Automate Your Accounting Workflow with Custom GPTs

So let's get started. Let's talk about ChatGPT. Let's talk about in particular custom GPTs. And if you want to learn more about getting, getting better results from ChatGPT, one of the things we set up just a couple of months ago now is a brand new community. It's on the platform called School. And you can join this completely free of charge. At the moment we are not charging for it. You can get free membership for, for life. To get access to it, just hold up your qr, your phone to the QR code or the link will appear very soon. And just let me tell you what happens if you do want to join and I hope you do because I can help you and help you get more out of ChatGPT. When you join the community you will go to a page that looks something like this.

You just click the join button. It's my gift to you. It's free. And in here we're on the platform called School and one of the things you can do is you can.

You can network with the rest of the community of other like minded accounting professionals and you can ask questions, you can help other people in the community. There's also a classroom and in the classroom there are a number of courses that I've created for people. These are, this is free training on, on how to get more out of ChatGPT. And also in the calendar you will get to see all of the live streams that we're doing like this one. So it's an easy way of you getting a reminder. You'll see, you can see right now it says live because we are live as I'm speaking right now. And then there's some other things you can do if you want. There's a bit of gamification. You can be on the leaderboard if you comment frequently. So that is our new community. We started in March 2025. We are up to 584 members and you are welcome to join. And I know that Marty here has just put in the comments, highly recommend the AI Academy. Okay, the AI Academy, which Marty's in. I will talk about that in a little while. Okay, let's get into the ChatGPT and custom GPTs. So what on earth are custom GPTs? Well, they are, they are really powerful and I'm going to, what I'm going to do today is I'm going to share with you a bit of a bit about what they are, but then more important, how can we use them? What are the use cases, what can we do with custom GPTs. And then to make it really practical, I'm going to do a live demonstration and then I'm going to show you how you build your own custom GPTs. This is so, so powerful. So custom GPTs, they were released back in November 13, 2023 by OpenAI. I'm doing this, it's about just over 18 months ago when they came out. At the time it was my favorite feature of ChatGPT. There's a few more things that have come out this year which I think are even better still, but nevertheless they are extraordinary. What are they? Well, think of them as a little mini version of ChatGPT, if you like, that are programmed. And when I say programmed, don't get think you have to write coding. No, you don't. It's really simple. They basically have a set of instructions, a prompt if you like, that will do something highly specific for you. And so what I like about these little GPTs is firstly they are really easy to build, easy to build, easy to access. And I'll go into ChatGPT later and show you that easy. You can have, you can have them to hand on your left hand menu bar you can create the instructions for them. Think of them as a prompt if you like, or think of them as custom instructions. You can have up to 8,000 characters I believe, unless they've extended it, but I think it's 8,000 characters. So you can write quite long instructions and be very specific in what you want it to do. You can also upload files into the knowledge base which that GPT will always have access to. So when you start a new, when you run the processor each time, you don't have to upload things, it's already there, the knowledge base and the other thing I love about them is they are shareable. And that's particularly useful if you have a team of people and you want to share your GPTs with your team, your fellow partners. It may even be that you want to have some GPTs that you create and share with your clients. So very easy to share GPTs. Now what they are best at is if you have a task that is a repeatable that you do frequently and secondly is very specific. And that's where I think GPTs are particularly strong. If you have a very specific task in mind, you want to do something regularly and specific. So I'm going to show you some use cases that will give you some ideas of what you can do. Now they are. The one downside at the moment is at the moment they are currently only available on the paid plans of ChatGPT. So if you're on a free account, I believe you can't create your own. And I think this is one of the many, many reasons why if you are on the free account, you will want to upgrade to either the plus account or the Teams account. If you're not sure, the plus account is $20 a month plus tax, which I think is incredibly tiny when you consider all the amazing things that you can do with the plus account. It's not just custom GPTs. There are so many more really powerful features that you do not the free plan. Yes, the free plan is great. You could do some amazing things. But if you are, the more you start to use AI, the more you start to automate things in your accounting firm, the more you'll realize that there are some powerful things you can do with the plus account, which is 20 bucks plus plus tax per month, or the Teams account. We have the Teams account. My Entire team uses ChatGPT every single day. And the reason we have a teams account is because the productivity of my team is extraordinary. What they can achieve every week is just mind blowing compared to what was possible two years ago.

