5 Prompting Upgrades You Need for GPT-5 and Agent Mode

Yes, very quickly. I always update you on any new stuff around ChatGPT and there's a couple of new things in the last week that you might not be aware of. So they, they every now and then they update, oops, wrong button. They update their release notes, two things that have just come out. Last week they introduced branch conversations. I haven't got my head around how this is useful, why it's useful, but essentially if you're having a chat, you then want to go off the tangent. You can go off and set up as another chat while preserving the existing one. And the way you find that is. Let me see if it'll work here. Yeah, so when you're having a chat, I've got a conversation here. I'm going to show you this one later. But during a conversation you will see there's three dots that say more actions. And one of those actions has always been to have it read aloud if you prefer it to read to you. But the other one says branch in new chat. And what that will do is set up a brand new chat where you can have extend the conversation without I guess polluting for a better word, the current one that you're in. So if you want to go for a tangent, that's what that feature is for. It's brand new. I've not had any feedback on it yet, I've not played with it, but just thought I'd let you know. And then also the day before on September 3rd, one of my favorite features is projects. I know many people in the AI academy love projects and if you are not yet on a paid for account, then the good news is that they released projects to the ChatGPT free tier as well. So if you're on free, you can start playing around with projects and that means you'll understand what the fuss is that I'm making about them because I think they are awesome. Okay, so today's session we're going to look at the new GPT5 model. So it all changed back in early August and it was a huge change. And what we're finding is that some people love it and some people hate it. So if you are in ChatGPT, then you will notice that at the top it by default you will have GPT5. It's the brand new model that came out in early August of 2025. And what that essentially did was consolidate all of the other models. In simple terms it's actually more advanced than that, but it's really built everything into one place. So I'm going to talk to you about what that is and how they use routers and so on. I also want to share with you in this session some proven prompting techni because we have to change the way that we prompt. And I'll give you an example use case. I'll show you something. We'll look at an example I think we'll probably pick. I'll talk about a prompt for bank statement analysis and how we could. Then I'll show you how you could improve a prompt using that as a worked example. I'll also talk a little bit about agent mode because that came out in July of 2025 and.

GPT5 is to some extent created to work with this more agentic process. And we'll look at some of the little tips that you might want to use in terms of your prompting. Creating prompts for agentic tasks as well, using agent mode. Right, I'm going to go, we're going to look at GPT5 and what's new. I'm going to kind of. I'm going to go through what's new with it fairly quickly because I really want to spend today focusing on being really practical. I want to share with you five practical prompting tips, things that you can build into your prompts. And, and we'll talk about how you can create a template for doing that. And I'll talk about the idea of meta prompting. And so if you are completely unaware of GPT5 or you haven't been using it, you feel like you're a beginner level. What I also did was I did a video just a few a couple of weeks ago now, and if I can just pull it up, it's called how to fix ChatGPT5 and I'm going to ask my team just to put the link to this particular video in the comments. So you might want to go and watch this afterwards where I talked about in More detail why GPT5 merged the models about the router or the router depending on how you pronounce it. I talked a bit about trigger words for better results. We talked a little bit about clear prompts. And I also did a worked example in that where I took a prompt that's always worked really well for me for transforming transcripts. If you ever take transcripts of the words you say, meeting notes, YouTube videos, whatever, it might well be podcast episodes, then I've always used a prompt to tidy that up. And in this particular video, I then use that as a worked example and show you how you might improve it to work better with ChatGPT5. So. So you might want to watch that video link is coming. If you are listening on the podcast, by the way. Then if you want to catch up with some of the visual things, you will. You can just head across to my YouTube channel and catch up on the live stream. So go and check out that particular video if you need to catch up with the basics. And.

I'm just checking to see that, yeah, the link is. The link is there, right? So.

One of the things that. So what they've done with GPT5 then is essentially we had a whole bunch of models before that. If we go back to prior to August 2025, we had the GPT series and we had GPT 3.5. It became GPT 4 GPT 4.0 and then they released their GPT 4.5 brief and then they went back and did GPT 4.1. It kind of got very confusing with all these GPT models. And then they launched, a while back they launched the advanced reasoning models, which was the O series. So they launched O1, O3 and these were from. For more a chain of thought type of prompts where you wanted it to think more thoroughly. This might be good for things like analysis, particularly analyzing, let's say financial statements, analyzing Excel spreadsheets. And so we had to get to understand which model was best for what task. And then when GPT5 came along, what it does is essentially it has a little router or router if you prefer it built into it and it's now just the one model. But it then when it reads your prompt, when it sees your prompt, it then makes a decision. It makes a decision. Should this be a.

