you in a very impressive manner. So I already had it fine. So my deep research query, I said, find the 20 biggest trending topics from May 2025 to July 2025 in, generative AI, large language models, et cetera. Make sure they're pinpoint specific. Do not generalize in open AI's deep research.

It asks you a couple of questions. I answer those questions, blah, blah, blah. There we go. And then we have this nice, this nice research report. It made 138 different searches. Sheesh. So I'm showing the activity. It did essentially 138 different Googles in order to, you know, I say Google just as a general

way to say search the web. and then it looked at 64 different sources, and then it created a document for me here. So it's great, right? So I'm looking at some of these things. You know, this should give me three, you know, sorry, some of the biggest AI news trends, trending topics over the last, you know, two or three months.

So here's, I want you to, as I'm going through my examples, I want you to think of your use case. That's going to make this so much better when we talk about contact stacking with GPT. What are the things that you're constantly doing? And a lot of times, I would even start, you know, before I even do a deep research query, I would usually start by sharing context and kind of warming the chat up a bit, going through the PPP process. But for demonstration purposes, I'm just like getting it clean, cut, and dirty and fast.

All right. So then what I can do here is context stack with a GBT that I built. All right. So now I have a GBT that I built. that's called Everyday AI Episode Researcher. So I'm saying based on the recent trends, I'm going to say above in your response, please look at my episode data, and I'm going to say in this GPT, in the Everyday AI episode researcher, GPT,

I love Typhon Live, nothing I love more, and plan out at least five episodes, for July 2025, use the entire context of this conversation. So you can hit, that's the big thing about GPDs. You can use them almost anywhere where there's a tax box. Well, so you can't actually use them inside projects, which is a big bummer. But essentially anywhere else, you can just click that at button. And then any GPD that you've made and you've saved, you can then use that GBT. So, okay, why would you do that?

Well, this GBT obviously has custom instructions. And it has. data in there. So I've shared all of my podcast data. I've given it custom instructions on, hey, when I use this, whether I'm using it in a dedicated mode or I'm using it in the context of a different chat, here's exactly what you should do. And I've kind of fine tuned it to do exactly what I want. And it's going to agentically use the O3 model. So by default, I have it using the 03 model. So then I say, now when I click this, you can see it's right away, it's running, It's running a Python script.

So it's doing data analysis on my file that I upload. And it's going to slowly go through here. It's going to start pulling out different ideas. Right. So I actually don't know how long it's going to take. That's the thing about doing it, doing it live here. So, okay, it already gave me my top five everyday AI episode ideas, right? How to actually use chat, GPT, AI agents, everything you need to know.

Will AI take your job, et cetera. So now it's going through there and it's planning out new episodes for me. All right. So I'm not going to spend like 10 more minutes, but this is context stacking in its simplest form. I use deep research, right? And now I at mentioned or used one of my other GPTs. And now because these GPTs can use the O3 model, they're capable of so much more, right? The difference between the GPT4, kind of, you know, GPT 40, quote unquote, transformer,

model in the O-Series model that you can use. I mean, they're agentic and they're absolutely amazing. Okay. Now, let's move on. And this one also took a while, so I just want to show you, you know, some examples of using GPTs with the O3 model and really what that opens up in terms of capabilities. So I have another GPT that I've built. And what I've called this one is the Evergreen episode Hunter.

All right. So I'm actually going to go ahead. All right. I'm going to go ahead and open this in a new tab and just kind of show you what this looks like. Okay. So this Evergreen episode Hunter is actually a little, little complex. Right. So now I've load up my custom configuration instructions where I'm telling it exactly what to do.

Right. So I'm telling it to open up, open up this file with all my podcast stats. And I'm trying to find outliers. Okay. So what that means is sometimes I'll have episodes, right? Most of our daily episodes, they get traction for about a week. And then for the most part, after a week, they just kind of fizzle there and they get, you know, maybe a dozen, you know, a dozen listens or something every day, right?

