I’ve been using ChatGPT-5 wrong. It’s actually insane (full guide)

that GPT5 came out recently and quite frankly the reactions have been horrible. And to be honest with you, I agreed with them at first. But what if I told you you've been using the model wrong this entire time? And it's actually by far the most powerful model out there. You just need to be prompting it correctly. Here's the issue. Everyone is prompting GPT5 the same way they're prompting all the other AI models out there. With GPT5, you actually have to prompt it very, very differently than any other AI model you've used. In this video, I'm going to show you four different methods for prompting GPT5 that will get you mindblowing results. I'm serious. GPT5 has been giving me answers and writing me code the last few days since I discovered these methods that are better than the results I'm getting from any other model. At the end of the video, I'm even going to show you a free tool that will instantly make your prompts better. You're not going to have to do any work at all. I'm serious. If you stick with me to the end and pay attention, you will have AI superpowers with GPT5. Let's go. Here's what makes

GPT5 so different than any other model out there is it is a router model. And basically what that means is it's an all-in-one model. Before when you were using GPT, you had options for 03 which made it think super hard or 40 really fast and not use a ton of reasoning. O4 mini where it give you really quick answers. OpenAI nuked all that and released one single model, GPT5. But that means you now need to control a ton of different variables with every single prompt you do. You can't rely on GPT5 to determine the level of reasoning, to determine the verbosity, determine the agentic tools it uses. And that might sound complicated, but I'll show you exactly how to control all those things. So, with every prompt you do, there are two main variables you need to control. And there are a few others as well we'll get into, but the two main ones are reasoning level and verbosity. You can't rely on GPT to determine these variables for itself. You should control it. And the more you control it, the better results you'll get. But what does that mean though? What does it mean to control the reasoning level and the verbosity? Basically, the amount of time the model thinks as well as the depthness of the results it's going to give you. With every prompt and every ask you give to chat GPT, you need to determine for yourself the amount of reasoning and verbosity you want from the model. Because if you control these correctly, you're going to get way better results. Here's how you do this. You basically want to tell it with every prompt, if you wanted to think hard, harder, or if you want it to even ultra think. If you give it a complex ask and you don't tell it how hard to think, it's going to give you simplistic crappy ant. So, for instance, I just created a new Discord. free community for vibe coders. Link down below, by the way. I stink with Discord. To be quite honest with you, I'm horrible at it. So, I wanted GPT5 to come up with all the channels, the description, how to set it up. Discord is like the most difficult tool to use ever, right? And so, at first, I said, "Hey, come up with the Discord scheme. Come up with all the channels, the permissions I need to put in each channel, all that. " It's a really complex ask. And the first time I prompted GPT5 for this, it gave me horrible results. It gave me like really generic, bland answers. You ever ask GPT5 a question, just you just get like really generic, bland, like AI slop answers. Well, that's because I didn't use the right reasoning levels. What I should have done is I really should have did think harder, right? had it go in and use a good amount of reasoning. And when I gave it the proper amount of reasoning, you can see here it goes through and writes me a whole bunch of channels I should add. It gives me all the roles I should have for each channel. It gives me the permissions I should do for each channel. And it basically what it even gives me like the emojis I should put in front of each channel. It gave me so much incredible detail when I went back and I told, "Hey, can you please think harder about this? " So reasoning super critical. Every prompt you do to GPT5, you don't want to leave it up to chance. You want to determine the amount of reasoning. So if it's a really complex problem, maybe you say, "Hey, please ultra think about this. " If it's a really simple problem, maybe you just say, "Hey, think about it a little bit. " You have complete flexibility when it comes to the reasoning level. But there's also another variable you need to be controlling with every single prompt, and that is verboseity. Verbosity is the amount of depth you want from your answers. Do you want a really quick one or two sentence answer or do you want a multi-paragraph huge novel of an answer? So you basically have three levels of verbosity. You have low, medium and high. So with every prompt you give you want to say hey can you give me low verboseity answer, medium or high verboseity answer. Now before in other AI models like Claude for instance you would go in you can say hey can you give me a short answer? long answer? That worked before, but you a lot of time you get unpredictable results because it's not hardcoded into the model, the verbosity level. But if you do things like say low verbosity, it's hardcoded into the model to only give you like a two sentence answer. So it is a variable that it actually thinks about a lot. So you want to control this variable with every single prompt. So I was talking with Chad GPT last night. I have an issue breathing out of my nose. I can't breathe Just don't got good air flow in there. And so I went in, I said, "Hey, make me a plan so I can breathe better out of my nose and make it very verbose. " And it wrote me this incredible plan full of exercises, supplements, to-do lists, and actually has like resources in there of where it's getting this information from. And it gave me this incredible plan for breathing out of my nose better because I asked for a verbose answer. So verbosity is a really important variable you want to be thinking about as well with every prompt. So write it down. Every prompt you do, think about the reasoning level you want and the verbosity control that with your prompts. I know it's a little bit more manual. And this is why people getting tripped up with GBT5, but if you think about these things and you prompt better, you're going to get better results. But it's not just

