Build an AI Writing Bot That SOUNDS LIKE YOU | John Tighe

What if you could train an AI to write exactly like you? Not generic AI slop, but actually content that sounds like you wrote it yourself.

Today, I've got John Ty with me, a nine-time best-selling author, and he's going to show us how to build your own AI writing bot that captures your authentic voice. So, by the end of this video, you're going to have a system that writes and edits in your voice so you can create way more content without losing what makes you you.

Hi, John. How are you?

Hi, good to be here. Thanks for having me. I'm I'm very well, thank you.

It's so nice to have you here. We are so excited to have you and just to hear from you. Someone who's literally a best-selling author is the sort of person who's going to be very particular about the way people are writing and expressing themselves, the words they use, and whether it sounds authentic. So, I'm super excited. What are we going to be diving into today?

Well, most people are using AI to write and and we we get this we call it slob. Now, there's a an actual word for it, but what it means is this generic AI stuff which doesn't sound like you. And so, the idea here is to uh train the AI in your actual voice. You know, what your preferences are, how you actually write, so that when you're working with the AI, it actually sounds like you. I mean, it really is as simple as that.

And what I try to say to people is in a perfect world, you don't necessarily get the AI to write everything for you even once you've trained it. Ideally, you would iterate the first draft and have the AI improve and polish it for you. But sometimes you're in a hurry. Sometimes the piece that you're writing is not quite so critical and it's okay to let the AI AI do its thing. Uh, and then you just go through and do a quick polish, quick edit. But you don't want the AI doing its thing if it's not going to sound anything like you because it's quite easy to spot. Now, AI has these tells which people have learned to recognize. And if you're sending stuff out that is written clearly written by AI, it does nothing for the trust in you. It does nothing for your credibility. And it and it it throws you in the mix with everybody else. there's nothing to distinguish your content from what they're getting from all the other people that are sending out AI slot.

So, it's really um whether you're doing it to write, you know, a book project. I work with a lot of non-fiction authors to help them write books to position themselves to generate leads, that kind of stuff. Whether you're doing content for a blog or a long form social media post, whatever it is you're doing, you want to take some relatively straightforward steps to train the AI in your voice so that otherwise it's just best guessing it. And AI is the the generic aspect of AI. It's not a bug. It's a feature. It's just the large language models are kind of taking an average of all the training that they have and giving you an average back. So it by definition it's going to be this generic AI sounding content.

Yeah, that is exactly it. It is definitely you can really notice when it's AI slop as opposed to something that actually sounds like the human it's supposed to be coming from. So how do we train AI to do this? And can you like show us like how do we build this bot? I'm super excited about it.

Yeah. So the Okay, so the core this there's a few stages. Um, stage one, I'll give you the high level overview, then we can go into more detail and and loop back if we need to, if anything's not clear. So, the first thing we want to do is give it a some samples of our writing. We need to the AI needs to be able to extract, you know, the essence of how we we write. So, that's step one.

Uh, step two is we want to give it some context. Again, if if we don't give it the context, it just has to guess. And the context is there are three key uh context layers I call them. One is a little bit about you as the author relevant to that writing project. Um the second one is the a description of the target reader. You know your ideal client if you're trying to generate leads, your target reader as the case may be. So kind of building out an avatar of of them. And the third one is the objectives that you have for that book or that blog post. etc. You know, what do you want to achieve with it and what do you want your reader to take away from it as well? What are they hoping to get?

Uh, and that so now what we've done with the by creating the the kind of voice guide, the style guide is it it's called different things, but it's that's what it is. You're telling the AI kind of how to write with the context layers, you're giving it its role and the purpose, who it's writing to, who the audience is, what the purpose is.

Um, and then the third element is really I call it just additional instructions. It's the things that that are important but aren't caught by the first two. So things like humanizing it. A lot of the humanizing actually is going to be covered through the style guide and through the context layers. So again, it's going to be much less like the AI generic stuff because you've done that. But you can also give specific instructions on definite dos, definite don'ts, to the extent they're not already picked up. And um things like, you know, what reading age grade are you targeting things, anything that's not already picked up elsewhere.