But for the first seven days, they're going to get usually 10,000 downloads. So I'm trying to find episodes that have evergreen power. For whatever reason, you know, there's a supply and demand paradigm out there in the podcast world. just in the information world where sometimes there's a lot of people searching for something, but there's not a lot of good high quality sources. So I don't really know that unless I'm spending hours every single day looking. That's why I built this Evergreen episode Hunter. So what it does is it goes through and it created a little short little algorithm or formula to find my average seven day, 30 day, 90 day kind of decay. And then it finds outliers in that. And it says,

these are great episodes that are going to stick around for longer than you may think. So this GPT not only does that, but then also when it finds these evergreen episodes. So maybe I had one from, you know, two years ago. And I haven't updated it. So not only will it go and find those episodes, but depending on what I ask of this custom gvety, it will then go agentically do research for me. Okay. So I already ran this one.

I didn't want to have to, you know, make you all wait. So in this instance, I said, give me a list of my. five most evergreen episodes and then an outline on how I could or should update them for new for new episodes based on recent happenings and you'll see I don't even have to say like oh look up information from 2025 these are the topics I care about etc it's got to go through it's got to find and extract that information from the spreadsheet and then I've already given it very specific instructions on how it should be researching so it doesn't pull in information from 2023 or 2024 it's only pulling up the most accurate and up-to-date

information and I've even given it kind of some very specific instructions on how it should agentically browse the web. So then I can go back and look at the chain of thought and I can kind of show you guys what happened here. Right. I didn't want to keep you guys waiting for too long because sometimes these GPTs, especially when you're using the O series model, they can take like just that GBT could take three to five minutes, right? A deep research query might take anywhere from two to 20. I love doing these things live as much as possible. We're still going to do one live here when we talk about custom actions. inside of GPTs.

But what happened with this evergreen episode hunter? Right? So I said, give me a list of the five most evergreen episodes and outline them on how I could or should update them. So first, as I look at the chain of thought, it did for three minutes and 40 seconds.

It's going through, it's writing some code, right? So it's using some Python code. It goes through there. It's familiarizing itself with the spreadsheet that I have in there. All right. A lot of code here, right? This is great that I don't have to do this anymore with my brain, even though I love spreadsheets and I love making little algorithms.

You know, it's funny. I was joking about this with my wife the other day. It's like Prime Day. Like I used to literally create spreadsheets like with like little weighted algorithms on like the best deals. Right. So I love doing this stuff. But it's also I understand it's very time consuming. That's one of those things when we talk about augmented intelligence and we don't always just want to kick off, you know, or kick.

certain tasks to AI, we really want to make sure that we're still becoming better at that task. So that's why I always look at this chain of thought because I'm looking what it's looking at in my spreadsheet and how it's taking that data to go in agentically search the web based on what it's fine, based on what it finds. So, you know, always look at that summarized chain of thought, see what it's seeing because then I'm always going to see what works well when I'm using this custom GPT and what doesn't. And then the next time I use that, I'm going to keep that in mind and I'm probably going to go back and update the GBT as well. So these are not just like, you know, use them once and forget it.

It's use it. Number one, look at the chain of thought to improve yourself. Then also look at the chain of thought to either improve your next prompt or query, but also to improve the custom instructions. All right. So as I keep going, it's identified some of the five episodes with the most evergreen value. I'm going to keep scrolling, keep scrolling. So pretty soon here, there we go.

So eventually it starts searching the web. So it makes one search of the web. It comebacks it thinks. Search the web. Thinks. Search the web. Things, right? Keeps going on and on.