So that's a little bit of an overview on what is a custom GPT. Let me talk about then. What could you use it for? What are some of the use cases? So anything, as I said, anything that you might repeat that's repeatable, think of it as little systems. Systems are so powerful. You create systems because.

It means that anybody can run that system and they get done consistently every single time. So a custom GPT is like a mini system. One example might be a report generator. So it may be that you regularly produce reports for clients, let's say. And what you could do is you could create a GPT that creates that report, but you could build into the GPT the specific instructions that tell it, for example, the writing style, the structure, the layout, what, what you want in that report. And we're going to go, I'm going to show you a practical aspect of that later so you can see how easy it is to do. So some of the things that I've built and taught in the Value Pricing academy and the AI academy is, for example, how you might, if you do, let's say diagnostics, let's say that you do cleanup work and one of the things you want to do as part of cleanup work is go into a business, go through and look at their QBO or their Xero account, whatever it is, look for the Errors, the mistakes, the things they do wrong. And then you might want to create some value by creating a report. And so one of the things that you could do is create a GPT for doing exactly that, where you click on a button, answer a few questions, and it then produces an amazing report for you. It might be that you want to create a standard report, let's say, for every end of year meeting with a client to go through their end of year financial statements. And what you could do is you could put in, let's say, an Excel spreadsheet of their profit loss account, their balance sheet, and it then produces in a predetermined format a report either for you to go through with the client, if you like, your agenda, or even to give to the client, where you could look at things like trends, potential risk, and so on. Just bear in mind, little. Just bear in mind, of course, in our profession, not only have we got to be aware of privacy and security, but also client confidentiality, which is really important. I recently did a whole video training course on how accounting firms can make sure they are compliant when it comes to AI privacy and security. And I think that, unfortunately, I think many accounting firms probably fall foul of the regulations, the professional regulations, GDPR and other data protection issues.

So one of the things that I say in that course is when you are, let's say you are uploading an Excel spreadsheet of financial data, that is okay as long as it's anonymous. So just in case you're thinking that yes, you could create a GPT for producing an end of year report, just my advice would be make sure you are giving it anonymized data. Okay, so that's use case number one. We could use it for creating standard types of reports. And you could have multiple GPTs for different types of reports. You could have a diagnostic report generator, an end of year financial statements report generator, and so on. Okay, what else could we do? Well, another thing that you might do regularly is if you want to grow your accounting firm and position yourself as the expert, you might have a blog and you might want to produce regular blog posts. One of the things that we can do is we can of course use AI to help us with that process. And taking that to another level, we could then use a GPT where we feed it with the prompt. We feed it with things like our writing style, the way that we write blog posts. And then every time we want to write a blog post, we just run the GPT. We give it a bit of information, perhaps some new, some News or something we want to build our blog post around and it will then do the hard work for us. Number three, third use case is it's really good for strategic thinking. So one of the things that I do is when I'm thinking about the strategy for my business, for the Value Pricing Academy, I have a GPT and inside that GPT in the knowledge base is information about my business, including things like the business plan, my customer Personas. It's got lots of information which means that I can open up that GPT and initially what it does when I run it the first time, it'll come up with some strategic insights and then I can ask more questions about strategy and it's pulling all the information in I've put into the knowledge base about my business. Taking that to another level. If you wanted to do business advisory, then in theory what you could do, and I've done this for a local business owner, is you could create a GPT for a particular client and use it for business advisory strategic thinking. What else could you use it for? If we think about marketing, if we think about copywriting, I have a couple of GPTs. One's called headline Generator and the other one's called Blind Bullet Maker and they are fantastic GPTs. What Headline Generator does is if I write something that's I don't have an email and I'm struggling with a subject line or I'm writing or I'm promoting a webinar and I'm struggling for a really attention grabbing title for that, then what I can do is I can put in the details into my GPT and what it will do is it will come up with a series of recommended headlines. And it does that because inside the knowledge base I've given it everything I've learned over the years of teaching copywriting how to create different types of headlines, different frameworks. They're already in the GPT so I don't have to teach you that again. Likewise, blind bullets are really powerful in copywriting for creating desire. They are basically the way that you create bullet points, which you do with curiosity. They're like mini headlines. And again, I have a GPT where I put in the learning points and it produces for me really powerful blind bullets. So there's a bunch of ideas for what you could possibly do and if you want even more ideas. Shortly after OpenAI released GPTs in November 2013.

