The Ultimate Guide To ChatGPT Custom Instructions

By using custom instructions in ChatGPT properly, you can go from a result like this to a result like this in one prompt when used correctly. This really is the biggest unlock in output quality that casual users and even more advanced ones can achieve within ChatGPT. You just need to know what to put into those two little magical fields. And that's exactly what this video is going to show you. So I'm going to give you a brief overview what's going on in there. Then we're going to talk through step by step how, how to use these fields to get the best results possible. And then I'm going to give you a formula on how to navigate it with some examples that you can copy paste right into it. So, first things first, they launched this feature. ChatGPT enthusiasts have been waiting for this since a while because as you might know, context is the thing that generates quality outputs. And what this really is is a super user friendly way to add context to your prompts. So accessing it is very simple. Three dots settings, basically beta features and turn on custom instructions. After you've got that, you can close the window and reopen this and it says custom instructions. All right, so now we have these two fields and the big question presents itself. What do we put in here and what does this actually do? Well, as mentioned, this is just a different way of adding context to your prompt. And concretely, the first one is here to specify your personal context, so anything that touches you. And the second part is responsible for modifying the outputs of ChatGPT. And don't worry, I'll tell you exactly what to write into these in a second here and show you how to customize that. But in order for you to do it properly, you actually need to understand what's going on here. And therefore I'm going to give you a visual description of what this context means. So picture this, this circle right here is all of GPT4's knowledge, all the data it has been trained on. And if you give it a simple one line prompt like write me an email, it's going to do that, but it's going to start using context from all these different spots to write that information. Now, if you get a lot more specific there and you say write an email to my boss called Henry about the coffee in our office, write it in the style of Ernest Hemingway, you start limiting the areas in which it finds these little circles, whereas before it might have been looking for them in an area this large because you didn't specify what exactly you wanted. So you Gave it more freedom. Now what you do by filling this out correctly is if this is all of GPT4's knowledge, you tell it, hey, I want to operate within here. And if you're even more specific and you give it more information to go off, off, you tell it, I only want info from within this little circle. And if you get super specific, it might actually be more like this. And within that little circle, it's going to look for the information because you asked it to, but you have to know how to ask it. So that is the next point. How do we set this thing up so we get exactly the type of results we're looking for? And look, this is no easy task because often we don't know exactly what we're looking for, right? But if we just think about the real world for a second, what, what do you do when you have accounting problem? You talk to an accountant. So the answer to the problem of not knowing exactly what you want to solve for lies within talking to the right person. Same thing goes for ChatGPT. You got to talk to the right Persona. And there we enter the territory of a discipline within prompt engineering called Persona modeling, where you lay out all the details in a very specific way. So you talk to a specific person and all the results that you get are specific to that person. So if you check out this notion page I have right here, these are the exact instructions you would want for, for a digital marketer. Now look at how specific these are. Right? We specify the profession, the current project, specific interest, values and principles, and so much more. And these have been carefully crafted because since the day of the release, we have been preparing a framework for you guys to use within ChatGPT. And this is not something that we spawned over the last week. We've been wrestling with this idea of Persona modeling at a very high and very precise level since months now. So a lot of the work that has been done we just repurposed here and I'm giving it to you right here. So check it out. You go into custom instructions, you post the first part, and then part two has been crafted for the second part, which asks, how would you like ChatGPT to respond? Well, right here we would like to specify the response format, the tone, detail, level, types of suggestions, and much more. For example, the problem solving approach is one that makes a massive difference. Bias awareness is a little more subtle but still critical to include. And again, all of these have been carefully crafted. And this is kind of the framework I recommend here. If you want to understand this In a little more detail, I'll just put a description of the various areas onto the screen right here. You can check it out. So these are the detailed description of what's going on here. So the values and the principles represent the key principles or