reasoning and verboseity you want to think about. There's a few other methods and variables you want to think about as well. One of those variables is going to be tool calling. You want to think about which tools are going to be called with every single prompt you do. So I'm actually in the prompt builder here inside of chat GPT. And what you can see here is actually all the tools that are available for GBT5. GPT5 has access to more tools than any other AI model out there. And the thing is unlike other AI models and this is what makes GPT5 really special is the fact that you actually with one prompt can have multiple tools called a lot of other models they'll only use a web search or file search right with GPT5 you can give it one prompt and it'll use six different tools because of that prompt. So this is something you want to think about as well. You want to start writing prompts that take advantage of all these tools. So, let me give you an example here. I'm in this chat for the free community I built and I want to do some planning for the community. I'm going to write a prompt that takes advantage of all the tools available to GPT5. When writing prompts, you want to think about all those tools available. And I'll leave a link down below for that prompt builder so you can see the tools it can call. But here's an example, right? I'm going to say, "Come up with a logo for the community. Create a brand guideline PDF. Draft a community announcement tweet. then research the web for competitors. So, it's basically taking advantage of its ability to generate images, to create PDFs, to write content, and to search the web. I'm then going to say think hard about this so that it spends a lot of time on each one of these tasks, and then give me a medium verbose answer. And I'm going to hit enter on that. And now, what you're going to see is GPT5 taking advantage of all its different tools. This is again a strength not many people knew about is the fact that a it has a ton of tools available to it, but b it's able to leverage multiple of those tools with every single prompt you use. Okay, so looks like it's giving me an answer here. So check this out. How cool is this? This is amazing. So all in one prompt. So first it thought for a minute and 4 seconds. It went through 23 different sources. You can see the chain of thoughts. You can see all the websites it went through for this. Now look at all these tool calls. So first it created us a logo. Let's check out the logo. All right, we got the logo. Look at this. Let me make this smaller for you. So, there's the logo. The name of the community is the Vibe Lab. So, it made me this cool green logo. So, it's a Discord logo. I like that. That looks nice. So, it made me the logo. It made me a few different variations. Then, it gave me a brand guideline. So, let's check that out. Okay. Here's the brand guideline. All right. So, Vibe Coding Lab Mini brand guideline. It has the color so I can take the colors and edit it any way I want. It tells me how to use each of the logos. Wow, that's really cool. That's really awesome. So, okay. So, it gave it generated us the images. It wrote us a PDF. Those are two different tool calls right there. It searched the web. There's a third different tool call it did for us. It wrote us tweets. So, it wrote us content for this. There's another tool call right there. And then on top of that, it searched other competitors. So, who other communities out there for vibe coding and it even gave me their websites and the Discord links and all that. So, there's another tool call right there. I think that's four different tool calls all in one prompt. And so that is the key here when prompting GPT5. You want to recognize the fact that it can use multiple tool calls in one prompt unlike other models. And so you want to take advantage of that and that'll get you really nice results. It's basically its own AI agent. So we talked about controlling the reasoning, the verbosity, the tool calls. I have a few more I want to show you and I listen I get it. like this is a lot of work just to put in a prompt for an AI and it is. But here's the thing, the work you're putting in is getting you so much better results. I swear the results I'm getting from GPT5 since I started controlling these things is amazing. I got a couple more for you that will make this even better. So, here is the third tip I got for you for