In 2023, shortly afterwards, in early 2024, they then launched the GPT Store where access publicly available GPTs. Now I would suggest Some caution there, because the problem is is they made it so easy to create GPTs that when they open the store there are now millions of GPTs and most of them are rubbish. Most of my rubbish. You could go down rabbit holes doing research to find great GPTs that are publicly available. Yes, there's some gold out there, but. But most are rubbish. However, what you might want to do is if you're struggling with some ideas, you could always go to the public GPT store just to use that as to generate some ideas for creating GPTs. And then I recommend you create your own, because when you create your own GPTs, they do it exactly how you want it. You can specify how you wanted to lay out the report, for example. So let's get, let's drill down a bit more practical and let's look at a couple, let's first of all look at one that I use and then let's build one together so you can see how easy it is to create your own custom GPTs. Now, I know some people are listening to the podcast, so I will try to explain this carefully. But if you are a visual person, you will probably want to hop across to my YouTube channel and catch up with this on my YouTube in my live stream section. So what I'm going to do is I'm going to demonstrate One of the GPTs that I use regularly. So let me go across to ChatGPT and here we are and you'll see the GPTs on the left hand side. Of course you have to have a. To create GPTs, remember you have to have a plus account or teams account and the GPTs are the ones with little icons next to them. And then under that you have your projects and then you have your list of previous chats. You can see I use it every single day, multiple times. It's permanently on. In fact, I use it so much that I have it on multiple computers. I am multitasking And I have two different ChatGPT accounts because I sometimes am doing two separate tasks at the same time. So I have a whole bunch of GPTs here. You may recognize some I've mentioned like Blind Bullet Creator, Live Stream Copywriter and many, many more. Diagnostic Report. I'm going to go into Email Responder. In fact, I used this just a little while ago. If I just click on Email Responder, that is going to open up the Email Responder. I'm just going to hide. Now the left hand sidebar. And so you'll see now we're into the normal, just like starting A brand new chat. But we see specifically this is email responder and it says in the description here, simply give me an email and some context and I will write an email response for you. And then what we can do to make it really easy for people to know what to do. I like to make these GPTs so they are in the main linear. There are, if you think of it as a step by step system, it's a linear system. And for step one what we can do is we can create up to four buttons if you like, four initial prompts. And I always put one in which is called let's get cracking. You'll see, let's get cracking. Because what that does by hitting that button, basically it starts the workflow. And this one is really simple. There are only two steps in the process. It's really easy. So how do I use this? Earlier today I got an email from somebody who wanted to appear on my live stream. He currently teaches dentists something and whilst his content's great, I don't have guests on the live stream very often and I want to make sure if I do they are highly relevant to you. So basically I want to say no to him, but in a nice way. And I wasn't sure how to craft that so all I did was I ran email responder and within seconds it produced a lovely, friendly, nice email email. A nice polite way of saying no. I copied and paste that back into my email and sent it back to him. So let me do an example just to show you exactly what I mean. If we hit let's Get Cracking, it's a two step process and what it's going to do is say that it's ready to start. Please give me the email you want me to respond to. Okay, that's nice and simple. So I've got a demo one here. It's not a real email, but just for demonstration purposes, have an email. I'm just going to copy and paste this in here, let me first read it to you. So this says imagine you've and you may have this. Imagine you've got a meeting coming up tomorrow at 3 o' clock in the afternoon with a potential client and you're looking forward to it. And the day before the meeting with the potential client, the prospect they email to cancel, which is always really annoying. Okay, so here we go. It says, hi Mark, I know we have a meeting booked for tomorrow at 3pm but I need to cancel. I'm really busy at the moment and I don't think I have time for this meeting, can we leave it for now? I'm still thinking of changing accountant, but since my year end is six months away, I am going to revisit this next year. Yours sincerely, Margaret. So perhaps you've had those before and perhaps you're thinking, I was looking forward to that meeting because I've got some ideas to save her some tax. I've seen. I know a bit about a business from, from the initial inquiry. I think I can save her about 15 grand in tax and it's a shame that we can't get to meet. And so what I'd like to do is send an email back that shows some empathy, some understanding, but also gives a compelling reason why she might want to change her mind. So all I do is I send the email off, I copy and paste it in, I send it off to the GPT and it's now going to run stage two. And what it now asks for is some context. So before I draft your reply, please provide me with any relevant background information or key points you want me to include in the email response. Is there a particular call to action you want the recipient to do? So all I would then do is I would just type, just very briefly, some of the things I might want to say. And we don't have to worry about grammar, we can make spelling mistakes, just have a brain dump into ChatGPT. You could even just click on the microphone and just speak to ChatGPT, the sorts of things you might want to say. And so when I did the email a little while ago to the person to say no, I basically said, I'm not. Just let him know I'm not doing live streams with guests right now and a couple of other points. And then it wrote this lovely email. So let's have a look at this one. What might we say to Margaret? So what I might say is, I might say this.