After searching the web thinking five different times, then it comes back and it analyzes more data based on what it found on the web. That's when like when we talk about how powerful these GPTs are now, it's because of this, right? Like the old GPTs, they were, right? Because they were only running GPT4,0, which is not a lot. model that can think, plan ahead, reasonably, reason, and agentically used tools as it decides to, right? So what that means is it found after first, number one, looking at my data and finding these evergreen topics. Number two, researching. After doing that, it decided, wait, I need more data,

right? I need to go run a couple more calculations to make sure that I'm doing the right thing. So it decided on its own, this O3 model, that it kind of had to go. go back and revisit step one, just like a human would, right? The GPT40 model could not do that, right? And that's why I think the GPTs originally not saying they flopped, but they just weren't super useful, right? Because the transformer models, to get the most out of them, you really had to like be a stud at prompt engineering, right? With the O-Series models, not so much because they are so much more capable and they think and reason and go through logic, like a human does. So then it went through after number one, data research, number two, browsing the web,

number three, goes back, does more data, goes and browses the web again, goes back more data. I mean, this is very impressive that this all happened with a GPT. Anyways, when I go back, I also tell it. Okay. Here's another tip and trick. When we talk about context window, all right? It only is going to remember things that are in. in the context window. So later, if I go back into this chat, I'm not going to be able to pull on that summarized chain of thought. So in the custom instructions, I tell it to put into the, its reply, its thought process and key finding. So then that way I can take advantage of that. That's a huge nugget. Sorry that came at the 25 minute

mark, but that's good. A good thing to know that to put into your custom instructions when using the O series models and to take advantage of context stacking. All right. And then I'll go through here. Anyways, eventually it came up with the insights and recommendations. It gave me, okay, these are, these are some pretty good ones. You know, regulation, reality check, explain how the EU AI Act obligations arriving in August 2025.

That's around the corner. And what small and medium businesses must do to prepare now. It's a great episode idea, right? So it went through, it found that some of my episodes that are talking about regulations and laws and legalities, right, that they perform well, even outside of that normal seven-day decay window. So then it went out, looked at all my data, did a bunch of research and said, hey, this is coming up. This is coming up. This is getting ahead. This episode is going to perform well.

So not only that is it did some research in here as well when it went to future episode ideas. and then it gave me all the sources and the headlines that I use as well, so you can have a little bit of confidence in it. That was a lot. Michelle just says, great tip. Keep it coming. All right. We'll keep it going.

Colby, what's up, Kobe? Kobe says, how do PDFs full of text, play and token usage? Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI assistant. now live in the Adobe Firefly app, the all-in-one creative AI studio. Powered by Adobe's creative agent, Firefly AI assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant.

The assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adop.com. great question colby the problem is i don't think anyone aside from open a i knows right i've done

manual testing in this uh the reason being is it uses different techniques to essentially tokenize or to analyze that pdf and sometimes it'll go back multiple times so i had a pretty good idea with the transformer models the o series models when you're uploading pdfs in there it is different each time. So I do not have a definitive answer. I don't know if anyone would aside from people at Open AI, but I would assume that even they would say that it varies, right? Because generative AI is generative, it's going to do something a little bit different each time. I assume it also depends on the specificity of your custom instructions, right? If you are having it read and analyze every single page, right? Or if it's just skimming through, parsing it and pulling out information in the middle,

and it's not actually ingesting all of that information. So I'd hate to use an SEO answer, but I think it depends. All right. So we've covered so far. Context stacking, number one, we've covered some tips with using 03 models when building custom GPs. And then last but not least, we're going to talk about custom actions. All right.

So yes or no, live stream audience, have you built any GPDs custom actions before. I want to know if people understand this or, you know, how in-depth I need to get. So when you create a GBT, most people know this. If you've never created one, it's actually simple because you can literally talk to chat GBT to build it. You don't even have to go in on the back end and tweak the settings. You can just say, hey, you know, my, you know, my website is your everyday AI.com.

You know, I run a, you know, top, top 20 tech podcast. go find out all this information and build me a GBT to do episode research, right? I could build it that way or you can build it by hand on the back end and configure. So if you build it on the front end conversationally with chat GBT, it essentially synthesizes that information and it fills in the configuration instruction. So if you're building a GBT, there's technically two panels and the left panel is split. There's a create tab where you can talk to chat GBT to create a GBT. And then there's a configured tab.

