S1. EP 2 ChatGPT Prompting Techniques for Business Owners

Katie: Welcome to the AI Automations for Business Podcasts. We're your host, Katie and Noel, and we'll be discussing how you can use AI for your business, along with the latest news, updates and automations to help you stay ahead of the curve. Make them running your business more effective and enjoyable. Please be sure to

Noel: subscribe and I really hope you enjoy this episode.

Katie: Hello, welcome back to another episode and today we are talking about prompt engineering for chat GPT so you can create content that actually sounds like you and Noel I know that you are a whiz at prompt engineering yeah

Noel: I've been using prompt well I've been working on prompt engineering for quite some time now I

Katie: feel like it's been and I feel like you were doing prompt engineering before people actually really knew what that was yeah

Noel: I think so yeah I when chat chip chiptt first came out that's when I first started using using it obviously and asking questions and I was like, this is one amazing. It was such a big technological leap. But then at the same time, it was all a bit samey and a bit, you know, it was not the best responses. But then I wasn't giving it the best instructions. So that's kind of how I was like, well, I need to figure this out. What I would say, like prompt engineering takes a lot of practice. So it's not like something that we can pick up and run with. You kind of need to train your brain to do prompt engineering. But anyone can do it for a bit of practice.

Katie: So I see everywhere on Instagram. So obviously my Explore page, when I go on to Instagram, is all fellow marketers and fellow, like, social media people. And I see a lot of posts about saying, like, this is the best prompt to use or use this prompt to get, like, X result. And I also, on the flip side, get emails from people. I probably get at least one of these emails every week and it says, I'm just going to kind of paraphrase because I don't want to. I'm not saying that someone's wrong. I just feel like sometimes we aren't given all of the information. I feel like sometimes people can use these scare tactics and they're not actually giving you all the information. So this one is about business owners using chat GPT to generate hooks for social media captions. and they go on to say, like, how do I know you're using chat chit? Now, for someone like me and you, we can pick up when someone's using chat jubit for, let's carry on with the social media caption example, because they haven't used prompts in a way to get rid of those telltale signs of chat chabit and those structures that just the basic chat GPT, you know, would have. So, yeah, and this person then goes on to say, like, you know, it's not very good for creating hooks, but actually I find that it is because of the prompts that I'm giving it. So then this person then goes on to say, you know, instead of using chat GPT, you know, to buy their program or digital download, whatever it is. But I'm pretty sure because there's hundreds and hundreds and hundreds of, they say, like, hooks, content, templates, that that's probably been generated by AI or chat GPT, Claude. So I'm kind of like almost fed up of people saying, you know, chat GPT works. It really does, but it just means that you're not. And I don't want to even say using it properly. It's just not, you're not giving it enough information when you're putting in the prompt. Absolutely, yeah.

Noel: From a social media perspective, if I came to you and said, could you create me a LinkedIn post for my electrical design company that focuses on, you know, the design tools I use as an example. You would go away, you would create the post, and you were sending it back to me and I'd look at it and go, well, that sounds nothing like me.

Katie: Yeah.

Noel: It sounds like you. Yeah,

Katie: because I don't have, if you're just giving me that information, that for me is not enough information for me to go, okay, I know exactly what to write and the tone, voice, like, everything. Yes, I could create some sort of caption with a, you know, a great hook, but whether it sounds like you or not, it's highly unlikely because I haven't asked or you haven't given me all the information that I require. Absolutely.

Noel: But if I give you like five or ten examples of like these are the ones that have worked really, really well for me, you know, you could then look at those and then get, right, well, if that worked well, I can repurpose it to do this, you know, but you kind of, it works all the same with AI. If you don't give it any examples or any sort of, you know, detailed instruction, then you're just going to get the basic stuff out of it. You're never going to get the good stuff. And you're just going to sound like everybody else. This is using it, whereas you could be clever. And what I would always do is give it, give it the, examples. I would also make, you know, really hammer home the tone of voice and say, well, look, this is my target audience. This is kind of like what I'm trying to achieve. And you've kind of got to, the way I look at prompt engineering is I've got to talk to it as if I'm talking to somebody on like Upwork or Fiverr that's never met me before. And so this is what I want. And I know if I don't give enough instructions, I'm just going to get rubbish back. So the more I give it, the better the output will be for me. Yeah.

Katie: Real key. Yeah. So really like you're saying to like part of your prompt should be, for example, like I want to create an Instagram caption on whatever you want to create the caption on. My ideal clients are and then give it as much detail as possible. This post like and then maybe like the aim of the post. Yeah,

Noel: definitely. Yeah.

Katie: And then. like what action you might want someone to take from reading the post.

