Top 7 ChatGPT Mistakes: The Pitfalls and How to Avoid Them

Okay, so what are we going to cover today? Let's start. We're going to look at what not to do with ChatGPT. What are the seven big mistakes? I will go through seven mistakes and show you how to avoid those mistakes. I'll make sure that you are using ChatGPT the right way. I've got some exciting stuff to show you. So I'm going to show you how to fix the seven mistakes and I'm also going to share with you some prompts because I keep getting told that people love the standard prompts that I create. I'm going to aim to show definitely one, probably two, one of which I use myself and I love. It's a really good prompt, really powerful. I think you'll have some fun with that particular prompt. So let's look at what are the mistakes people make with ChatGPT and how can we then fix them. The first one I think people make is they treat ChatGPT like Google. And the problem with that is it if you use it like you would use Google, you're going to get real surface level answers. They're not going to be great answers. What I always recommend, and this is I heard this when I first got into ChatGPT in early 2023. Someone once said treat it like your new hire or a junior team member and I think that's really good advice. I just treat it like another person. When you're delegating work, when you're asking your assistant, your junior team member, whatever to do something, you'll give a reasonable amount of information to get the answer back that you are looking for, which is a very different into very different to what you might put into traditionally a search engine like Google. So I always think of it as having a conversation with somebody else in the AI academy. A few weeks ago I posted a an interesting chat about or comment about what do people call their ChatGPT. Mine came up with NIA. It came up with some suggestions and that was the one I chose. So ChatGPT is called NIA. I think of it as NIA giving me advice and having a conversation with nia. So that my first tip, think of it as having a conversation with a human and that will then get you thinking different about how you communicate. The second mistake is using one sentence prompts. One sentence prompts are not going to be very useful for most of the serious stuff that you want to do. Yes, it's fine for a bit of fun. For example, if you said write a blog about tax planning, you will probably Get a fairly poor result. Garbage in equals garbage out. If you don't give it a great prompt, you're going to get something poor coming out. So when we craft our prompt, we want to add some structure, we want to add some specification. Who is the audience? Who are we aiming this response at? Is it going to be something that we're going to use with a client, for example, or is it just for our consumption? And so that might get us to think about what about the tone, what about the format? So this is about creating structure, creating some specificity, and this is what we learn in prompt engineering. And I'll talk a bit more about some of the ways that we can construct great prompts. The third mistake is to ask ChatGPT stuff without telling it who you are you need to create. The more information you give it. And we talk a lot about context. The more context you give, the better the results you'll get. Who are you? You'll find that when it understands you a bit more and what you are looking for, you'll start getting better results. Tell ChatGPT who you are. What is it you're looking to do? Who do you help? And one of the ways that we can do this is in a great feature that came out early on with ChatGPT, which was custom instructions. And so you really need to be using custom instructions. You find it in the settings where you can customize ChatGPT and the two primary boxes to tell a bit about you. And the other one is, how do you want it to respond? That particular one I find really important, because if you want to use ChatGPT to create written responses that you might use, for example, write a blog post, social media post, write an article, a report for clients, then one of the things that people often complain about is they say to me, mark, when I get chatgpt to write something like a blog post or an article, it's got better over the years. And certainly Claude is very good when it comes to more human like responses. And also the latest models of ChatGPT. 4.1 is a big improvement on 4.0. It's definitely getting better at creating more human like responses. Sponsors we can go further than that by making sure we tell ChatGPT about our writing style. How do you write? How do you structure your sentences? What's your tone of voice? Do you like short sentences? Do you want long sentences? Is there any particular word choice that you use? We can put this in the custom instructions. The great thing about custom instructions is you put it in there once and that will apply to all of your chats going forward. So make sure you put in your writing style into custom instructions. But also in the other main box you can put some information about you. For example, if you tell it who you help, who you serve. If for example, you're an accounting firm and you serve people in the medical industry, then that means when you want to write, let's say an article, a blog post, a social media post, it will know that information and therefore if you forget to tell it who is your target audience, it will automatically build that in anyway. So use custom Instructions to tell ChatGPT a bit about you, about your writing style. There are a couple of other boxes in there as well. You can tell it your name if you want. I did that. And now in a lot of the responses it says to me, hi Mark, that's a great question. And then it would go on, which I find a little bit irritating. It never used to do that. So sometimes in my prompts I tell it just to get straight to the point. The fourth thing, and this is a really big one, and that is wasting time writing prompts from scratch. Because I said to you