A quick thinking prompt like we've been used to with GPT4, or should it be more of an advanced reasoning one? Should it think through more slowly? And it would make that decision.

On your behalf based on the prompt, which in theory means that we don't have to worry about all those old models and which one should we pick. But what happened was a lot of people got very frustrated because they lost control. Sometimes it would pick, it would do things fast when you wanted to think slowly and sometimes it would take a while thinking when you just wanted a quick response. And it wasn't always doing what we wanted. So they made a quick change to it. What they did is they then added in a drop down box which would mean that you could manually choose whether you want, as they now call it instant, whether you want the answer right away. And instant is more like the old 4O model or whether you want thinking, which will then do the advanced reasoning that we were used to with the O series. And if you want to go to a higher level, they have the Pro one as well. But that depends on what plan that you are on. But let's just put that to one side. Essentially the options are do you want it instant like GPT4, or do you want it to think things through like the O series? And now you have that element of control by picking the one that you want. But what they also found was that a lot of people got frustrated because they still loved the the previous versions. And so they also enabled in Settings, a setting where you could, if you want, bring back the old legacy models. And for a little while that was all the legacy models you could access 4.1, you could access 4.5,03, etc. I don't see those now. I think that's now changed. As far as I can see on my version it just gives me the option of 4.0. So we can still have the model that most people like the best before.

So you've now got a bit of choice. Some people love GPT5, some people hate it. Now the reason why people, I believe don't like GPT5 as much is because it's changed the way that we need to talk to ChatGPT. In fact, they, they brought out a Back in on the August 7, 2025, they brought out a GPT5 prompting guide where their research team researched well, how do we get better results out of. Out of GPT5? Now you can go and read this. It's actually quite a long article. It also gets very technical because it covers in here a number of aspects. It covers agentic workflows, it covers coding performance which may not be relevant to you, and then also instruction following. So you don't need to read this because I've read it and picked out the bits that I think are most useful for you. But if you do want to read it, it's there. It's the OpenAI Cookbook, the GPT5 prompting guide and I'll ask my team to put a link to it in the notes that go with this session. And interesting. One of the things that they said was that when it first came out, GPT5, I've spent several a couple years now learning prompt engineering and how to craft prompts and I was starting to think, does this mean that prompt engineering is now less important? Because does GPT5 now make it much, much easier? And what they said in their guide. This is from OpenAI, they said. However, it's careful instruction following behavior means that poorly constructed prompts containing contradictory or vague instructions be more damaging to GPT5 than to other models. In other words, prompt engineering is now more important than it was before. 4.0 was fairly forgiving, but with GPT5 we have to learn a few extra tips and skills when we if we want to get better results from ChatGPT. So what I'm going to do is I'm going to share with you 55 really powerful tips that you can use in your prompting and we'll then make it practical. And as I mentioned earlier, if you want to go back to some of more basic stuff I did in another video, go and check out the video that I mentioned earlier. It's on my YouTube channel and this will nicely follow on so let's look at five things. The first one is something that's actually not totally new, but the OpenAI research team have stressed this is more important than ever and that is having clear section headers. So it's always been the case that in prompt engineering is that the more that you structure your prompts, the better the result that you are going to get. And kind of on that point, I'm going to show you a template for this to illustrate some of the points. Of course, some people might be thinking this now sounds like it's a lot of work to craft prompts, and I often talk about the two different types of prompts, the quick prompts and the prompt templates. Sometimes a quick prompt is all you need. If it's a quick, simple task you want, you don't need to necessarily write really long prompts. But nevertheless, even with a short prompt it's good to use some of the things I'm going to talk about. But if you're going to do a task where a it's really important and or it's something that you might repeat, like analyzing financial statements, like a prompt to answer emails, things that you might do frequently, it's worth spending a bit of time crafting a better prompt, saving it somewhere in a prompt library in a database so that you can then use it whenever you want. These are prompt templates and I do recommend you build up a library. I think of these as being mini systems that are going to save you time. So going back to the idea of clearly creating some structure in your prompts, what that means is you want to break it up with sections. And what I've done here is I'm using Markdown. If you're not familiar with Markdown, then it's a really powerful language that we can use when we are talking to AI systems because it creates that clarity. So in this particular case with the headings, we can use hashtags to denote heading style 1, heading style 2, heading style 3, etc. In other words, if you put a single hashtag, then a space, then a, then something, then computer systems including ChatGPT and other large language models see that and Understand that's heading style 1. If we put two hashes in front and then a space, then that's heading style two. And what that does is it creates clarity. It creates clarity for us when we're reading our prompts back or when we're crafting our prompts, but it also creates clarity for the model because the model then knows when a section starts and when a section finishes, it makes it much more clear. And OpenAI have confirmed that having this structure and this clarity will help contribute to getting better results. So you can see here, and I'm going to show various elements of the one on the screen. This is a sample template structure that you could use. And we've got a section where we talk about the role and the objective. What role and object do we want ChatGPT to play? If you followed my training before, you'll know that in prompt engineering, we get better results if we tell the model what role to play. If we say, for example.