The tax planning idea I wanted to discuss will save you more than £15,000. And I know cash flow is tight. If you change your mind, I'm available on Thursday and Friday of next week. Okay. So that might be the sort of the gist of it. So I hit this button now and what the GPT is going to do. It's already got my writing style, the way that I write my emails. And so it's now opened this up in a canvas, which I love canvas, because I can then. I can always change it then with AI. But let's see what it says. It says, subject, read tomorrow's meeting. Hello, Margaret, thank you for letting me know, and I completely understand how busy things can get, so I always like to show some empathy beforehand. I just wanted to quickly mention that the main reason I wanted to have the meeting was to share a tax planning opportunity that would save you more than 15,000. I know cash flow is tight at the moment, which is exactly why I think this idea could really help. Of course, if now is not the right time, I completely respect that. But if you do change your mind, I have some availability on Thursday and Friday next week and we'll be more than happy to talk it through with you. Let me know what works best for you. Best wishes, Mark Wickersham, fca. And if I was in practice and I got that email, I probably wouldn't need to edit that in the slightest. I would just hit the copy button in the top right, paste it into email, job done. So I don't know how long you spend dealing with emails in a typical day or a typical week. In fact, one of the firms that I was with in Portugal last week, he told me his inbox was. Neil will remember. Mel will remember. It was. It was over 2000. I know some firms spend so much of their day dealing with email. So what if we could take what might be a 10 minute task and make it. This is probably less than a minute. I know, I've been talking through as well, but it probably would have taken less than a minute to create an email like this when you have a GPT. So there's my. The first thing I want to show you and just to remind you, hopefully you can see GPTs are best when you have something that's highly specific. So in this case it's a simple system. I just want you to read an email, I'm going to give you some information about how to respond and then write it in the way I write my email. So highly specific task and secondly, repeatable. Yes. Do we do, do we want to respond to emails regularly? You probably do it 10, 15, 20 times a day, a week. Just imagine how much time that you could save answering emails with something like this.