And once we've gone through these three key phases, now the AI's got a really strong idea of, you know, who we are, how we write, what the purpose is. And is it going to be perfect straight out of the gate? No, it's it's not. And what I say to people is don't get hung up and paralyzed on trying to get it perfect. You won't best guess these things. Um, the AI can give you a lot of information. It can pull apart your writing and give you really good feedback. Um, so take advantage of the AI, guess the things, you know, best guess the things you're not sure about, kind of brainstorm. If you want to help help with the customer avatar, you're not sure. AI can interview on that. So AI can be really helpful there. It won't be perfect. When you get the output or when your stuff is being edited, you'll see if there's something that's just off and then you can go in and fix that. And you can and again there's a a way of doing that working with the AI. So uh don't think perfect straight out of the gate. Try and just build a a prototype quickly. Use it, test it, and then you can fine-tune it. And over a fairly short period of time, you'll find that, you know, there's less and less that it's getting wrong. It's it's it's um it starts to understand you because you've iron out all those little um things where you just haven't quite given it the right instructions first time.

Okay. So, how long is this document? Firstly,

let me just give a little bit of background first rather before we dive into the um you know the the word counts and and the the stuff. So, there are there are different ways to do this. So, the the sort of quick and dirty way, if you like, is to just take a sample of your writing um and put that into the AI and ask it to create a style guide. I'll show you the prompt for the style guide uh in a moment. And what I can do is is um give a link to the PDF so people can download that because there it's pretty long. There's no there's zero chance of of sort of being able to scribble it all out or even copy and past uh you know, screenshot it.

Um, so the quick and dirty one is just to take a sample of writing. Let's say you write a newsletter. You you might take um depending on the length of the newsletter, you might take three or four articles that you've written in the past that you're happy with that are really good reflection of your voice and feed those in and then ask AI to kind of extract your your essence in terms of uh your writing. The better way to do it is to recognize that actually we we write and talk slightly differently depending on what mode we're in. You know, are we in storytelling mode? Are we in step-by-step how-to mode? Are we describing something, etc. Are we coaching somebody where we're trying to uh encourage and nurture them? They're all a little bit different. So, if you're willing to spend a little bit of time up front um pulling out some different samples that reflect the different ways that you might write in your newsletter or in the book that you're working on or in that long form post, you're going to get much better results. And the nice thing about this is once you've done it once, it's kind of set up. You might you might go back if your writing style evolves, you might go back in in three or six months and and and tweak a little bit. But basically, you're setting it up once. So, if it sounds like a chunk of work at the beginning, it's not so bad. The AI does the heavy lifting for you anyway. And once it's done, it's set up.

So, you know, find an example of where you're telling, let's say, a story, maybe a personal story. Uh, find an example of where you're talking about a case study, maybe a client case study, something that you've you've done with a client. Find a story where you're giving, let's say, there's a three-step process and you're explaining that three-step process. How do you do that? Maybe there's a time when you were running a a nurture email and trying to coach people. So find those different things and then we're going to feed those into AI.

And what I'm talking about here, this is worth mentioning here. It's kind of model agnostic. I get asked all the time, you know, what's the best model to use? I've used focus mainly on chat GPT in the past. I'm using Claude more now. A lot of writers like Claude. Sometimes I'll run stuff through two or three different models and just select the best output or mix and match. But um whether you're using Claude or Gemini or uh Chad Gypt or something else, the building blocks for this are the same, which makes sense if you think about it. The large language models work in a similar way. We need to give it the same kind of foundational stuff, the same instructions.

So, uh find a number of writing samples. We're going to put that into the AI. We'll run a prompt which gets the AI to extract first of all the core voice. Okay, so these are different modes of writing. They're also sometimes referred to as registers, but it's the same thing. So, but there are going to be things that are common across all of them in terms of how you write. So, the first step is to get the AI to extract what is common to all of those different types of writing. And then the second uh step is kind of an obvious step is to extract the things that are specific to the different modes or the different registers. And the AI is going to do this for you. You do want to obviously review what it's given you and proof it and you can quiz it and you can tweak things if you're not happy.

So, just as a a little side note here, this is a really useful exercise even if you didn't go ahead and build the bot because you're going to get that really high quality in-depth feedback from the AI on how you write. You'll see all the things that you're proud of and that you think are great and make you a great writer, a great communicator. And there might be some things in there which you didn't realize you do and you're not so happy with. Well, the good news is you can you can iterate in real time. not only can you uh adjust what you put into the AI, but you can actually iterate your own actual writing styles. So once you know you're making a mistake, you can you can build in a habit of not doing that thing or doing the thing that you were missing. Um, and then you can tweak the instructions as well so that it more reflects how you want to write going forward. So it's a really useful exercise from that point of view.