I want you to act as a financial accountant skilled with analyzing financial data. If the task we're asking for is to analyze some numbers and give us a report, then it's going to give us the sort of response that we'd expect as a financial accountant, it's likely to be a better response. And if we say act as a financial accountant, as opposed to leave it blank if we don't say anything, the default is a helpful assistant. If you want to write some marketing copy for your website, then the role might be I want you to act as a copywriting expert. So we give it a role and also by giving it an objective or goal or an outcome that we're looking for, that then gives it that clarity as to what are we looking to achieve from this prompt. When we give it an objective, a goal that we're looking for, then even if the rest of our prompt is not so well written, it knows what we're looking to try and achieve. And so that's why it's good practice with GPT5 to start off with the role and the objective and have that in a clear structure. And you can see we've got further sections here, we've got workflow checklist, I'll come back to that, instructions and so on. So that's my first tip with prompt engineering. You want to start breaking up your prompts into sections. What it also does is the more you do this, and I've been doing this for two and a half years now, the more you do this, when you start even creating short, quick prompts, your brain starts to then get better at crafting prompts because you, in my case, I always just naturally start off with what is the role, what's the goal, what I'm looking to achieve here. And so that structure is now in my brain for how I craft things. Even if I just want to do a quick task, I will use some of these elements. Okay, tip number two, and this one is very new to GPT5, and that is planning before execution. So the OpenAI research team have told us that one of the things that will help us get better results, particularly if it's an important task, if it's, if you're doing some research, for example, and also if you're using agent mode as well, then what they say is that if we ask it to plan out how it's going to do the task in advance, if we ask it to plan things out, then it gives a better result. So that's why they say in their prompting guide is that a good thing to do is to have a section that. I've called it workflow checklist here, but it says present a bulleted list, three to seven items each less than 12 words. Dealing, dealing the. I've missed a word there. Dealing with the recommended workflow steps. What that does, and you'll see that in an example very soon, is it forces the model to think through from your prompt. Okay, so how am I going to tackle this task? What are, what are the steps that I'm going to go through? And when it thinks that through, it then more accurately follows the steps and does the things you've asked it to do. And then what we can do, as well, as you'll see in a bit, we can get it that before it then gives you the result, the output, we can get it to validate itself and check it has done all of those things. So that's best practice. Number two, when it comes to using GPT5. Number three, the team at OpenAI tell us that it's good practice to have a section about the output format. So this is where and this is nothing new. This one's actually been good practice with prompt engineering since the start. It's good practice to specify in your prompts what sort of output are you looking for. For example, do you want it in Markdown language? I often ask for things in Markdown because I'm going to then turn it into a standard prompt and I want the hashtags and the asterisks, I want embolden and so on. So you might want it in italics. Sorry, in markdown, you might want it to put it in a separate Canvas. Canvas is another one of their tools. One of the features that I love in GPT and if you are in GPT and want to use Canvas, you can do it manually. You can go into the tools, it's the plus symbol and you will find there is one called Canvas. If you can't find it, by the way, just put a forward slash in the start of your prompt that pulls up all the list of tools. So Canvas is really powerful. I won't go into the detail of that right now, but I often, if I'm doing written work, I'll want it to put it into a canvas in the output format. Other things you might specify is do you want it to be broken up into headings, do you want your output to be in bullet points, do you want it to be bullet point style, or do you want to be written paragraphs? So always think about the output that you want and have that clarity in the prompt. So if you were thinking about building a template GPT5 prompt, you might have a standard section that says output format, where you would then just list some bullet points to say things like open a brand new canvas, I want you to structure it with headings for each of the main sections and bullet points, etc. Number four. Number four. So one of the things that they've done with GPT5 behind the scenes is they've given us more control, particularly if we're using the APIs. Now you probably are not using the APIs. It's more the coders and the software developers using the APIs. But there's a new API setting and that is verbosity. Now, verbosity is basically how long an answer do you want? And now we can control that more carefully. So it may be sometimes that you just want a short, quick answer, you want something in overview, but sometimes we want more detail in our response. And so we can specify this with verbosity.