values that guide their work. Right? And the devil is in a detail here. You really need to follow this closely for it to work well. But then on the other hand, you will get rewarded with very precise results. Okay, so now that we have this dialed in, what can we do? Well, simply put, we say save, and now we have a custom version of ChatGPT set up right here in the browser. And I can use both GPT 3.5 and GPT 4 to generate outstanding results which are very specific to the thing we're solving for. So, as mentioned in this case, we set up a digital marketer Persona for ChatGPT, and now we can use this to get better results. So let's do a little comparison, shall we? I'll just turn these off for a second and run a basic prompt. So right here you can disable for new chats. I'll open a new chat GPT4. And what we're going to do is in the free notion template that we provide, we're going to head on over to writing and use one of my favorite prompts called Write anything. And down here in the examples, you have one of these prompts that is just a basic blog writing prompt, okay? And we're going to run it for the first time without custom instructions in GPT4. Like so I'll go to custom instructions, enable these newchat GPT4, run the same prompt, and as you can see right here, the differences couldn't be any more different. Important point here is look at the structure, look at the vocabulary, look at the tone of this thing. So, folks, it's time to stir things up. Catch you there, fellow earthlings. It's these little snippets that in my opinion, are just a little too much. This is not really dialed in, and a lot of these parts are just very chatgpt typical, like picture this or here's the cherry on top. Whereas this result is way more formal. And the reason that is is because in the custom instructions, we told it to do exactly that. So if you look over here, it says to maintain a professional tone that balances friendliness and formality, and these instructions override the prompt up here. Okay, so that's a big takeaway already because remember, when we gave it custom instructions, we told it to operate within A certain area. And no matter what you do in the prompt, it's not going to go outside of that. Okay, let's do another example. If I pick an entrepreneurship character here, let's take the serial entrepreneur. And this one is one of my personal favorites because this Persona is experienced, it's been around the block, it's managing a portfolio of diverse ventures, and it just has this very, like, macro overview of things. And its communication style is straightforward and strategic. So same thing. First part, we copy over. Second part we copy over. That took five seconds, save. All right, and then I'm going to head back to the free notion template. And under entrepreneurship, I'm going to pick one of these prompts. We're going to be evaluating business ideas. All right. Okay, so what about this one? This one is actually quite complex. It's going to be evaluating the viability of launching an online language tutoring platform by conducting a SWOT analysis. All right, why not? Let's test this inside of GPT4 with the custom instructions enabled and compare it to the vanilla version. Okay. And again, second chat, custom instructions off, new chat. And we run the same prompt. And here it doesn't take an expert to see the difference right away. So let's have a look at the SWOT analysis. Okay, so as you can see, on the left side we have custom instructions on. And on the right side, they're off. So what the entrepreneur really did here is it concisely gave you the essence of what's happening here. It gave you the information and no extra words around it. Right. Something I personally would prefer when doing a SWOT analysis. So now let's look at the content of these points. Unlimited geographical reach, flexibility and affordability. A platform like this has the advantage of being accessible to users worldwide. Users can learn at their own pace and according to their own schedules. In many cases, online tutoring can be more affordable than in person instruction. Fair enough. What does our seasoned entrepreneur say? Potentially scalable business model due to online nature. Same as over here. Lower overhead costs compared to physical institutions. Aha. See, he's thinking about the finances and the business side of things. That's what the Persona is set up to be. Good, good. And the flexibility is same as over here. Okay, so pretty similar. Okay, so let's head down to the threads and the market size, because I think these two points really illustrate the difference nicely. So first of all, it talks about competition. That's the same over here. Then it talks about the regulations. Same over here. But here's the thing. Here it Says it's dependent on tech infrastructure and it states that cyber threats could negatively impact the platform. I don't know, might be just me, but that doesn't seem as one of the big threats threats to an online learning platform. Right? And here the entrepreneur talks more about oversaturation, so he talks about the supply demand dynamics, which I believe to be the preferable topic when we're talking entrepreneurship and starting a business. Right? All right, you might consider that minor. But what about this next point? So here without the custom instructions, it first goes ahead and states something like as