And so what we're going to have having done this is uh yeah the AI will extract a essentially a summary like a lot of kind of bullet points that it can then use in the future to write in the way that you're going to write and it will know or and if it's not usually it will pick this up but if you're telling a story it will pick all the common things from the core voice but it will it will pick out the specifics that you use when or telling a story, etc., or when you're doing a step-by-step how-to and so on and so on.

So, this is the and again, I'm not going to read through all this. This will waste too much time, and it'll be on the PDF, but this is the prompt or a prompt for extracting that core voice profile. Uh, I'll read the first paragraph. You're a voice analyst and style guide creator for a non-fiction how-to author or whatever, newsletter, writer, etc., whatever the case may be. Your job is to extract the author's authentic voice from the writing sample or samples below and produce a core voice profile that can be used in future writing and editing.

Um, and it says paste writing sample here, but you can have multiple samples. So, I can just change that to samples plural. Um, and um, you c you can uh, you know, just title them. This is me tell you literally just plain English. This is me telling a story. This is me sharing a client case study. This is me explaining a process step by step and and and paste them in there. And it can extract those common things. And this is a list of the kind of stuff that it's going to pull out. Like what's a voice snapshot? The distinctive features that it can picks up from your voice. Again, it can do this probably better than pretty much any human can. Um what's the relationship you have with your reader? You know, how do you talk to them? Are you do you are you the guru? Are the the expert? Are you the coach? Are you the mentor? Etc.

Um, again, I'm not going to go through all of these. They'll be on the on the PDF, but you know, what kind of vocabulary do you use? How do you like to explain things? Is it very detailed first? Do you open with a story and then go on from there? Do you um use analogies, etc. What sort of frameworks do you like to use? And so all the different things, authority and conviction, are you very confident when you're explaining it or do you hedge the way you explain it and and give people the if this or that? All these different things that are kind of unique to you as a as a writer. and it will extract those from the document and then give you uh basically that kind of summary of how you write which then is going to go into what basically think of it as a master prompt. We're going to build a master prompt. So what is the bot? The bot is is simply just a master prompt that you've saved that understands all the things it needs to to know and it might have some knowledge documents loaded up as well. We'll come to that later but that's what we're actually building here. So this is how we extract that core voice.

And then I mentioned the register profiles or the modes. These are the stories or the how-to and um descriptions things like that. Um, and this is the same one same type of thing but for that. So it's doing doing that one thing. Um, and again, it's it's it's going to ask similar similar questions of that. What are the um you know what is your tone? What is your stance with the reader? What is the cadence? The rhythm? What kind of language? what kind of rhetorical devices do you use? All of these types of things.

And we do that. Uh, let's go back to the top of that. So for the register profiles, rather than putting all of the different samples in at the same time, and you might have five or six samples, and this is just a rule of thumb because this changes all the time, and so you can and you can ask the AI what the what the best practice is when you're setting this up whenever you're watching this. But you know the kind of rule of thumb right now is if you can take five samples, three, five, six samples depending on what you what you have and what you're writing. You don't need to put more in for the sake of it. Uh, and they might total in at anywhere from say 4 to 8,000 words. Put those in, upload those, and then run the uh run the core voice. But when you come to the register, separating out the different types of uh writing, you do them one at a time. So you'll end up with a a set of um instructions for your master prompt for how you you know what you do differently when you tell a story, what you do differently when you're explaining something to somebody. And so you might have three, four, even five of these different things depending on and you can always add these later. Again, you don't have to have everything on day one. You can always if there's something that you realize you missed just go back do this process again for that and then drop it into the instructions. In fact the AI can add it to the instructions for you um as well. So you don't even have to you don't even have to copy and paste.

So um that's how you uh extract the um uh your voice. I do want to ask because I've never heard this idea with um I always thought when you were training your AI, even though you might be doing a storytelling post or whatever, you're just training it to sound like you. I never knew that you could break it up to sound like different versions of you, like the storytelling version of you or the different version of you. Am I understanding this correctly that your one bot is able to recognize and break it into four or five different styles? It's not like five separate bots.