So it's a useful little tool that you can use. And even though it was designed for the APIs, it's a setting in the API where you can set it to be brief, normal or longer. There's the different settings. We can also use this facility if we're just creating our own prompts. And so one of the things that you might want to do as you're crafting your prompts is have a section in there called let's say Controls. And then what you could do is you could specify the verbosity between, let's say, brief, normal, detail. How long do you want the output to be? Now, that's not the only control that you can use. There's also reasoning. You can tell it whether you want it to be quick, balanced, thorough. This is how much you want to think things through. So this, if you like, is the equivalent of going into the drop down box of the drop down box and specifying, do you want to use instant, do you want to use thinking? Well, if you're creating a prompt template, you want to cut and paste in every single time with a standard prompt, you probably don't want to have to always remember to then also select the appropriate model. So instead we can specify that inside the prompt itself. So if you're analyzing financial statements, for example, and have a standard prompt, you might want to always have reasoning depth to be thorough. As a default, I don't go into the other two right now. Essentially, autonomy is to do with the agent mode. I'll touch on that later. And parallelization is basically a setting we can control if we want to, as to whether we want it to do tasks step by step by step, or we're happy for it to go off and multitask. That's more advanced stuff. That's perhaps for another day. So let's move on to number five. That is something. This is new with GPT5, and that's planning and verification. So what the team have said is that when we are using GPT5 for more complex tasks and also with the agent mode, then one of the things that we can do is we can get it to tell it to think things through in steps. They make it very clear that stepwise thinking is really powerful. And so I often now build steps into my prompts, particularly the more complex prompts, but we can also ask it to verify. And this is interesting because one of the things that has always been a problem with large language models, particularly in 2023, was hallucination. It would make things up. Well, one of the great things about GPT, and that's of course for accountants, a big issue, particularly if we're doing things like tax research, we can't afford for hallucination, we can't afford for it to make sense, stuff up. And fortunately these models have got better. And GPT5 is in a lot of the tests now, the one that hallucinates the least. And so that's reassuring. But nevertheless, we can ask it to then double check things and think things through, and that's the planning and verification. We can use that, for example, to further reduce or capture any possible hallucination issues. If you don't know what hallucination is, essentially that's a term that in AI means that it's when a model gives you an answer that's wrong, but in a highly convincing way, that's what hallucination is. It's less of a problem now. But all models, whether it's ChatGPT5, whether it's Claude, whether it's Gemini, they do still hallucinate from time to time, which is why, particularly as accounting professionals, we need to double check everything that we do. You don't create a report and send to the client without checking it first. So some of the things that we can do here in planning and verification, we can ask it to think step by step before drafting findings, get it to think through step by step what it's doing. We can also ask it to verify the accuracy of what it does. And then, and this is a bit more advanced, but we can ask it to create a rubric. What we mean by that is it can create a scoring framework where it will then score itself. And it doesn't have to tell us the scoring framework, but what we say to it is based on what the goal is, the objective, based on the task. And if you remember, if I, if I go back to, I think it was number two.

Number two, one of the things that OpenAI tell us is that if we get it to present.

A list, a planning list of what it's going to do before it then does the work. Once it creates that, it's more likely to follow it through. But we can also then get it to score itself against that and we can tell it to create this rubric. Now you don't need to worry about what the rubric is necessarily, it's something that you could create it yourself. But actually ChatGPT is fairly good at creating its own rubric. We just tell it to do it. We say create a private five to seven point rubric, in other words, come up with a scoring scale across a number of different criteria such as clarity, accuracy, Tone, audience fit, completeness, things that it will decide based on what the objective is and the prompt. And then we tell it to iterate until it passes, till it gets the right score, till it feels it's done those things correct. It happens behind the scenes, we don't need to see it. That's why it says create a private four to seven point rubric. But that's what it'll do behind the scenes. It will go and come up with this scoring system which it will test itself against with the output. And if it doesn't pass, it will iterate and think things through more deeply until it feels it's past its scoring system on all the things that are important in the output.

So there's, I want to go through those in those five points, but let's now just pull that together and make it just a little bit more.

Practical for us. So what I recommend that you do is perhaps create a template for your prompts. And those of you who are, if you're in the AI academy, you'll know that you get access to my prompt library and one of the prompts that's in there, if I can just move across to here. And one of the prompts that I just put up there is called GPT5 agentic metapromptor. And I'm going to show you what this does in a little while. Also, I'm going to use a worked example.