And once you make save it, saves, or sorry, once you make changes, it will render on the right hand side, which is the preview pane. Okay. So when you're configuring something, that's where you can, you know, type in those custom instructions, conversation starters, upload knowledge. The good thing is so you can use on your GPTs, you can decide what model to use. There's a model selector. But you can also set preferred methods.

So maybe if you do want to build these for your team, that's something I should have started with at the beginning, right? For enterprise teams, people on chat, GPT teams, et cetera, you know, this is great that you can set a recommended model. Sometimes you might not need 03 pro, right, if it's more of a simple, you know, hey, rewriting content, right? Because if you're using like 03 pro, depending on what your query is, you might be waiting 10 minutes for a response when you might only need to wait 10 seconds. So having a recommended model that the creator can set is great, but also the users, whether you are using the GPDs that you create, whether you're sharing them with a small team, whether you're sharing them with, you know, we just trained a international

organization that's going to have, I think, 500 enterprise license, 500 enterprise seats. We've trained companies with thousands of enterprise seats, right? So you can then share these GPTs across your organization. So that's good as well. But let's look at the actions. Okay, this is getting a little bit technical, but hey, you all said you wanted the advanced episode. So when I click create actions, so remember, when you do this, unfortunately, at least right now, and this could change, if you're using custom actions, bringing in information from an API, then you are only able to use the GPT40 or the GPT 4.1 model. So there's some tradeoff. So essentially, if you're going out and fetching dynamic data via an

API, then you can't use the thinking models. Hopefully that changes in the future, but maybe there's a technical reason that you can't do that. All right. So I'm going to click Create New Action. So this might seem confusing, all right? Because what you're looking at here is something with schema actions, right? There's, there's JSON, YAML, right? And you're like, what the heck is this?

All right? It's not as hard. you would think. All right. And the reason being is chat GPT has an actions GPT. So when you, a lot of people don't know this. When you go into actions, right, there's a get help from an actions GPT. So I could just say as an example, right, you know, I want to create a GPT that pulls in information from my, or that not pulls in information that, right to my Google calendar. How do I do this?

All right. So this Actions GPT is created by OpenAI. So then it's going to give me a step by step overview. And in most cases, it will eventually give you the code that you need, you know, essentially this schema to put in to this schema bar. So, you know, if you're looking at the word schema and you're like, wait, what the heck is this? Don't worry. It's not as difficult as you may.

think. You can literally just use the actions GPT. It'll give you a step by step guide. You know, if you're using, I don't know, a very, you know, popular API. I don't know. I'm trying to think. Like, I don't know. If you're in digital marketing, maybe you're using, you know, Semrush or H-Refs, right? Like these like SEO content platforms, you know, you can literally just grab their API documentation, paste it in and say, I'm trying to create a GPT that will pull my website's, you know, keyword rankings instantly, right? And then And here's all the API information and then the actions GPT will essentially write that schema for you. And then all you have to do is copy and paste it.

All right. But I'm going to give you an even easier way. So yes, if you are an organization, you probably have someone technical on your team who knows how to work with API endpoints. Right. But all that means is your GPTs can essentially via these actions. And, you know, so if I go to. authentication, right? Any, not any API, but essentially any API, depending on how it works, there's some intricacies that I don't necessarily want to get into because it would take

hours. But essentially any API out there, if it has the correct endpoints, if the way it handles the API key is correct and you can keep it secure because you never want to hard code an API key inside the custom instructions for AGVT, because again, If you make this public, someone could, in theory, extract your API information. And, you know, if you're paying for something, they could run up the bill. Right. So you never want to hard code the API in here. So there's different ways that you can authenticate the API in here.