earlier, mistake is one sentence. Prompts aren't very good. We have to create longer prompts. We have to make sure we have a specific structure, the writing style, the writing tone, and so on. But of course the downside of that is it takes more time when you are delegating a task to a junior team member. The more detail we give in terms of what we are looking for them to do, the the more likely we'll get the right response coming back. And it's just the same with ChatGPT. We do need to create that detail. It takes time writing prompts, so let's not waste that time writing from scratch. Your time's precious, you are busy. So let's make sure that we think about what are the standard things that we do regularly, what are the standard tasks, the things we want to do with ChatGPT and let's simplify the process, let's automate the process, let's create what I call some prompt templates. Let's create some pre built prompts that we can use over and over again. At this point, I think we'll talk about one of those prompts as an example and then get on to mistakes 5, 6 and 7. So let's imagine that you're an accounting firm owner and one of the things you might want to do for your own firm, but perhaps more likely for your clients is you want to come up with some ideas for your clients of how they could create some key predictive indicators. KPIs. Every business should have KPIs. So if we're not sure what KPIs to use, why don't we use Chat GPT to do that? Or different large language models? So everything I say today pretty much applies. If you prefer Claude, if you prefer Google's Gemini, whichever is your preferred. I usually talk about Chat GPT. It's the most common one. Unless you imagine for your business you want some ideas on some KPIs. Things that you should measure if you are the owner of an accounting firm, this is something you might do frequently because you might want to do this for clients. It could be that you do this for every client, either as part of a business advisory service or simply to add more value once a year or as part of your onboarding process. There's lots of ways that we could do this where we advise on KPIs. If you're doing this for clients, you don't want to rewrite the prompt from scratch every time you want a standard prompt. Where do you keep that prompt? My advice is you keep it inside a prompt library. Let me show you what I mean. A prompt library is essentially a database of prompts. Here is the one I use. Some people use Notion to store their prompts. Some people use Trello. We used to use Trello. Some use databases like airtable. But having been teaching prompt engineering for two and a half years now using prompts regularly, I did use Trello and there were a few things I found a bit frustrating. So I decided to create my own prompt library because I know the power of this in terms of the time it can save. This is something we are rolling out to our members of the Value Pricing Academy and the AI Academy. It's a prompt library. It's a way of managing your prompts. And so if you want to do a task, we're going to look at KPI. We're going to look at creating KPIs. I have one called KPI Creator. It happens to be second on the list. But I could have searched for this in a range of different ways. I could search on the type of prompt. It's a business advisory prompt. In this particular case, there are prompts here for marketing, for copywriting, for content creation, SOPs, ideation, and so on. We could filter down and find the prompts. I happen to favorite some that I'm going to show today. And so we find the prompt we want, and I want to create some KPIs. And so we click on it and when we go to the prompt we can see what the prompt does in the instructions at the top. Most important is the prompt itself. And my prompts are always very long. Why are they long? It's because the more detail you give, the more specificity, the more guide guardrails, the more you give examples and templates, the more predictable the answer, the better the answer that you will get. And so this is why we take our prompts, the things we do regularly, and we turn them into prompt templates that we can just reuse. Think of these as mini systems. So we have here a mini system that to come up with KPIs for any business. And so all we have to do to use it is really simple. We find it in our prompt library, the filters or search. We open it up, click Copy as markdown to clipboard. That's going to copy the markdown version to my clipboard. And then I can come out of this and go across to ChatGPT and I can do control voice, paste this in. And this by the way, is a prompt that nobody has seen yet because I wrote this one specially for today's session. If you are in the value pricing academy, you will not have seen this before. It's a pretty cool one. And I'll talk you through some of the things in the prompt in a second and how they overcome some of those mistakes we've talked about. But the first thing I'm going to do is go to the end because what I always do in my prompts to save time, to make it easier, all the prompts are designed so you don't have to do anything with them apart from add a few variables which I I always put at the end, so always do is put in a few details into. So I am just grabbing a fictional website that I can use and rather than me typing these in. So if I just. If you bear with me a second, I'm going to use a. We'll do KPIs for let's do an engineering business. So all I need to do is grab a website of a business. This happens to be a UK engineering business called Saxon Engineering and the industry sector, well, that's fairly obvious. It's engineering. Okay. So we can. That's all we need to put in. So there's a few things that I won't go through everything here, but I want to pick up on a few things. So it starts off with the role what role do we want chatbee to play? And it says you are a performance analyst with expertise in identifying and evaluating key performance indicators for businesses. We know that when we give ChatGPT or any large