And I'm going to use a worked example. If I just type in bank a prompt I wrote some time ago now, probably about a year, year and a half ago, was a prompt, a standard prompt you could use for giving it a PDF of a bank statement. And it will analyze it and create a CSV file of the transactions. And this is the prompt and it works reasonably well. It worked very well with GPT4, but I thought, let's have a look at this and now let's make it work better with GPT5. So I took this prompt. So this is what meta prompting is. You can take a prompt that you've got that's perhaps used to work, but it's not working so well. With GPT5, you could take a prompt that you've got already and then you could get chat GPT to improve it. And one of the things that GPT5 is very good at, it's very good as a prompt engineer at improving your prompts. So I created this particular prompt to do exactly that, to instantly improve your prompts. So let me show you, let's make it practical and so if I go up here, here is the prompt. Let me just talk it through and you might want to create your version of this. You are a prompt engineer. That's the role with advanced expertise in GPT5 prompt optimization. I want you to carefully review the prompt and follow these steps. Now, the prompt the way if you follow my training, if you're in the AI academy, if you use my prompts, you'll know that one of the things I like to do is make these easy by putting the stuff that you have to copy and paste in, in this case, the prompt at the end. So there's a section here called prompt, and this is the prompt that I want it to improve. So let's just have a look at the prompt first that I want to improve. You are a data analyst skilled in organizing and structuring financial data. Your task is to structure the transactional data from the attached bank statement into a CSV file with specific columns. So the steps to go through is to review the attached bank statement, extract the following transaction date, description, amounts, etc. And then create a CSV file, ensure it's accurately formatted and free of errors. And then we have a few rules. So this was a prompt that I created a while back for accounting firm owners if they want to analyze a PDF bank statement. And it worked very well under the old GPT4 models. So that's the prompt. So going back here, you're a prompt engineer with advanced expertise in GP5. Carefully review the prompt that's the bank statement turning into a CSV file and follow these steps. I want you to convert that prompt into the required template structure. This is a template structure for GPT5. So I want you to take that prompt and put it into the format, ensuring all fields are actually completed. If any information in the prompt is missing, unclear or incomplete, clearly identify the gaps and generate concise clarifying follow up questions. So it may be when it looks at your existing older prompt, there's things that just aren't clear in there. And so it's going to ask you some questions and then present your clarifying questions in a dedicated questions for you section format as a bulleted list and also provide a dedicated recommendation section with actionable suggestions. So in other words, what I want it to do is I want it to look at my old prompt, which I've copied and pasted in, and I want it to look at the prompt, put it in that new template structure for GPT5. Also, if it's missing anything, ask some questions and also give me some recommendations for how I might improve that prompt. So then we have the template structure, which I've gone through. So I wanted to set out with a role and objective. I've got this workflow checklist where it plans it out in advance. I've got instructions, output format, controls, verbosity, for example, and planning and verification. That's the template. The five things that we've gone through so far, and there's a few other things in there as well that are a bit more advanced, but all the five are in there. And then we have the prompt. So I ran that and it thought for 1 minute and 11 seconds. Okay, so it's because I wanted to think this one through and get it right. And after 71 seconds, here's the revised prompt. Roll an objective. You are a data analyst. Your task is to extract instructor transactions from a bank statement into a clean CSV. The outcome is a validated file ready for import or analysis. The audiences, finance and operating teams who need accurate, tidy data. It's then come up with what success means.

Because this will then create a rubric against it. It'll score itself. So it's suggesting that success is every transaction is captured with no emissions. Columns match the required schema and order dates and numbers are consistently formatted. Amount equals debt, debit minus credit for each row. A basic validation report confirms totals and row count. So that sounds great to me. That, I would say, means a measure of success. It's then got a workflow checklist of the things it's going to go through and then it's got the specific instructions. So it's going to go through and plan the process and then it's going to execute.

And this bit here is blocked. It will let you know if there's no. If you forget to upload the PDF, it's got the output format there, which is the suggested one. We could change that. So it's suggesting that we have the CSV with header rows with we're specifying in the output, what headers do we want in the CSV file, what the date format should be and so on. This is stuff that we often don't specify. We create our quick prompts and then worry why the Excel spreadsheet, the CSV file isn't in the format we want. Well, it's because we didn't tell it what we want. That's what output format is for. We've got some controls and then we've got the planning and verification here that we've asked for and what it then did as well, because I asked this prompt to do is it's got questions for me, of course I've got to upload the bank statement, but also it's confirming, for example, are we happy with that date format, what should be the currency symbol, and so on. And this is great because this is telling us things that if they're important to us, then we might want to change the standard prompt that we're going to use. It may be that perhaps it needs to be a different format. Perhaps we want to specify all amounts are in euros or US Dollars. The more we specify in our prompts, the more predictable, the more consistent, the better the output. So this is now what this, this prompt is doing is coming up with some questions for us to consider.