Yeah, absolutely. So if if we think about it, we've what we're going to have at the if you think of it as a hierarchy. What we have at the top of that hierarchy is the core voice. So that is when it's looked at all your different um you know speaking or writing styles and extracted what's common across all of them. How do you always talk? Uh, and that'll be at the top and then it'll have like little almost um subunits underneath for this is the this is the story specifics. This is the how-to specifics. This is when I'm coaching and encouraging and nurturing somebody and anything else that you use in your writing. And again, you're not necess you know, some people don't do the coaching thing or some people um uh well, they most people probably would tell the stories and do the how-to type stuff and do the explanation and do the but uh and if you if you're not sure about this, by the way, just take a sample of your writing, take a a chapter from a book or take a a series of uh um blog or or newsletter posts, put them into the AI, and just ask the AI to to kind of pull out what are different styles of writing here in terms of stories, how to, etc. And it can it'll do that for you and then and point you at what which bit is what. If you're really sitting there looking at it thinking, I don't know exactly what to do here. Have the AI guide you through the process and just asking plain if you can articulate what you want, you don't need to be able to craft the prompt. The AI can do the prompt for you. If you can just articulate the outcome you want, what you're trying to learn, the AI will be able to do it for you. So yes, you've got this hierarchy with the core voice at the top and then the different kind of subvoices. And the AI pretty much in it's pretty smart. Hopefully still non-scentient, but it's pretty smart right now. Um, so as it's uh if it's trying to tell a story, it knows to go to the story box. if it's um uh if it's editing and you've told a story, it knows to go to the story box. Uh and the beauty of this, and again, if you're doing this, you you're probably in the top 1% of people using uh AI to help them to either write or draft or edit because nobody's setting this up. Um and and people are, you know, we've been trained and I was, you know, this is how I first learned to do it as well. you know, you take a writing sample, you drop it in, and you and you get a set of uh instructions. But the problem there is because we write in different ways depending on what we're trying to communicate, you end up, it's not the same as the AI slot problem, but it's kind of like a personalized version of that where you get an amalgam of all the different writing styles. So, it's not great at telling stories the way you tell stories. It's not great at doing a step by step the way you do a step by step. It's this mean average of all the different ways that you uh all the different styles you have of writing or speaking. And that's why this is so important. The AI is more than capable of knowing which one to pick at the time depending on what it is you're trying to explain to somebody.

To me, it's just so cool because as you said, like if I'm writing in like my friendly style, that is going to be completely different to if I'm writing in my professional newsletter style or whatever style it is. And I think it's really cool to even think if you're someone who's new to blogging or writing, you can literally just create four bits of content. You can create a how-to bit of content. You can create a personalized bit of content. And it might take you 2 hours to sit there and write four pieces of content, but once you've done that, you can actually get that bot to identify those different traits in you. So, this is to me it's mind-blowing cuz I was doing what you were doing as well, just doing the one style. So, having these four different styles, I think it's just going to sound so much more authentic.

Yeah. And the one style, that's the kind of like like I said earlier, that's the sort of quick and dirty version. And it's certainly better than not doing it at all. and it'll be much less generic if you do even if you did just that. But if you want to get if you if you're willing to spend a few extra minutes setting it up at the beginning and I say you don't have to analyze this, you don't need to be an English professor. Um the AI can do a much better job than pretty much any of us at extracting this stuff. All you have to do is paste in the samp run, you know, paste the prompt in, put the sample in the little sample box and hit enter and it's going to give you the the uh the raw material that you need to put into that master prompt.

Um, and I'll mention something else as well because you mentioned your you talked about your professional newsletter voice versus a different voice. uh if you have particularly different writing styles that aren't that go beyond just you know the storytelling or the narrative etc. If you have a let's say you have a blog and it's a very kind of professional orientated blog and then you have something else and it's a bit more kind of conversational bit more humorous you probably want to build out two style guides one for each. Um, because again if you try and have you know the serious stuff with the the the conversational stuff you're going to get an average of the two and it's not really going to sound like either. So if you have let's say you're working on a book project and you've got one your voice is a little bit different for that to let's say your social media posts create one for each. If you're doing something for email and you have a particular style for email it's a slightly different way of communicating with people. So it might be a little bit different. It maybe it's a bit more salesy. Maybe there are a few more calls to action. So maybe you create one of these for your email. Once you've done one, by the way, it again, it's starting to sound like a lot of work, but it's not really because again, the AI does the heavy lifting every single time. And once you've built one of these, it's really easy. You know, you you know the process. It's quite straightforward to build something else out for uh you know, for a different type of writing.