language model a role to play, we tend to get a better answer, the sort of answer that in this case we would expect from a performance analyst. Every prompt I do has a role that I give chatgpt. Then we have the task. And the task is to analyze the industry sector and identify the most appropriate KPIs for client name, client website, and present them in a table format followed by detailed explanations for KPI. It's very clear there. It's got the variables embedded in it, which means you'll find them in many places in the prompt. So rather than having to go through the prompt every time and put in a new client name, it knows that's a variable we only have to put it in once, at the very end. That's the power of variables. It makes it easier to use your prompts. We have a task in prompts. I break it down into action steps to be really clear. The more clarity you give to the large language model, the better and more predictable the response. So I want it to identify the key KPIs first, and then I want it to create a table specifying exactly what I want to be in the table. I want it to have the name of the KPI, how it's calculated, what it measures, why it's important, and tips on how to measure it. To make it really clear, I include an example or a template of the layout I am looking for when we're constructing prompts. I found that giving it templates, giving it examples, are two of the most powerful things that you can do if you want to get some consistency in the type of output. Particularly important if, for example, you've got a prompt that creates standard reports for clients, perhaps it's end of year financial review. It could be a report based on tax planning. Then you want to make sure that in your prompt you've been very specific about how you want it to lay out the report. What headings do you want? What do you want to be in the report? Because then you will have some consistency as to what you're going to get. All of this is just giving it the specific response I'm looking for. I want a table with six headings and detailed KPI descriptions below that. And then just to create some additional clarity, I'm just telling it instructions for use. Replace the variables in curly brackets with the relevant information for each client or industry, specificity is always powerful. Fill out the KPI table with the most appropriate metrics for the client's business and expand the detailed KPI descriptions to give extra clarity, context and actual tips. And then I've got some constraints and goals. In this case, I'm telling it to ensure the KPIs are relevant and also to avoid including irrelevant or obscure KPIs that do not provide meaningful insights. So that's the prompt. Run it. Let's see what happens. If we think about how long this is going to take, it took probably 10, 20 seconds to locate it in the prompt library, perhaps 10 seconds to hit that copy to clipboard and paste the longest bit will probably be putting in the variables, which is simply a client name. Copying and pasting a website and then an industry so far probably takes a minute. And then I hit this button here and ChatGPT usually takes matter of seconds. In less than two minutes you can generate a really powerful report of ideas of KPIs for any business if you're working with clients, their business as well. So if we look at this, we've got the tailored KPI dashboard for Saxon Engineering. Based on the needs and benchmarks of the engineering sector, these KPIs help track productivity, profitability and operational efficiency. Operational all critical to engineering firms. So what's it suggesting? Project gross Margin? It's suggesting and it's telling us how we'd measure that. Why it's important to show whether projects are priced and delivered profitably. Tips on measurement, use job costing, tools to allocate labour and materials. We've got engineering utilization rate on time, delivery rate, rework rate, quote to win ratio. That's an important one for any business that does quotes, revenue per engineer, change order frequency, client satisfaction score, bid accuracy and cash conversion cycle. So we have this wonderful little table with a whole bunch of really powerful KPIs specific for engineering firm owners. If you're the owner of an accounting firm wanting to add some value or doing advisory work, you then print this out and then go through this to the client to see which ones they are measuring, which ones they're not and the ones they're not. Do they agree they should be measured? If so, you could offer to help them measure those things. Below that we've got more detailed KPI descriptions for them. So take Engineering Utilization Rate. It says what? It measures the proportion of time engineers spend on billable tasks versus total bill of available hours. Why it's important. High utilization means you're making the most of your engineering team. Low rates suggest overstaffing or or inefficient workflows and tips on how to measure Require team members to log time daily, categorize time as billable versus non billable to calculate accurate utilization. For all of those KPIs, we've got more detail as well to help you with measuring them. Within two minutes you've got something really powerful. Hopefully you can start seeing the importance of why one sentence prompts are not going to give you this. You would not get this from a one sentence prompt. You would not get this from a searching on Google in the traditional way, this kind of analysis. So we have to just make sure that we know how to craft our prompts. Let's move on to mistake number five and mistake number five is ignoring ChatGPT's creative strengths. Some people use it for very limited uses, perhaps just admin tasks, but there is so much that we can do with large language models. I use it every single day, my team uses every single day, and I have it on all the time. One of my members of the academy had not thought of having it open on multiple browsers at multiple times. I typically have it open at least three different places. I