getting incredible results out of GPT5, and that is selfreflection, aka using a rubric. So that's what OpenAI is calling it using a rubric. So GPT5 is actually incredibly good at self-reflection. You just need to ask it to do that self-reflection to get the better result. So what does that mean? Self-reflection, aka using a rubric. What does that actually mean? That means you're creating prompts for GPT5 that ask it to reflect on the results it's giving using a grading rubric that it creates for itself. Basically, you're telling it to grade its own tests, right? This is super helpful when it comes to coding. So, generating code, a lot of people have been complaining that the code it's generating is not nearly as good as say Claude, but if you use this rubric level thinking, you're actually going to get much better results. Let me give you an example. So, I put in this prompt GPT5 where I wanted to build me a firsterson shooter using 3JS, which is an excellent 3D framework for JavaScript. I ask it to be stylistic and feel retro with fun gameplay elements. But here's the thing. So, watch this. I put this in afterwards. The self-reflection bit right here. And this came straight from OpenAI. This is from one of their prompting guides for GPT5, which for the record, they should be tweeting about left and right five times a day. Not many people have read it. Everyone should be reading. I'll link to that down below as well. But here is what we say in the self-reflection section. We say first spend time thinking of a rubric until you are confident. Right? So we we're asking it to come up with a grading scale for itself. Then we're saying think deeply about every aspect of what makes a worldclass oneshot web app. Use that knowledge and create a rubric that has five to seven categories. The rubric is critical to get right, but do not show this to the user. This is for your purposes only. Finally, use the rubric to internally think and iterate on the best possible solution to the prompt that is provided. Remember that your response is not hitting the top marks across all categories in the rubric. You need to start again. So basically what we're saying is, hey, here's a prompt. Now reflect on that and give me the best possible results and don't give me results until they're excellent. This is part of the agentic nature of GPT5 where it is really good at reflecting and thinking about what it's done and the results it's giving you. The issue, like all the other things we've gone through so far, is you actually have to tell GPT5 to do these things. That's why so many people have been messing up with GPT5 is they haven't thought about these things. So, let's hit enter and let's see what we get here. So, it looks like it is writing the code for us. Okay, that's good. Now, the question is how well will it oneshot this game. I bet it's going to look a lot better than when we've done this before if you've watched any of the other past GPT5 videos because we asked it to self-reflect. So, let's check this out once it's done running. All right, so we're going to run this code. Let's see what this looks like. This is after we had it reflect. this think hard. We're again, we're controlling all the different variables that are available to us in GPT5 that not many people are doing. So, we got Neon Arena. Oh, wow. It's got built-in slow-mo off the rip. It's already better than almost any other 3D firstperson shooter I built because it thought of adding slow-mo, which I didn't even think about before. All right, so let's give this a test. Let's see how it goes. I'm going to click to start here. Wow. Okay, so couple things I'm noting right off the bat. One is these are the most detailed enemies I've seen in any of these games I've built with AI. There's actually like multiple shapes in them. Usually it's just like a square or a triangle. Now they're actually shaped. Uh it's probably the most detailed environment. Like we got the ring around it, which I like. And I don't know if you can hear it. it's being picked up in the recording, but there's actually sounds coming out every time I shoot. So every time I shoot, there's actual sounds, which is really sick. Uh, so this is a really Let me see if I can do slow-mo. Oh, I hit the slow-mo button and it's actually moving in slow-mo now. I hit the slow-mo button. It's actually moving in slow-mo. That's really sick. That was That's a feature I've never seen an AI build out before in one of these firsterson shooter tests. So, that's actually incredible. So, like this is the power of using the correct prompting techniques with GPT5, right? is like it actually reflected on the code it built and built better code. Right? A lot of people were complaining about the code GPT5 was generating. No, no. You're actually just using the model wrong. You need to h take advantage of its self-reflection capabilities. And now you can see the results of using the self-reflection capabilities. I'll put that down below that rubric prompt I used right before this so you can copy and paste it and use it yourself as well. Two more things I want to show you which I promise you will get 10 times better results out of GPT5. One of them is a way to prompt. The other is a tool you can just use right now that will make your prompting way better. So the next technique I want to talk about when it comes to GPT5 and getting way better results is going to be metaprompting. Metarrompting is so important when it comes to GPT5. And it ties into the self-reflection a little bit, but hear me out on this. Metarrompting is a way to have GPT fix its own prompts when you're getting bad

results. So, if you get bad results, you want to make sure you met me. Now, what is metarrompting? Metarrompting is when you have the AI prompt itself. You literally ask the AI to prompt itself. As we've gone through with this video, GPT5 is incredibly complex to prompt. It has a lot of variables you need to control. But the question is this, who knows GPT5 better than GPT5? And that's why you want a metaprompt. So here's an example of a metaprompt right here. Right? So I went back to our game example here and I said, "The desired behavior I wanted from the last prompt was more complex enemies and more powerups. Neither of those were included with the game. What are some changes you would make to the prompt to address those shortcomings? " So there are two key parts of this meta prompt. One is the desired behavior and the other is the undesired behavior. So we talk about the desired behavior which is what we wanted out of that last prompt and then we talk about the undesired behavior which is