So how do you build, how do you build a GPT? Easy, really simple. They've made it so simple. It's one of the reasons why when it came out in November 23rd, I really got excited by GPTs. And I have dozens of GPTs. If I just close this one back down. What you see on my sidebar is just the ones I've chosen to pin to the sidebar. And it's one of the things I like about GPTs is your favorite ones you use regularly, like an email responder. You can pin it. So it's always to hand straight, straight away. But if I go and look at, you can see at the bottom it says explore GPTs. If I explore GPTs, what I can do is I can, I can explore firstly the public GPT store with the millions of GPTs out there. But also I can explore all the GPTs that I've created and I've probably created 30 or 40 over the last year and a half, partly because I teach other accounting firms how to do it themselves. So some of them are just demo ones, but the ones that I use most frequently I can put on the sidebar. So let's build, let's build a new one. And so in fact let me first go to explore GPTs. And what you will probably find is GPT is sometimes a bit slow. Actually was a bit quicker today. Okay, so it's now opened up GPTs and at this stage I can now see there's Seymour. I can see all the ones at Value Pricing Academy because we have a teams account so we can share them amongst our team very, very easily. And then we can also see all the ones on the public store, which is a bit overwhelming. But if I want to create a brand new one, all I do is hit Create in the top right. I hit the Create button and that's going to go to a screen that is split into two halves. And actually the default is normally here. So the left hand side is, is where you create your GPT and the right hand side is called Preview because that's really a testing site. So you can start to build a GPT and then you can test it without leaving this screen, which is one of the things I love about it. You can test it and if on the right hand side and if it's not giving you what you want, you can go back to the left hand side and tweak the instructions. Now, on the left hand side, when you're building a GPT, there are two ways to do it. You will notice there are two buttons at the top. One says Create, which is normally the default. One says Configure. Create is for beginners. I do not recommend you use Create unless it's just the first time and you're playing around. Basically, Create means that you have a natural language conversation with ChatGPT and you can see it starts off, it says, hi, I'll help you build a new GPT. You can say something like make a creative who helps generate visuals for new products or make a software engineer who helps format my code. What would you like to make? So I could. I could say I want a GPT to help me answer emails. And what it will do is it will write some instructions for that, it'll write the instructions and then it'll perhaps ask some clarifying questions and we can add, we can have a conversation and build on that. The reason why I don't recommend it is, is because firstly I tend to find that what it builds behind the scenes is the instructions are a little bit basic. And also as you can continue the conversation, it keeps changing the instructions and so you might find you get a situation where the first time it seems to work quite well and as you have the conversation, it changes it and it gets worse and you've lost the original set of instructions. So for that reason I prefer to go to configure and write the instructions in the instructions box. In fact, what I do is I will write the instructions perhaps in a Word document, a notes document, and then copy and paste across. And the reason I recommend that is because if you get into the habit of keeping your instructions in a separate file, if something went wrong, if your GPT was to crash, I don't think it's ever happened on me. But let's say it disappeared for whatever reason. If you've got all of your instructions somewhere else, you can always then just start a new GPT and copy and paste. So let's go through the process of configuring. The first thing you'll see is a plus with a circle. That's where you can upload little image or you can use Dall? E to create it. I don't know why it says Dall E given that.

They'Ve now changed that two months ago with their new image generation. But hey, doesn't really matter. So that's how you create the image. I just use Dall? E. It's simple. But I would do that as the last step. I don't know why it's at the top because it will create an image based on your GPT. So let's do the GPT first and then do the image last. That's the way I choose to do it. You can give it a name. So what am I going to call this one? Let's build a GPT. Let's imagine this and this is a, I think a really cool use case for you that you might want to use. Let's imagine that you help clients, help your clients sometimes with installing, using financial tech it may be you get them using QuickBooks or Xero, or it may be there's other things that you use in your tech stack or you recommend they put in their tech stacks. Perhaps for example, expense claim software, employee management software, work in progress, inventory, stock management. So many different things that your clients could have in their tech stack. And it may be that what you do is recommend tools and then from time to time the client will either make a mistake or they won't know how to use it. And one of the things that you could do to add value is to create a little mini software guide specifically for them on how to use a specific feature relevant to them for a specific tool. And so you could create a GPT that will just very quickly and simply allow you to create a report for your client a software guide for software they're using in their tech stack. So I'm going to call this, because of that, I'm going to call my GPT Fintech instructor. Okay, then we might want to give it a description. This is the description that will appear underneath the name when one of your team decides to use it. So I would recommend you give it. This is a short description of what it does. So I just put in here a training manual assistant that helps you structure, outline and write clear, step by step software guides tailored for non technical business owners. I know you can't see all that. The box is a bit small.