have two different computers and I'll have it open on Chrome, perhaps on my desktop. I might have it open on Safari because I am regularly using it and multitasking with it. I might be getting it to do one thing on one computer and something else on another computer because some things take a little bit of time. If I'm doing deep research creating images, I'm often using it for different things. We should not be ignoring its creative strengths because it can do so much than just basic answering questions or admin tasks. I think its real power is ideation and brainstorming. If you ever want ideas for stuff like KPIs, if you don't know which ones to use, then it's great for doing that. So you I recommend that what you do is get into the habit of using it more and exploring all the different ways it can help you in your business. Things like ideation and brainstorming, analyzing financial data, creating your marketing material. In the last few months they brought out the brand new AI image generation which is so much better than the old Dall E3. There's some great stuff we can do with images once we figured out what are the right prompts to do. Mistake number six is not reviewing or refining the output if you are using it to create content to send to customers to send to clients, then please do not send the output straight to a client or to a customer. There's a phrase that's being increasingly used which is human in the loop. And what that means is that we should be treating AI as a tool to accelerate and not replace us. And so we want to make sure that we are using as a tool. If, for example, you were to create a report for a client, make sure you read it first and use your professional judgment. I was using it just a couple of days ago to create a complete workflow action plan for valuing IP in a business and it did a really good job tax and valuing IP. I was thinking, I'm not 100% sure. I would then go back and ask a more detailed question about that bit of advice and I would specifically ask it to use its search facility and cite the sources. In two instances the other day, it wasn't necessarily wrong, but there was a better route. Once I'd steered in the right direction, it then said, yes, you're right, Mark, this would be a better way of doing it. So please do not just blindly rely on ChatGPT. Make sure you review the outputs, refine the outputs. That's really important. AI is not going to replace us just yet. It's a tool to be used, a very powerful tool, but nevertheless, there's some things it can't do. It can't replace us just yet, which is good news. Number seven is breaching client confidentiality. So we have to be really careful about how we use it. We have to be aware of the privacy rules, security rules, privacy and security in particular. Many here are owners of accounting firms. You need to be aware of the rules and what you can do and what you can't do, what you can put into ChatGPT and, and what you have to be careful about with personal information. And so I strongly recommend that you have an AI policy and you should definitely get clients permission. Otherwise many accounting firms are unwittingly falling on the wrong side of the regulation. So let's just be careful about AI. It's an extremely powerful tool. We just have to make sure that we are using it in the right way. I'm going to look at one more prompt to illustrate some of the things that we've talked about. And it's one of my favorites. It's a really powerful prompt that I've used a few, few times now. And the prompt is this a legal agreement to sign. And when I look at legal agreements, I find them so hard to read. My eyes start to get really heavy as I'm reading a legal agreement. But I also know that we should read things and not just blindly sign them. They could be things like non disclosure agreements, it could be a rental contract, it could be a business contract deal. There are so many different legal agreements that you might come across both your personal and business life. What do you do when you get an agreement? Do you read it word for word and analyze every word? Which I don't, I have to confess because they just bore me completely. And I also find the way they're worded confusing. I really struggle to follow sometimes what the clauses are talking about. So one of the things that ChatGPT is really good at is helping us understand legal agreements. And this is a huge time saver. So let me show you what it is. Let me go to the prompt library. I believe I just need to change something. So we're in the prompt library right now. So I'm going to go to legal review and execution. If I copy it to Clipboard, which is the first step I go across to ChatGPT. I'm going to create a brand new chat. I'm going to put it into here. And then there's one variable. It's really simple. There's two things I need to do. One is the advising party. This could be you or you are advising a client. Perhaps a client said to you that they are thinking of renting some new premises and they've got this complicated legal agreement and so they want you to, the accountant look at it. You just put in who is the advising party. Because what ChatGPT is going to do with this prompt is advise one of the parties. So this is the tenant in a legal agreement. And the other thing I have to do is feed it the legal agreement. So I just need to grab that. It was on my desktop, nice and easy to find. I have a commercial lease agreement which I am just dragging into ChatGPT. So you should see that this is a lease agreement with loads of clauses. I think it's about a 12 page document. Standard lease agreements. We have the role and that is you are a legal expert specializing in contract law with a focus on ensuring that all legal documents are thorough, clear and properly executed. So we have this role first. We wanted to play the role of a legal expert because we know that we'll get a better response as a result of that. The task is to conduct a comprehensive review of the attached legal document, ensuring that