Things like the sorting order, do we want the sort them ordered in date order? Do you want from the first date to the last date, the last date to the first date. And then it's also giving some recommendations for us that we might want to think about as we craft our prompt.

And this works really, really well. Creating a meta prompt using GPT5 as a prompt engineer is a great way to improve your existing prompts. This just happens to be one version of a metaprompter. This bit here, there are other meta prompts that we could use to improve the prompts. You could even just keep it really simple. You could say, here is a prompt I've written. It's not really working for me. How can I improve the prompt? Act as a prompt engineer and it will improve your prompt. But the better we get through the prompt, the more detailed the prompt, the better the output. And this one I've been testing a few times, I really like this. I like the way that it takes a prompt, puts it in that GPT5 framework that OpenAI tell us is a better way to craft our prompts.

And also then come up with some questions and so on. So let me just wrap up with a couple more things. One is I mentioned that there's this agent mode. So if we go into chat GPT and we go to craft a. If I craft a brand new, let's craft a new chat for a minute. We have the plus button, which is the various tools we have access to and one of them is called Agent mode. It says new next to at the moment, Agent mode came out in the middle of July. And this is extraordinary when we know how to use it. So what agent mode does is when you craft a prompt, what it will do is firstly it will clarify a few things. It will do that. This is what I'm going to do, or at least we should tell it to do that. And then it'll open up a little window, if you like, a Little desktop within ChatGPT. And what it does, it goes away and does research and it will go and click on things. And you can see it clicking on things. It does take a bit longer, typically 10 minutes to go through with Agent Mode. But where it's powerful is, let's say that you. Let's say you wanted to launch a new product, then what you could ask it to do in your prompt is you could ask it to go and review similar products and review forums, you could review feedback pages and so on. And what it will do is it'll go through and click through to different web pages and do research. It'll act like a human. If we were doing research, we'd often go and click on things. We might do a Google search, we click on a particular link and then we might want to click on sub links and work our way through and do things. That's essentially what Agent Mode does. And then when it's finished doing the end result of its analysis and its research, we can then specify that we get it to create a CSV file for us or PowerPoint presentation. There's a number of interesting things that we can create. And I'm going to do a complete session on Agent Mode next week. I wanted to mention it right now simply to make the point that some of the things we've talked about in terms of how we do prompt engineering in GPT5 are particularly important if we are going to use this Agent mode. So that's the two things I want to specify particularly are the. I think the second thing I mentioned where in our prompt we ask it to.

Come up with the steps it's going to go through in advance, think through what it's going to go through and tell us what the steps are. And also then the validation, the validation. Those two things are really powerful when it comes to crafting prompts using Agent mode.

Right? So we covered a load of stuff there and hopefully you're going to take something away from that. Hopefully something's useful there in terms of either those five tips. If nothing else, at least make sure you start structuring your prompts, because that's really powerful. Use Markdown, it's really powerful. So use some of those five tips. Even better still, create a template that you can use for your prompts, which gives you some guidance, some structure, and then consider using meta prompting where you perhaps feed an existing prompt of yours and the template and you ask it to improve the prompt so it works better for GPT5. And I think that's particularly important if you are going to create prompts to use the new Agent mode because it can do some really powerful things, but we just have to make sure we know how to craft those prompts. Okay. If you've found something valuable so far, then please do hit the like button. I'd really appreciate that. If you're not subscribed, please subscribe to the channel. That would be awesome. In terms of some other things coming up next week, I think it's a week on Wednesday the 17th, September 17th I'm doing a live stream on Agent mode also.