Absolutely. I totally agree with you. And that's the thing is like if you if you don't spend the time to do this, you're going to be writing the posts anyway. So, you may as well do do the extra, you know, might be an extra two to three hours of actually building out these four bits of content and then making potentially the first bot and then okay, yeah, you might have to make a second bot, but you've already got 80% of the work done because you've already made a first bot. So, I just think all of this is amazing and mindblowing. And like I totally agree with you in the saying of if you're doing something that's really um professional, that is going to be so much different from something that's a little bit more personalized. Would you say we could use the same um bot for say a blog post as a Facebook post and social media post or would you even separate out from blogging to social media?

I think that's a per the answer to that is is kind of personal because it depends if you write the same way. You know, have a have a read of you probably know if somebody's listening to this, they probably know kind of intuitively whether whether it can be the same or not. But if you're not sure, just go back and um read a couple of posts and if you're still not sure, put them into AI and say, are these in the same sort of style? Um, and if they're not, kind of tell me why they're different. Um, so again, we don't want to over finesse this. Um, so if it's if it's broadly similar, you you might be okay just having the one. But, um, yeah, if you have and again, you want you you want a certain amount of congruence if it's the same audience. But if you have a Facebook style, um, if you have your newsletter style, if you have your email style, you can build out a separate writing bot for each one of these different things.

Um,

I love this. How do we actually build the bot?

Okay. Well, we got a couple we got to touch on the context layers before we actually put it all together. We're we're doing the we're kind of building it as we go along. Um, so what I will say at this point is that uh you know, we've got the the master style guide and the and the kind of subunits. Uh we're about to do the context layers. what we what we do when we build it, we just take all these things we're making as we're talking now and we just we basically put them into this master prompt and then that is your bot.

Um, and and I can do a little screen share on on where that goes. Uh, I mentioned the three context layers. Again, I've sort of explained what they're for. These are these are telling the AI who you are, who you're writing to, who the audience is, and uh what the objective of the book is. But I'll just flesh those out a little bit. You can have AI, some of this stuff you're just going to write yourself, like your your bio. You can have AI interview if you're if you're stuck, if you're really struggling with it.

Um, but the bio is not your life story. It's just relevant information to that particular writing project. So, you know, what's your expertise? You know, how long have you been doing it? What kind of projects have you worked on? Maybe what relevant qualifications you have, what is your motivation? What's that story that got you into it? What's your sort of origin story? Your your your mess to success story, things like that. So that when the AI is trying to write, it has a it's not guessing uh at who you are. It has a little bit of background as to you know why you know the subject, what motivates you, why it's important to you personally. So it can start to try and impart some of those more personal touches. So that's style guide one, sorry, context layer one.

Context layer two is your reader avatar. Again, if you're stuck, have AI interview you. uh core to this. Yes, we have demographics, we have psychoraphics, there's lots of material out there that people can look up on building out avatars, but I think the key for this is where is your reader now and where do they want to go? What's, you know, what's what's the issue, the problem that they're dealing with that has caused them to pick up your book or go to your blog, etc. What's the pain and and what's the the end goal? Where do they want to get to? And again, so we're giving the AI the the the journey so it can it knows where to meet them today and it knows how to kind of guide them towards wherever they want to go.

And then finally, the uh the objectives they they kind of build on the first two, but maybe a little bit more specifically. So what is the objective that you have for that post or that book etc. You personally as the author, what are you hoping to get from it? Um, and what is the uh the reader hoping to get from it? And also what are you hoping that the reader gets from it, which might be slightly different things. Sometimes the reader doesn't know what they don't know. So maybe you want to give them what they they're looking for, but maybe you want to give them something extra as well that they didn't even know they needed. And if you can articulate that, and this is a personal thing, so yeah, you can have the AI ask you questions in interview if you're stuck, but probably this is going to come from you working with your audience for how long you've done that. And you know the it's like any it's like marketing 101. The better you know your audience the better job you can do uh uh serving them. So it's getting those things so the AI knows who it's writing to and what the purpose is.