It's not that critical, that box. The really important one is the instructions. This is where you write your custom instructions for this GPT and we'll come back to that in a second. I'm going to just show you the other things that you can do. You can then have up to four conversation starters. So these are buttons that somebody will, when they start the GPT, they're buttons that they could click on that if you like, have a prompt behind them. So the prompt to help people when they're not sure what to do. So you could have a conversation starter, a button that says simply tell me.

What this process will do. And somebody then could just click on that and it'll say this is what it does. Personally, because I tend to do linear things. I have one like let's get started because I'll tell it what to do straight away on step one, we don't have to put a prompt in there. Let's get started. Sufficient so I like to have that. Then we can have a knowledge base so we can upload files. We can upload up to 20 different files. So in this particular Case and I will do it now before I forget, so. So we can upload files. These could be things like example outputs or templates. There's so many things it could be. So I'm going to find a different one. Just bear me a second. I'm going to go to my.

I'm going to go to my desktop, I'm going to go to for ChatGPT demos and I've got an example. It's just an example report of teaching of, of what I'm expecting from this GPT. So I'm just going to upload that. But you could upload all sorts of files. If you were doing, for example, strategic planning for your firm GPT, you could upload your business plan, you could upload your ideal customer profile, your customer Personas about your services you offer. You could upload all sorts of different files and they are in the knowledge base that every time you run the GPT it will always have access to that knowledge.

And then we can specify which of its capabilities we want it to use. Do you want the user, when they run this GPT to build a web search? Do you want them to be able to open up in a canvas? In this case, absolutely yes. Do you want them to create images and do you want to use code interpreter and data analysis? Now, by default you will always find the first three are ticked on because why wouldn't you want those? Unless you decide this is for your team and you don't want them playing around with images, you could just turn this off and that means inside the GPT they could never then create an image. Personally, I leave them on, the fourth one is off. And the reason for that is because if you ever share your GPT, for example, put it on the GPT store or share it with clients, then when code interpreter is turned on, it's not that difficult to hack a GPT and access the instructions. And with a little bit of ingenuity it's possible to also hack and get the knowledge base as well. And that's why OpenAI made it off by default. So I would always recommend that you only turn this one on if a you want it to use code interpreter. That might be to analyze Excel spreadsheets and you are not going to share it on the public GPT store or with clients. And I would recommend that there's no reason I can think of why an accounting firm would want to put it on the public GPT store. And then finally we have a section called Actions which I'm not going to go through this is really advanced stuff. It's about hooking up APIs to link to things like Zapier. That's advanced and to be honest, you do not need to know that for pretty much most of the practical use cases you will want to use. So really that's it, that's building a GPT. You give it a name, perhaps a description. The instructions is the most important thing which we'll do next. You can choose what buttons appear on the GPT to help people get started. I like just a simple let's get started. You can upload knowledge and you can choose the capabilities. Okay, so let's build out some instructions and rather than me showing you now step by step, just going to copy and paste and then just briefly show you what I've put into that box. So not wrong button there. Let's go back to me. So when you, when you are creating GPTs, I do recommend that you have some.