all amendments, addenda and appendices are included and correctly referenced, verifying signatures and execution and documenting any verbal agreements. Important note, you are acting for advising party. You must conduct your review as advice for advising parties. So in this case I just put at the end, the tenant could be that you put in your client's name, whoever you are advising, and that would be you. If it's from the personal point of view, you perhaps had something to sign as a document and you just want to make sure that you're okay to sign it. So you might put me or whatever it says as you in the agreement for clarity. So then we have action steps. Because one of the things that's good with prompts is you can create better results if something's a more complex task. Break it down into actions. Chat GPT will follow the actions. So I want it to firstly summarize it in simple plain English so it's easy to understand to a layperson like me who doesn't understand these agreements. And then I wanted to list any potentially onerous clauses that may adversely affect the advising party. So I don't care about onerous clauses for the other party. I'm just advising the tenant. And items that are vague that the advising party should seek clarification on from the other party. And unfair clauses. Consider each item of the agreement elements for each, where appropriate, suggest a fairer or clearer clause or section the advising party could propose by way of amendment and then conclude with other advice considering each item of the further advice. As you can see, we're being really clear here. You can see that advising party is used several times in this prompt. And what we don't want to have to do when we use these prompt templates is go and fill every single place. That's why we use variables. It saves us huge amounts of time. In this case, I've got the agreement elements. So I'm being very specific here as to exactly what I want ChatGPT to look at in the contract. I want it to make sure it understands the parties that are involved. I wanted to review the purpose and the scope. I wanted to examine key terms and definitions, check the obligations and responsibilities. I want to assess the payment terms, identify potential risks and liabilities, check for termination clauses. We should always look at the termination clause in an agreement so we know how we can get out of it. We want to review confidentiality and non disclosure. We want to examine dispute resolution mechanisms, evaluate any non compete or non solicitation clauses, assess intellectual property rights, look for automatic renewal or evergreen clauses, review governing law clauses, consider any penalties or damages clauses, check for compliance with relevant Laws seek clarification on ambiguous language. So these are all things that if you're reviewing a legal document, you should consider. And so we want to make sure that in our prompt we are being really specific. We are telling ChatGPT we want to look at all of these different areas. And then we've also got a section on further advice. We want it to review amendments and appendices, seek expert opinion, verify signatures, document verbal agreements and so on. And then we have the variables. So it's a long prompt. Why? Because we're going to get a much, much better answer. So let's see what it does. Let's hit this button here. It's going to analyze this commercial lease agreement. It might take a few seconds because it has to read. It was a really long document, but it read it pretty fast. I would have probably spent an hour reading it and still wouldn't have understood it. So now it's telling us a plain English summary. This lease allows the tenant to occupy and use a property for warehouse and power generation. It started on December 1, 2013 and was set to end in 2020. However, the tenant has the right to renew it and so on. I can understand that. Then what it's doing is going through critical issues and suggested changes. So we've got potentially onerous clauses. I think this is US based one. I don't understand all of this, but real estate taxes. The tenant must pay all property taxes, assessments. Then suggest a suggestion, ask for a cap or cost sharing. So it's picking up on all sorts of things we should look for that could be potentially onerous with suggestions on what to do. For example, late rent fee, a 5% penalty after just 10 days. Suggestion, request a grace period extension to 15 days or reduce the fee. If we're still negotiating, it's got vague or ambiguous provisions. So areas where we might want to seek some clarification. Landlord controls, insurance, trust fund and approval of architect. Clarify by saying the tenant should have equal say in selecting the architect and managing proceeds. We have unfair or imbalance clauses. We've got a number of those options. Purchase, assignment, landlords, title warranties and governing laws. Further advice and checklist, review of the things we should consider seeking further advice and so on. Then we've got a summary. I think you'll agree that's a pretty good review of this particular legal agreement, which is very long and it just saves me so much time because I can't be bothered to read these things. In fact, I've just been sent one today. It's only a non disclosure agreement I've been sent. They're usually fairly straightforward. I've never come across one yet with onerous clauses. But I'm going to run it through this and get Chat GPT to review it as my lawyer and tell me if there's anything that I should think about before I sign this NDA. I'm hoping now that you can see the power of creating better prompts and also of having a prompt library. If you found something valuable in this session, please let me know by giving it a thumbs up. I would really appreciate that. Also, please let me know in the comments what's the best thing you've learned today. If there's anything else that you specifically would like me to teach, let me know in the comments and I will see you all soon. Bye for now.