And then finally just additional instructions which is anything that's not covered by the above. So it might be specific dos and don'ts. A lot of that will be covered by things like your style guide. But if the AI is still putting in uh you know m dashes, it loves to do that. Oxford commas, it loves to do that. It loves to construct sentences in a particular way. It'll say not X or Y but zed. So you you put in there don't do this. Don't kind of preface sentences with in the world today and all these kind of AI tells which are sloppy writing anyway. But AI is is just creating this generic middle ground from every page that it's been trained on. So those sorts of things can go into the additional instructions. Uh things like what what's your target reading age for that particular piece of writing. And against you know as a as a rule of thumb it's often better to have a slightly lower reading uh grade than you might normally write without dumbing it down. We're not trying to insult our audience, but the but the easier it is to understand, the more mental bandwidth your audience can spend learning whatever it is you're trying to teach them and and and understanding the key takeaways. The flip side might be if you're writing to a particular industry audience or or an academic audience, they they might hate that and you might have to make it deliberately a little bit more complex to kind of fit in with the that particular room. So again, things like that, they they're very personal. The AI can't really tell you that. That's comes down to you knowing your audience. And again, if you're not sure, you know, best guess it um talk to the people that you're actually working with or writing to. Um get their feedback and and adjust as you go. So, so those are the three kind of really important things that we need to put into our our master prompt.

So, we've got our first document, which obviously has all that information in it. Those other three things that we were just talking about there, are they a separate document or are they still part of that first document?

We're going to build each part out separately. So, you know, you sit down, you do your your author bio, your backstory, then we sit down and we do another thing which is the avatar or the ideal client profile. So, we're building them one at a time, but then we're going to put them all in the master prompt so that the AI has that uh all of that information to draw upon.

What about using things like um podcast recordings of like things that you've spoken that you've now got a transcript of? Or would you recommend not necessarily doing that because we speak differently to how we write?

Yeah, that's a really good point. Um, you yeah, you you certainly can use uh transcripts of interviews, podcasts, things like that or or you teaching something. Uh, the um and and in some way in in some ways that's almost better. You can argue it either way. You know, when we write we kind of mentally edit as we go. Uh, and so it's it comes out it's we write much more slowly. It comes out more polished. it's much closer to the finished thing, but when we talk, it's much more authentic to how we actually communicate. That's not a bad thing because if you're speaking to people in a room or having a conversation with somebody or on a Zoom with somebody, there's more congruence uh with how how you come across when you're talking to them if you've used that more raw material, so to speak, to to um to train the AI. So what you can do is you can take a transcript and you you know again there's a prompt where you can uh have it either clean up or ignore the the ums the rs the pause where you were going to say one thing and then you changed your mind and went off in a different direction. So you can have it keep all of the vocabulary, all of the meaning, just polish up the grammar and and a little bit of the syntax so that it's it's more uh a a little bit more like it was written than spoken. And then extract you your essence from that. You still have the the the same issue of, you know, storytelling, how to coaching, etc. You still want to create the different subunits. But yes, if you haven't got any written stuff, you can use you can use transcripts uh of you telling that story and use that to train the AI.

That's amazing. Thank you. Yeah, for all of the podcasters and YouTubers kind of got a cheat code, but no, I like that you that it's still important to add the writing in as well so that it does get those different styles. So, I love that. Thank you.

Again, I'm not going to go through all of this. We can put it on the PDF and it would take way too long. We have this master prompt at the top and these are the you know we talked about this already. So voice hierarchy we talked about building out that corvos pro voice profile all the things you have in common with the different communi you know communication styles and the register profile for the different things explanation story selling instruction myth busting that's something I have forgot to mention earlier so if you take a if you're taking a contrarian stance on something in the industry you know how do you do that do you get angry and fired up are you a little bit more measured about it you know we're all a little bit different how we do these things so you'll have all of those in in there and we're going to just paste those in.

Uh, these are the context layers we talked about. You know, your author bio, your reader avatar, your book's objectives. Um, and and it goes on and and you know, where are we? Um, vocabulary from the voice ban list. You know, these are the things we're telling it not to do. Certain words, certain phrases, certain ways of punctuating and so on. All the things that we've talked about.

And then this is so this is the master prompt. And then we're going to paste in underneath um the uh the the different elements you know our um our core voice our our subvoices are uh three context layers and then the sort of humanizing slash additional instructions underneath. So and just put a heading for each one of those. I mean, the AI is pretty good at figuring out anyway, but just put this is the core voice. This is the author bio backstory. Um, just to give it a little heading so it knows what it's actually reading. And you just paste those component parts underneath there.

And then what we're going to do, I'll show you inside chat GPT, but the process is almost identical um for Claude or Gemini. Again, like I said, because you know, we whatever model you choose to use, you're putting the same building blocks in the same foundations to train it. So, if you're in chat GPT, you have you can put all these instructions into a custom GPT or you can put them into a project. In Claude, uh you have um projects or you have also now skills. In Gemini, you have gems. They're all broadly similar.