If you want to go deeper on this then I know many people here now in the AI Academy. The AI Academy is where I help accounting professionals, accounting firm owners, business owners, masterchat GPT. I try to make everything as practical as possible. We have a great community with people like Randall in it, which is awesome and you can ask questions, you can share your successes. The goal here is to help you with whatever you need to know about getting better results from AI systems and particular chat GPT. There is also a classroom where I keep adding new training videos for you. There's a whole course on prompt engineering and in the next few days there will be a complete chapter on how to use GPT5, how to craft your prompts with meta prompting templates and so much more that's in the prompt engineering course. Some of the other courses that I know people love is if you are an accountant in practice you must be aware of privacy and security. I have a detailed course where I tell you about how to make sure you are compliant with the various privacy laws like GDPR and so on and even how you could run a local large language model. Other things that you might want to check out is the course on agent mode if you want to take things to another level and get to do things for you or creating custom GPTs. And one of the GPTs that I've recently shared is how to answer emails really really quickly and so custom GPTs is one of my most popular training courses and then also projects which now projects you can use on the free account. Now this is my favorite feature in ChatGPT. Projects save me hours of time with with Chat GPT. I use them every single day. Projects are like folders, like smart folders. They allow you to organize your chats, but it goes deeper than that. They are smart folders, you can add project files and so much more. So I'm going to ask my team to put the link to tell you more about the AI Academy. When you go to the link, you can find out much more about what's in there in terms of the courses and so on. And also, most important, what people tell me they love about the AI Academy is you also get access to the prompt library software. It's a free bonus with dozens and dozens of prompts specifically for accounting firm owners, for accounting professional, for business owners covering things like financial analysis, business strategy, copywriting, marketing, and so much more. We've looked at a couple of them in this one today, the Metaprompter and the bank statement analysis, and there are many more prompts. And this software, you can also add your own prompts. It makes it really easy to find prompts and then copy and paste them into ChatGPT. It's a huge time saver. And why should you join the AI Academy now? Well, the reason for that is because in just a few weeks time we are putting the price up. At the moment we have a very special Pioneer price and if you join today, you get the crazy low Pioneer price and you get to lock in the price. When we put the price up in a few weeks time, it will not affect you, it'll only affect future members. So if you want to lock in and preserve the special Pioneer price, today's the best time to join the AI Academy. Okay, I'm going to answer questions now and there's loads of stuff going on in the chat, so just bear with me a second while I find where we got to with questions. I think the first question was from John, I believe. Yes. Okay, let's do what John says. I was listening to a webinar yesterday and the presenter said that from a security aspect, Copilot is safer than ChatGPT. He agreed, not as good, but confirmed we need to be careful as to security. Okay, I can't answer that because I've never used Copilot and I've never used it simply because all the things I've come across over the last year or so have been that it's not as good as ChatGPT, but I know it's getting better. I know Microsoft are putting a lot of effort into Copilot. So John, I can't answer because I don't use it. But if anybody else does use Copilot as and has researched it from a security point of view, please let John know. Please put in the comments if you have Any thoughts on that? And John, I know you're in the AI academy, so if you put that in the AI academy, I will do some research for you. Happy to do that.

Okay, next question, what's the next one? I'll also just give some shout outs to people on LinkedIn because I don't see them. So John is here from New York. Rashidat is here from Georgia.

Okay. Clive Stewart, I think that's referring to. He loves GPT, I think GPT5 question. If we contribute through your channel with your money, do we get. Okay, that's the B1 G1. I don't know the answer to that one. I don't know if Sarah, you know.

I don't know if B1G1 issued receipts or not, but I know there's many people here who are members of B1G1. Perhaps you could answer the question there. That would be awesome.

Okay, we've got loads of great comments from people like Randall Sunita. Oh, this isn't. Remember to put two capital Q's in front, otherwise I might miss the question. How do I access the AI library that was presented last time we were live? That awesome library. So Sunita, you need to be, I think you are, you need to be in the AI academy. And just so you know, let me just make this, let me just explain for anyone. When you are in the AI academy, when you've joined, go to the classroom and do the start, go, go to start here. And the step number two is how to register your free copy of the software. Just watch the video, there's a link below. You can register for the software. It's a separate piece of software. And then we need to just approve you in the back end, allow 24 hours and then Sarah will let you in. So Sunita, hopefully you can find that. Okay. And if not, just message the team, we'll sort you out. Right, we have seven minutes left for questions. How many more have we got?

Okay. Sunita also says. Oh now, now I know why it kept on asking questions. It was frustrating. I presume you're talking about GPT5 and asking questions. Yeah, but let me just mention that just triggered off a thought in my mind. Sometimes my mind goes off in crazy places. I'm like a squirrel. So I did notice something that I've not actually told anyone yet. I came across it by accident the other day and this is a new, I think a relative. I don't know when it came out. If anyone knows, let me know. But in ChatGPT and I am going to explore this myself. If you go into ChatGPT and you go into Settings and you go into.

Actually go back to there. If you go to Settings and go to customize ChatGPT, where you can put your custom instructions, which are really powerful. I love custom instructions. They are great. But one of the things they introduced that I only found myself is in the custom instructions. You have a section called what personality should ChatGPT have? And it defaults to default. But there's four others, Cynic, Robot, Listener and Nerd. So you might want to play with some of these because depending on which one you choose.