So also with O-off as well. So like your log-in information. Anyways, so you can do that on the more technical side. If you understand API endpoints, even if you don't, you can use chat, GBT's actions, GBT, but I'll tell you an even easier one and do a demo of it here. real quick. Zapier. Zapier, y'all. All right. So I will have this kind of blog post or page in our newsletter today and give you an example of how this works. So Zapier is great. If you don't know Zapier, it essentially connects to everything on the web. So maybe you don't know or understand

APIs or maybe you don't have access to it at the organizational level, but you have access to it at your level. That's really all you need. So it's a little bit easier if you have a paid Zapier account, even if you have a free Zapier account, you can do the basics and it's actually really impressive. So all you literally have to do and is this page gives you a step by step. So it says, copy this link to your clipboard. Okay, I'm going to copy it. I'm going to go back into, let me do a new one here. There we go.

So there's an option to import schema from URL. So with Zapier, it's the same URL. I click that. I paste. I click import and bam, it. loads up the schema, right? There's some placeholders in here that you obviously have to update, right? So the next part of that Zapier kind of blog posts is it gives you these instruction templates. And then these are the rules then that you would copy and paste into your custom

configuration instruction. So as an example, I'm going to go ahead. I'm going to copy this, right? And then I would put this. I would click the back, button here. Okay. So I essentially imported this schema from the URL, which is a great feature from this GPT builder. I'm going to go back and then I paste that other thing in custom instructions. And then from here, literally the only thing I have to do is I have to update what I actually want it to grab and I would put that in the custom instructions. And then the other thing is I just have to go to Zapier's actions and then just enable them. Right. So, So as an example, it's just actions.zapier.com slash GPT slash actions.

Again, I'll have it in the newsletter. And then you just build them visually. So you don't even have to be technical. So I did this, you know, different things. I can send an email, right? Which is great because in GPTs right now with the new connectors, if you're using deep research, you can read emails. You can read your calendar, right?

But you can't write. And there's sometimes I'm like, man, I wish, like, while I was in, in here, I could just add something quickly to my calendar, or I could, you know, use deep research and go through my g-mails, but then actually reply to them. Well, you can do that going through this process with Zapier. So then you just have to add the new action. And then essentially, this reply to email, that would be when I go back to the custom instructions, there's two or three places that you would need to put that action in the custom instructions. And then that's it. So literally, you import via the URL.

It loads the schema up by itself. You go to the Zapier page. You get their template. You update it based on the action that you build. And bam, that's it. All right. So let me show you an actual use case example. Sorry, live stream audience.

I know I'm jumping all over. And if you do have a question, we're wrapping up here. So get it in now. If you have any questions, I'll try to answer them if there are any at the end. So now I went through this exact same, this exact same process. I just didn't want to have to show you guys like my email, my credentials, you know, trying to be somewhat safe, even though I just give you guys all my data anyways. So I literally just went through the process that I just walked you through via Zapier,

importing the schema via the URL using the custom instructions from Zapier's website, going in, configuring the action, which is literally no code. You just click a button and then updating the action name in the custom instructions. that chat gpd gives you it takes like literally two minutes all right so now i'm just going to say something simple i'm in the gpte it's called add to calendar i don't need to use the four or the o three model because it actually won't work so i have the four oh model and all i'm saying is add a 30 minute meeting with jake this saturday at 5 p.m i'm going to say to my calendar and i'm going to say no location or description because if i don't say that it's going to come back and

ask me and I just want to show you guys literally one prompt. It's going to add it to my calendar. And just so you all see here, I'm zoomed in to my calendar for this Saturday. It's blank. Nothing's on there. All right. So hopefully this will work. But demo demons, right? It may ask me to reauthorize once. Oh, great. I literally just did this before I started. Give me a second, y'all. I just got to re-sign in, even though I literally already did this. Okay. All right, we're back. So all I did, I said, add a 30 minute meeting with Jake this Saturday at 5 p.m. to my calendar, no location or description.

And you'll see here, I don't know why it makes you reconfirm. I think in the new context window, you have to do it once. And I started this fresh. So you'll see it just says talked to actions.zapier.com. And here in a second, it should be done. There you go. It says your 30 minute meeting with Jake has been successfully added to your calendar for this Saturday at 5 p.m. And then it says view on Google Calendar. I'll just show you over here.