And um, let me again for people who aren't familiar with GPTs, I think of GPTs as like a an an app within your chat GPT account or a skill if it's clawed where it's you've you've sort of programmed it, set it up to do one thing specifically but do that one thing well. So it's a bit like going to the app store. So if I go to GPTs in my sidebar, I can click on explore GPTs and then this is the GPT store um for ones that are public. Yours doesn't have to be public by the way. Um, it can be, you know, it should be private to you. Uh, just click on the create button and you get this blank uh template. Top tip, don't do it under create. Go was you I won't go into why here. You just don't get as good a result. It's not as it's not as nuanced. You're going to lose a lot of the nuance that you've created doing all of the work that we've just described. So you must do you must do it under the configure tab. and then simply put in the name of the GPT, you know, something that something that's memor able to you so you can find it. Again, it can you can pin it in the sidebar and it can live there, but you can look it up under your list of whatever GPTs you've created. A little description. Again, this doesn't matter so much if it's a personal one. This is more if it's a public facing one, but you can put a description in there.

And then this is where the instructions go. and the last the Google dot that I was just sharing with you which had those uh prompts on literally all you're going to do is copy and paste that into this box and then um once you've uh put that in there let's let's imagine we're going to do that now. So uh I won't put the whole thing in take too long but let's pretend that's the whole thing. I'm going to paste that in there and then I can close that. it's in there. Once I've filled out all these different boxes, this will not be grayed out. I'll be able to click that button to create it and save it. And again, if you're updating it, because you can update this at any time, like I say, um, remind me to talk a little bit about feeding back to the model in a second when I've I've done this, if I forget.

Conversation starters. You don't really need to worry about these because it's personal to you. This is again, if it's public facing, you might want to put a, you know, uh, create a logo if it's a if it's a graphic design one. some things that people can click on and start, you know, um have a have a starting point for them. And then upload files. So, you might want to upload files into um into what they call the knowledge component. Uh, I'll mention this because this sometimes this is a problem. This instructions box uh on CHBT has a character limit of 8,000 characters. That includes spaces. So that's actually not a huge uh window for your instructions. So if you are running out of space and this might you might be building a GPT for something completely different. So if you are running out of space, what you can do is have your core instructions here and then supplemental information. You can upload those as a as a text file or as a or as a PDF and you can put those into the knowledge section. So if you do run out of space in the instructions, um, you know, maybe you you you would could put the the the style guide, for example, into here. Now, ideally, you want the style guide to fit into the instructions box, but that's the workaround if it doesn't.

Now, this is a little pro tip. I haven't mentioned this yet. Let's assume that the uh the instructions plus this the elements you put underneath all fit into the uh instructions box. So you have your style guide in there, the core guide, the sub guides. What you can also do is take the original raw material, you know, the story that you used to extract that information and upload those as files into the knowledge. And the reason for doing that is that what we can have the AI do is have it it will default to the instructions first. So, if it's if it's helping you to either draft a story from scratch or if you've told a story and and and you're having the AI edit it for you and you want it to maintain your authentic voice, uh either way, it's going to let the sort of core plus the story specific instructions. However, if it's not sure about something, it can go and do a little deeper dive and go back to the original source material. Again, this is something that almost nobody is doing. So, if you do this, again, this is, you know, top 1% stuff. So most of the time it's not going to look at the the the actual story that you uploaded as a file in the knowledge section. It doesn't need to. But if it's stuck on something and it's not sure which way to go, it can go back to the source material, read that through in a you know in a microscond um and and then make a decision based on that uh and and that gives it a little extra steer. So that's a little pro tip for um just for giving the AI basically that you know as much as you can give it and having that kind of belt and braces.

So So it's it really is this simple. You just paste the instructions in here, upload any files that you need to do. Hit hit create or save if you're updating it and and and it's done. you you've built that out and and it's the the box for Gemini, the box for Claude uh or the window.

for Claude is is is virtually the same. If you can do it here, you can do it there as well.

It's amazing to think how easy it actually is to achieve this end result. Like everybody thinks I'm going to build a bot and that it's going to take them forever, but it's actually just a couple of clicks at a button and some really great prompts. So, loving this. And you did say that we could go back and make changes to the bot afterwards. How do we go and do that?