Knowledge and capability of prompt engineering. Prompt engineering is the skill of how do we communicate with AI, how do we write and construct our prompts. And it's something I've been teaching for the last two years. It's a really important skill because the better you are at writing prompts, the better the results you'll get, the better the outputs you'll get. So I spent a bit of time earlier today crafting a prompt for this particular GPT. So if I just go back here and by the way, it's a small box again, but we can expand this box which is useful. So I've just done that and I'm just going to do Control V. So here we go. Here's the prompt I did earlier and let's just go through it and I'll show you what it does. So it's got a goal or mission. The GPT is going to act as a technical writer specializing in SAS user education, with expertise in creating step by step software guides for non technical audiences. Your job, Mr. GPT, is to help write a training manual for business owners learning to use financial software. Your mission is to simplify technical processes into clear, actionable instructions that build user confidence and promote independent use of the software. So I've got there the mission and the role I want it to play. If you've watched my other training on prompt engineering, you'll know that it's very important to give it a specific role to play because you get better responses. It then goes on to say, you will accomplish the mission by following the steps in the instructions and you will accomplish the mission by Always following the rules. So I've then broken down a couple of sections here. I'll come back to rules later. They're at the bottom, but I've also put the instructions. And one of the things that is great about GPTs is that I think they work really well when you've got a series of steps. Not too many steps, Ideally less than 7. My email responder I think was three steps. Step one is to ask for the email, step two is to ask for some context. And step three is to open a canvas and write the email. It's a three step process. So GPTs are very good at following a series of steps. Just don't have too many. It gets too complicated and you sometimes get unexpected and unpredictable results. But as a general rule of thumb, 3, 5, 7 steps can work really well. So what are the steps that I'm going to do in this particular one? So it says, work with me to perform the following steps in order. Do not move on to the next step until I've confirmed I'm ready. End each step by saying, are you ready to move on to the next step or do you want to explore this further? You don't have to do that, but one of the things I teach with GPTs is have feedback loops. In other words, every single step of the process, process. What you might want to do is ask the user, are you happy with this? Should we move on? Because they might not like the response to the first step. And then you might have a conversation with it to tweak it before you move to step two. That's why feedback loop is really, really useful in these steps. So what does it then do? Step one is to ask for the name of the software. So what's the manual we write? Is it qbo, is it Xero, or is it something else? What's the purpose of this manual we're writing? Who's the target audience? The tone? The personality? Then it's going to ask for any specific tasks or topics to cover in the training manual. And then what it will do is it will create a proposed structure for the training manual, grouping tasks into logical sections such as setup, daily use and reporting. And in step four, it's going to present to us that proposed outline and ask for feedback or edits before proceeding. So essentially that's the third part of the feedback loop. And then if we're happy, it's going to open a new canvas, name it after the final report and then it's going to begin writing clear, step by step instructions for each Task ensuring the language is simple, the tone is appropriate and the format is consistent. And it then says, look at this, it says refer to the document called Example in the knowledge base and use it as a guide to the required output. So this is I uploaded that you might have remembered. So one of the things that's useful to put into the knowledge base is either examples of what you're looking for example reports or template reports, and then say in the instructions, tell it to specifically refer to the knowledge base and then it will. That's how you start to create predictable, consistent responses. If I went back to my email one, I have my writing style guide and I also have examples of how I write my emails and that's in the knowledge base. So then the response format, this is I wanted to lay out the responses as follows. I want you to again, I'm just reiterating, refer to the document called Example in the knowledge base and use it as a guide. And also I want you to lay it out with this template. So templates are really powerful. We can specify in our instructions, always have it with a section title, an objective, why it matters, and then the step by step instructions. And then finally I said at the start, I said, you must also follow the rules. Now why rules are useful is one of the things you'll find is when you start writing longer prompts, sometimes it forgets things, it doesn't behave. And so here's a great tip for you. If you say early on you will accomplish your task or your mission by following the rules, it might be initially you just have one or two rules, but then what will happen? What happens is having now written it, which I've done, we can now test it over on the right hand side, we could say let's get started and we can go through the process, we can see the results we get from it. And if we don't like the output, we can then go back into the instructions and add a rule and say, I always want you to do this or whatever it wasn't doing before. So this is the iterative process of creating, testing it and then just configuring it by tweaking a few things until we are happy. Once we're happy, we hit create and we've got a number of options for who we share, we can share with our team. For example, I won't share it. At this stage we hit update and oh, I know what I didn't do. I didn't do the image, I didn't do the image for it. But never mind the image is really simple because all you do is you click on the plus button and it says use Dall E3. You wait about 10 seconds and it will create a little image for you. For.