Noel: Yeah, and you could also put it in a bit of variety. So if you're giving it the examples, you could also say, well, these are the kind of structures that I like to follow. So I'd like to start off with a strong hook at the start. You know, I then like to do like a little bit of this in the middle and a call to action and maybe something else after. Or you've got different structures for different types of posts that you're putting out. So if you give it that as well, you know, put it in there and say, well, you could then just give it a general instruction, like give me a marketing post about X, Y, Z and it would go, marketing, this person likes this structure, they've given me these examples, example two and three in that list, match. I'll then structure it in that sort of way and give you that sort of post. Yeah,

Katie: rather than just putting in, I'm a fitness instructor and I've got a new, can you write me an Instagram caption on my new class for Pilates?

Noel: Yeah, they'd have no idea about, you know what Pilates is. You know what Jim is, but it doesn't know you. It doesn't know your business.

Katie: Yeah, and that's when I feel like then that's where other business owners are getting frustrated and thinking that AI isn't, you know, going to help them in their business.

Noel: Yes, absolutely. Yeah.

Katie: So I always as well give it my like brand voice. So I might say, you know, this. needs to be in like a it could be like a friendly and polite tone you might want to go for sassy and spicy it might be professional and corporate i mean not me ever but someone might so yeah it's like that's really important as well because then that's how you train it to sound like you

Noel: yeah absolutely yeah you get

Katie: rid of those like familiar chat you structures that are really obvious when people haven't put in the right prompts. Yes.

Noel: And also like the telltale signs of AI with their double emojis and things that it's like you've then got to instruct it and say, well, look, these are all the things that I don't want you to do. You've got as good as giving examples of what it should do, you've got to give it, you know, instructions of what it shouldn't do. I find those like, they're like the superpower of a prompt. Yeah. You've really, that's the point where you can really hammer home and tailor the output. So it might give you an output. It's like, no, that's not really. It's 80% there. But if I added in this extra constraint into my prompt, that would then get rid of that bit. I'll give that another go. And then, you know, you kind of iterate on it as you keep going. But yeah, you kind of got copy, paste that. And don't forget where you've left it in your chat history. So you can use it again.

Katie: Yeah, absolutely. Yeah. And we were having a look, weren't we at a app. I won't mention it because I don't want to get sued. But there was an app and it said that it could like generate hooks, keywords, hashtags and captions for you. And so you were able to have like a free trial with it. So I downloaded it and like you had to put in like your, I think was it like can you remember what it was? It was just kind of like I think I just put in like content. marketing, didn't I? And then it gave me some options of hooks that I could then turn into like a caption. And they were so like basic in terms of what the prompt must have been. It was all the telltale sign. Literally, I tried it a few times using like different words and also different industries like in nine out of ten of the prompts that had given me it used and here's the kicker

Noel: yes it did yeah it loves that say does it

Katie: loves that and then i feel like as well if you're using chat jpt for say like a blog it will always have in conclusion yes or in or just um to summarize it always has like those key phrase that you need to look out for Or it says, but what else does it sometimes come up with? It's like, but here's the thing or something like that. Just watch out for things like that as well.

Noel: Yeah, because I think the thing with that app is it could be incredible.

Katie: Yeah.

Noel: It won't take a lot to do it. Like there just needs to be a box where you could say, well, here's my brand voice. And then use that to then create the hooks. But yeah, they need to do some work on some prompt engineering. It's a great idea. But, yeah, it was like, hmm. Some people might find that incredible, but yeah.

Katie: Yeah, I just feel like you could, with apps like that, they're really great because like they really pull you in because they can help me with hooks and hashtags and keywords and, you know, things like that. And, you know, people really want the help with that. But then the result is so disappointing because actually you could do that within ChatGBT, BT for free just using the right prompts.

Noel: Yeah, yeah. And it gives you the freedom as well in chat chvety to actually look at it and go, that's rubbish. I'll change the prompt to do this with apps like that. You've got no option other than to re-roll and try again. Yeah,

Katie: yeah. And the thing is with an app like that, you can't give it more information.

Noel: Whereas with chat chepti,

Katie: you can go, no, this is not what I wanted. You know, I and then like you can add in more information. information or more details.

Noel: But I think one thing, if you're like really serious into like creating, you know, or using outputs from chat GPT, then I would definitely suggest like getting the plus okay. So then you can create your own custom GPT. So you could input that prompt that you've created within a normal chat and then have that within your own custom GPT. And then you don't then have to then worry about it ever again. You can always add to it. You can always add examples. So if one, post one day went absolutely crazy. You could go, right, that needs to go in the GPT and you copy-paste it in and it sits there. And then all you need to do is just open up your GPT and then just say I would like a post about this and then there it goes straight off.

Katie: Would you kind of give it more information rather than just copy and pasting the caption in? So would you say this post performed well and it was, I don't know, like a storytelling piece of content? Like would you give it that much information or you

Noel: can do and there's nothing wrong in doing it like for me i usually don't i would just say well look this is this is good a known good so use things you know in that sort of area and then yeah just use the prompt then to constrain it and make sure that gives you the same sort of stuff because there's also another good one that most people forget about especially as english people we forget about it is the most people would then just copy paste it and then not realize that things like revolutionise may have a Zed in it. Yeah,

Katie: specialised. Yeah.