It has an impact on the way that it answers you as its default. So if you choose Cynic, then that's good. If, for example, you wanted to critique stuff, if let's say that you are often writing copy, like emails copy for your website and you want it to give you be critical, highly critical, then as a cynic, it will do that. If you select Robot and this might be for you, Sunita. It's efficient and blunt and so basically it'll just do what you want it to do because I find it frustrating and annoying. Thank you, John, for joining the Academy. I find it frustrating sometimes when I just wanted to give me the email or the report, but it then says, oh, that's a great question, let me go and work on that. And at the end it'll say, do you want me to also do this and this? Sometimes I just want the answer. And that's essentially what Robot would do. Listener is if you want to be more supportive, perhaps if you want to use it as a, you know, have a conversation, perhaps you wanted to. Perhaps you want to think through an issue and so on. And then Nerd is where it gives you more detailed answers effectively. So that's giving your chatgpt a personality. You might want to play around with it. I certainly intend to do that perhaps over the weekend. Right. What other questions have we got? Loads of comments. Again.

I'm looking for qqs. Morning, Jan. Great to see you.

Thank you for that comment, Sunita.

Welcome to Bob, who's over on LinkedIn as well. I'm still strong to see Qqs. Ronald says, Are there any handouts? No, there are no handouts. But this session is on my YouTube channel under the live Stream section. So if you want to revisit anything, you can always go back and revisit Ronald anytime. Okay. It won't be going away. Randall's just posted a comment about branching in chat. I'll read that with Interest. Shabir says love the AI Academy. If you're not in it, sign up Incredibly. Thank you Shabir. I really appreciate that. That's the link for it. I'm still looking for questions and I can't at the moment see any questions. We do have three minutes left, so if you're very quick then get your question in while I'm waiting for questions and let me just see other comments. So, Sybil, Hi there Sybil. Hope you're doing well. She's using Copilot and she's happy with it, so that might be useful. Shabir says Copilot is GDPR compliant, but there are setup considerations to help ensure it stays that way. That's awesome feedback. Thank you for that, Shabir. That's really helpful. Great to see Neil here. Randall says the AI Academy is well worth the price of admission. Thank you so much for that, Randall. I really appreciate it. Randall is very active in the Academy and very helpful posting great stuff. I'm still strong to see any questions and I'll just go back through the chat to see if I missed anything from earlier.

Otherwise we can wrap up in a bit.

Also, while I'm waiting for questions, let me know in the comments please. What has been the most useful thing today? Of all the things that we've covered, what's been the most useful? What's your key takeaway today? Let me know. Okay, a question has come in and it's if I'm a member of the VPA level 2, do I also have access to the AI Academy? No, it's a separate thing. The AI Academy is completely separate and perhaps if you want more information, Sarah can send some information out or you can reach out to her. Thank you so much for that feedback as well. I know we have a lot of people here who are in the AI Academy.

And the reason I set it up as a separate thing is because this space is moving so crazy fast. AI is changing at the moment. There's new stuff every day and so it's easy to feel overwhelmed. So my job is to help you manage the overwhelm by telling you what you do need to know. There's a lot of noise out there with new AI tools every single day. It can be really distracting and so I try to keep it as practical as possible.

So Michael says the most valuable thing for him Today is having ChatGPT update current prompts. So that's the metaprompting idea. I will be working more on some metaprompting stuff. It's really powerful to have a number of different types of metaprompting. There's one that OpenAI recommend themselves. I will be sharing that sometime very, very soon. And okay, Randall's after points. Okay. If you're in the AI academy, Randall likes getting points because he's number one on the leaderboard. You get points when you post good stuff and if people like your content. So Randall's stuff people love. Hence he's top of the leaderboard, right? I think we're pretty much out of time. I can't see any more questions. If you do think of a question afterwards, then best thing to do is if you're in the AI academy, that's where you can ask me questions and I'm more than happy if the questions for me when you post it into the AI academy, happy for you to tag me. That way I'll notice your question and I'll be able to answer it. I try to get on there every single day to answer questions and help you navigate this minefield of AI. So thank you. And there's another comment from Sybil around Copilot. Thank you so much for coming along today. Thank you for your amazing comments. I love that I'm going to play out with some music as I normally do because it gives me chance just to read your comments and if you are new, if this is your first time. I live in Portugal by the way and some people like the end of show music because they get to see an insight to behind the scenes of life in Portugal. Anyway, I will say goodbye for now and hopefully see you next week on the 17th when we look at agent mode and how you can use that. See you soon.