It's there. There's my 30 minute meeting. So again, you might be saying like, all right, Jordan, like, what's the point? Again, one of the downsides with chat GPT in general is it doesn't look or work with your live dynamic data, which is okay. That's fine. Right. They're improving that with connectors. They have some dynamic connectors as well. But now, you can't, right? Zapier, you know, it's the simplest way to do it. You know, it connects to, I think, more than 7,000 different services. So even if you're a non-technical person, you don't have to understand API endpoints and be writing schema from scratch. I showed you how with just a little copy and pasting, using Zapier, I think is a great way to do it. And then even if you get caught up,

I mean, Open AI has that actions GPT. Just paste in the information. from whatever API you're trying to use, or you can even say, hey, I'm using this on Zapier. Here's the documentation. Here's what I'm trying to do. It's giving me this error and you can probably fix it in like, you know, a minute. But it's so simple. So I hope this was helpful, y'all.

I'm going to get as we wrap here. Let's go, let's go to a couple questions. I said I would. Jay, what's up, Jay? Jay says, assume could add it to one of your specific Google calendars in your account. Yes. So that's one of the steps when you're setting up the actions inside of Zapier. So first, you have to authenticate your account if you've never done it inside Zapier.

Most people, if you have a Zapier account, you have. Then you can actually, if you have, you know, five different calendars inside your Google calendar or your outlook. Yes, you can choose which calendar goes to. Great question there. Let's see. Anything else? Wouter. Sorry if I got your name correct.

Great question. When do you know when the context window is full? Well, you don't. I wish front-end AI chatbots would include this. The only one that I know it's not front-end is if you're using Google's AI Studio, which is more for developers, it tells you how many tokens you've used. There's plenty of Chrome extensions or whatever browser that you're using that will tell you how many tokens you've hit.

But if you're not using those, which you always should be, and if you listen to the show, we've talked about that a lot and we've shared in the newsletter a lot, you'll otherwise know when it starts to forget stuff, right? So that's why you should always just use a Chrome extension with a token counter. It's way simpler. Let's see. Do we get it? Okay.

Let me just confirm this one here. And then I think we can wrap. So Kathleen says, so when you switch the model in the same chat, the context window resets. So no. It stays the same. Okay. but it retains the information in that context window. So when I switch models,

it's not like, oh, I'm back at zero, right? If I had 26,000 of my 28,000, you know, kind of quote unquote, you know, context window consumed, it's not like when I switch the model, it goes to zero. No. And you wouldn't want it to because you want it to retain that information. So it keeps the same context window. It just allows you to swap model.

in there. So great clarifying question. All right. Y'all, was this helpful? Let me know. Like I said, if you want access to all those other GPs that I created last week and they're actually pretty good, just repost this episode. I'll make sure to share those with you. Also, let me know whether it's in the comments here or in the newsletter. Number one, are you liking this AI at work Wednesdays and what should we tackle next? There's so many different things. One of the reasons why I started this segment is people always ask me, Jordan, like, oh, how are you using AI? And I'm like, okay, well, every day. Right.

So on Wednesdays, I try to either, whether it's something new, it's something that people are confused about, it's something, oh, so many people are asking about blank, right? So I try to bring you into my world on how I'm using all of these different AI tools. But if you have specific requests, ask, y'all, I work for you. So in the podcast, show notes, I always leave my. contact information. It might take me, you know, a day or three or a week. If I miss your message, sorry, I stink at that, but I eventually get back to everyone. It just might take a while. But give me your suggestions. I want to hear and I want to make this new weekly Wednesday

AI at Work segment very valuable for you. So I hope this was. So as a reminder, make sure to tune in tomorrow. It's going to be a banger episode having one of the original OGs of AI, the actual Dr. Ben Gertzel himself, who coined the term AGI. I hope this is helpful. Please go to your everyday AI.com. Sign up for the free daily newsletter. And I'll see you back tomorrow and every day for more everyday AI. Thanks y'all.

Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one, conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly. Adobe.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us.

If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.