Noel: So, you know, you've then got to make sure that within your constraints that thou shall only ever output UK English or whatever, you know, whatever different dialect or whatever. Yeah. You've got to really hammer that home because you could have, you always spot like a UK company and you go, well, they haven't written that. Yeah.

Katie: And also one to look out for as well is the M dashes. So it's the little dashes. Now, not a lot of people, more so I think in the UK, don't naturally use m-dashes in their writing. And I feel like m-dashes are like for people who are doing like a lot of more professional writing, I want to say. I mean, don't come at me at this one because obviously I don't know, like, how they are taught, like, in other countries. But I don't ever remember being taught m-dashes at school.

Noel: No, no, I don't either. If I did, I was, my mind was on something else.

Katie: Yeah, I was off six that day. I was sciving clearly that day. But I feel like, you know, they would have been taught for more than, like, one lesson.

Noel: Yeah, yeah, they did.

Katie: Yeah, so just be careful as well. They're m-dashes. They're not hyphens. They are used in a different way. And I feel like that is a very telltale sign if that's not how you write naturally. Absolutely, yeah.

Noel: But if you've given it examples, it should already know that, depending on how many, if you only give it one example, that it's never really going to understand it. But if you've given it 10 and you've never used it, nine out of 10, it would never think about it because it knows how you structure. And I guess another good thing, we're talking about the voice. So there is a bit of a cheap, code that you can use without having to really go crazy on building a prompt. And that is to ask to increase burstiness and perplexity within the writing. So the way that the AI works, it's in a structured way. And it basically guesses the next word or what it should be based on what's come before it, whereas we don't speak like that. You know, we're not robots. So when you're asking for those burstiness and perplexity, it's making it sound as if it's, you know, a human, so it uses human speech patterns within the output, which is really quite clever. But yeah, not many people are you ready to use that.

Katie: No, I feel like that was a real gem of a tip that you have just given us.

Noel: Well, drop it in there.

Katie: Yeah, and like you just totally went in and dropped that in, like it was no big thing. And I feel like that's going to make a huge difference.

Noel: I've been doing it for years.

Katie: Yeah, thanks. Thanks for passing that on to me. I'm just finding out the same time as you guys listening. There's no perks.

Noel: Definitely not.

Katie: No.

Noel: Yeah, you could have that one. Okay, thanks.

Katie: Any other amazing tips that you want to share, or do you feel like we've covered most of what people need to know to get started?

Noel: Yeah, I think that covers the basics. One thing you can also do is that sort of quite helpful is using markdown language within the prompt. So for those that I haven't seen it before, you can use like one hash would be like the H1 tag for like a title. So, and then you can then have double hash would be the, you know, the second level of a heading. Yes, yeah, the hashtag, yeah. So you could have, yeah, the first one would be like a big title, then you can go down to like four. And basically, if you include that, it would then say, look at that title. And they say, that could be like constraints. And that would be like a number one. And then it go, ah, well, that's important. I need to know about this. And then it knows that everything else underneath that is part of that constraint until it hits the next title, which could be, you know, the examples, they go, well, now I understand that, the next bit. So it's all kind of, you've kind of got to build it in building blocks. You know, you could type out a prompt, you know, in just plain text, in paragraph format, you know, with the ramblings of a madman, you know, and it's like, well, it's great. You've given it all the information, but you haven't, you haven't quite structured it. So I would say, you know, how to structure prompts is usually really good. So I, for my prompts, I always start off with a role and then do goals, have constraint, samples, and then. example outputs. And I work through that each time I build like a, especially on an automation prompt. That's the kind of structure I follow. But you could use that anyway. There's another top tip.

Katie: And also we were mentioning about, you know, putting in captions because we were using captions as an example, you know, so if there's captions that you've written and they sound like you to add those in. But it's really important as well. Like if you want to use chat GPT for hooks, then to do the same. So any hooks that you've written and they're like, yes, this is the structure that I want for my hooks to add those in as well. Otherwise, you are going to get the very like standard structure of hooks. And so it's a couple of words, then a colon followed by like more words. And that's when I feel like that's a very telltale sign as well,

Noel: because I

Katie: feel like not a lot of people naturally would write like that.

Noel: No, no.

Katie: Um, that's very. Yeah, so as well as captions, if you want it to create those hooks, then just give it examples of the hooks that you've written and then it will understand your structure and the way that you write them.

Noel: Absolutely, yeah. Yeah. More you give it, the better you'll get.

Katie: Well, thank you so much for listening to this podcast episode. I hope you liked it. Please do like and subscribe and leave a review if you feel called to. and if you've got any questions about prompt engineering that you would like us to answer on future episodes then just send them over and we will try our very best to answer those for you and we will catch you next time.