of my knowledge cutoff in September 2021, the Global Learning market was estimated to be worth approximately 60 billion in 2020. And then some growth projections and the estimate by 2023 however, conduct updated market research whereas the seasoned entrepreneur actually goes ahead and he quotes a specific type of research. According to Grant View Research, the global learning market was valued at 12.49 billion in 2020 and expected growth from 21 to 28 was 18.7% per year. And once you go into this bottom paragraph, these subtle differences start becoming not so subtle anymore in the market size. This actually quite quotes are researched that after a few minutes of searching I couldn't exactly find. But that's an inherent weakness of GPT4. It doesn't have direct references, but I'll give it bonus points for at least attempting to quote something precisely as opposed to over here it doesn't mention a source whatsoever. Down here in the competitive landscape analysis it mentions the same competitors, but here it directly outlines the biggest benefits of those platforms. Again, in my opinion a better answer than this over here. And then in the end there's just an extra idea in here. Plus the conclusion gives you actually an actionable step in the end that you could continue with, whereas this just gives a generic ending. So as you can see, sometimes it's subtle, but the deeper you go and the more it really matters that your results are not a 3 out of 10 but a 6 out of 10 or a 8 out of 10 every single time. The more you will want to go into these custom instructions and specify exactly what you need. Because when we look at the serial entrepreneur, everything that we just saw reflected in the answers you're going to find in here. Right? Checks and balances. Please cross verify market data and trends, resource references, reference reliable industry reports and credible news sources. It at least tried, right? If we had plugins on this would perform a whole lot better. And you can do that logical analytical approach and concise and action oriented responses. We actually saw all over the place. So overall, I would rate this as a 7 out of 10, whereas this is more of a 3 out of 10. And this is the experience you're going to have across the board if you specify the following characteristics within your custom instructions and you do it well. So that is an important one. Right? So if you're specifying the professional role, you can't just go ahead and say, I am the CEO of a pet food company, I have 20 years of experience. Because what you're going to do here is you're going to be creating confusion because this is very precise and it yields very precise results. Whereas if you do this, it didn't ask you for your experience, you could maybe add that within the specialized knowledge, but not within your profession. And that's the point here. If you're going to go this specific, you got to craft it carefully. So the best thing you can do is either take some time and play around with this until you start getting results that you actually like, or you will just use the ones that I will provide in the description and you can mix and match different parts. So what I'll do is I'll provide the two that we looked at plus the marketing analyst in the description below and you have the whole framework. As you can see, all of these are repetitive. And yes, we absolutely use GPT4 to generate this. We had our eight sequence prompt to arrive at this point, and I'm not afraid to say that the results are stellar and this works. So what you're going to want to do is you're going to look at these and you're going to pick various parts of these that fit your needs, and then you're going to mix and match them and extend them with your own context. Just make sure not to go above the 1500 character limit. And if you don't want to do that by yourself and you'd rather just copy, paste and then maybe tweak a minor thing, we're also going to be offering this as a product where you can go ahead and you can pick any one of these that we created for you and you can get going right away. You can find a link to that in the description. But really I just gave you all the tools for you to craft it yourself. Just make sure to have a separate one for every use case because these are very specific. If you, if you go into this level of detail, you're going to be getting that very focused answer. So if you're trying to write something over here, you don't want a Persona that is situated somewhere over here. And obviously this is a gross oversimplification, but it's a good way to think about these things so you understand how to use Personas. And the most exciting part about this is that every single prompt you ever used is going to get transformed by these custom Personas that are now super easy to use because you can just set it up once with a high level of detail, and then you can breeze through your daily tasks with a level of precision that up until now was only possible if you did sequence prompting with very specifically casted Personas. I hope this was helpful. If you have any questions, I'm happy to help you out in the comments below. And if you enjoyed this content, chances are that you would love this video with some advanced prompting techniques to which you can also apply Personas. And now that you know the secret to better results, go out there and experiment.