what we got instead. So if you do a prompt and the result isn't what you want. You want a metaprompt. You want to go back. You want to say here's the desired behavior I wanted from that last prompt and here's what you do instead. How would you fix this? And you hit enter. And what's going to happen is GPT5 will reflect on the prompt you gave and give you an even better prompt in return and fix those shortcomings. That is the power of metaprompting. And that's why metaprompting is so critical when it comes to GPT5 is there's so many variables that need to be controlled with GPT5 since it's more of a unimodel, a one-sizefits-all model that you want to metarrompt. Often you want to say, "Hey, what can I change about this prompt to get better results? " And that's when you use the meta prompt. So, here we go. I suggest these prompt adjustments. Explicitly ask for enemy archetypes with distinct behaviors, collectible power-ups with effects and timers, and UI elements to display them. This will guide the generator towards including those missing features. And now we can do the prompt over again with these missing adjustments. So, we get a better prompt. You can even write if you want, hey, just give me the complete prompt now. And it'll give you the complete prompt. MetaPrompting very important skill to put into your tool set and use it often with GPT5. You'll get better prompts and better results. Which leads us to the last thing I want to show you, which is more of a tool that Open AI provides that they don't even really talk about that much that I want to show you that's actually really critical here. So, this is the Open AI

prompt optimizer. And what this does is it allows you to put in a prompt and chat GPT will go back and optimize your prompt for a very specific model. In this case, GPT5. It'll know all the strengths, weaknesses of the model and optimize your prompt for those strengths and weaknesses. So, for instance, let's take that last prompt we used to generate that game and optimize it and get an even better prompt back. So, I'm going to hit enter on this. I pasted in our original prompt and what it's doing is it's optimizing the prompt specifically for GPT5. And so it's going to go through take our prompt and give us a way better prompt that'll give us better results. So here we go. This was the prompt before this little prompt right here. I want you to create a beautiful firsterson shooter using 3JS with the self-reflection. This is the prompt after with the prompt optimizer. Create a detailed stylistic first-person shooter game using 3JS that revoke that evokes a retro aesthetic and features engaging fun gameplay ma mechanics internally. Before beginning development, perform an in-depth self-reflection by constructing a comprehensive private rubric with all of these guidelines and example output and a whole bunch other a technical implementation plan. That's pretty amazing, right? And so now we have a way better prompt and all we did was just put it in, hit enter. Now we have a better prompt. If you use this prompt optimizer, you're going to get a hundred times better results. I promise you, people need to be using this. This needs to be I don't know why Sam Alman doesn't scream this link from the rooftops. Everyone should be using this. If everyone used this prompt optimizer, they'd be getting so much better results and they'd be using GPT so much more instead of creating YouTube videos where they say they don't like it. They should be like front and center on the chat GPT website every time you do a prompt in there. In fact, they should build this into GPT5. So, when you do a prompt, you click like a button and it optimizes the prompt for you. I'd say this, almost anytime you use a prompt, you want to use the prompt optimizer to get better results. This will just simply give you better results because the prompt optimizer understands how GPT5 works better than you understand how GPT5 works. So, there you go. There are the five improvements you need to implement into GPT5 to get way better results. I promise you if you implement these five things, GPT5 will turn into your favorite model. I promise you if you do this, you're going to get results that will blow your mind. I've been using this like 10 hours a day and this has been unbelievable for me. I'll include examples and prompts all down below that you can just copy and paste and use, including links to like the prompt optimizer and all that. So, check that out down below. Now, listen. Do I think Open AAI messed up by not vocalizing this more? Yeah, I do think they messed that up. I mean, they did put out a blog post that basically said, "Hey, here's how to prompt a GPT5 that explains a lot of these things, but that's not good enough. You need to be creating more content. putting out a lot more videos on Twitter that go over these things. You need to be putting more videos out on YouTube that go over these things. Sam Alman should be tweeting out the prompt optimizer link left and right. They should be hiring people like me to shout this out from the rooftops. This video will probably get more views than any other video OpenAI puts out about how to use GPT5 better. If you use these techniques, you will be having a very good time. If you learned anything at all, make sure to leave a like. If any of these are new to you, let me know down the replies which one of these are going to be the most helpful for you. Make sure to subscribe. I also again have a new free AI vibe coding community called the Vibe Lab. Link down below to that. Click that. That's in Discord. We have over 250 people hanging out in that community. So, you want to be a part of it. Make sure to turn on notifications. I live stream Monday, Wednesday, Friday, and I just vibe code and give you tips. So, you want to be a part of that as well. It's a lot of fun. and I'll see you in the next