Yeah, literally what I just showed you. So, um just go back to your, you know, I had there was a little uh icon um top right of the screen as you're looking at it saying, you know, my GPTs um or if it's Claude and you've set up a skill, you know, my skills, etc. You click that button. If you've only got the one, it'll be there. But if you've got a if you've built out a whole bunch, just find the one that you're looking for, click on it, and that exact same window will come up and you can you can either go in and manually change the instructions. Um, but I'll talk about uh uh okay, let's so let's go back. Let's say you've just written a piece of content or just had the AI write or or edit a piece of content for you and it's got something wrong. You might be able to look at that and it might be obvious. It might be just a way it constructs a sentence or a piece of punctuation and you can say, you know, okay, I don't want you to do this in future. But if you're not sure how to describe it, um, what you want to do is open up a separate chat window, copy and paste the bit that it's got wrong, and then underneath that, just write out how you would like it to have been written and ask the AI to and and say literally um, like this is the output that I got. paste that output and that this is how I would have liked it to have been written. Describe to me how I would explain that to AI and and get it to write the instructions for you rather it it'll do a much better job than I I write prompts and I've been writing prompts every day, every working day at least for you know about three years now. I mean I got my chat GPT account about a week after it went um kind of public. uh the AI can do a much better job than I can of writing the prompts. Plus, it saves you an absolute ton of time. So, you so so you do it in a fraction of the time and uh you get much better results. So, I yeah, why would you ever do it yourself? So, have the AI analyze the piece of writing that's wrong, compare it to what you would like, and then it gives you the instructions and then you can go across to uh what I like to do, again, couple of tips. I haven't mentioned these before. When I'm putting the bot the the master instructions together, I always like to do that in something like a Google doc. So, it's so I've got it outside of my AI as an easy reference because the filing system within AI isn't isn't isn't the best. It also helps with things like version control. You know, I can make it make a new version but keep the old version. I can always swap them in and out if one gets if the old one was getting better results. Uh so as a best practice I was kind of assemble the instructions in the Google doc and then paste them across into the uh into the GPT. Um but also because I'm probably going to be having this conversation with AI helping it where it's helping me to build out the uh you know it's analyzing the text is building out the instructions for me that are going to go into the master prompt in the first place. I can go back into that chat where I was putting, you know, doing this analysis or the I wasn't the AI was doing it for me. And if I've got that issue where it's it's got something wrong, I can I can have that conversation inside inside that chat or inside that project where I've built out the instructions and so it has that context of of it knows what it's doing without me having to explain it. Uh, and it gives me the um gives me the output that I want to change. And I most people probably know this now, but just in case anybody's not aware of it, you can you can create like a canvas inside a chat GPT and the others have equivalents which are documents that are that you can edit as well as the AI. So if you have a canvas for your for your instructions, what you can do is ask the AI to drop it into the canvas to edit the edit the instructions. So again, you don't even have to go into your instructions and find the place to, you know, change something or drop it in. It'll do it for you. And then you just re-upload, you know, put those in your Google doc, paste them across into your uh your custom GPT or your gem or your skill depending on what model you're using. Uh, and it's done. So, there's you need to know how to use the AI, but you don't really need to know how to craft this, how to how to explain it. The AI is going to do it for itself.

This is amazing. I feel like you've given all of us at least eight or nine different like pro tips that you just don't hear about. And so, like, this has just been phenomenal. Is there anything else we kind of need to know when we're trying to get started with sounding authentic with AI in our writing?

Yeah, there's there's just I I think I've covered all of the really important things except one more like real pro tip. Um the single once you've built this out, this is this is great, but the single best way to stop the drift again um because you've baked all of these rules and instructions and guides into um a permanent set of instructions, you're not going to get the kind of drift that you would if you were just prompting AI. So that takes care of a lot of the issues anyway because obviously with context windows, you know, how much AI can remember, the longer the conversation gets or the bigger the chapter, the the more it drifts away from the original core instructions or we have a a set of master instructions which it can keep referring back to. So that that helps a great deal. But what I encourage my students, my clients to do if uh if if they can, sometimes we want to, again, we just want to do it really quickly and get it out the door. But if it's a if it's an important piece of writing, the best way to do it is to um you know, do what you would normally do, create a framework, and again, AI can help you with with fine-tuning the framework so that you've got something to work from, and then speak it, dictate it. Um, it's not as easy as just having pushing the button, having a AI write it for you, but it's going to be much more authentic to you. And then the AI can can essentially edit that for you with all of these instructions with the style guide and all the rules and the dos and the don'ts and do a really good quick job of that. And those two things together are really going to make sure that it's always going to be in your voice. And again, um, if you don't have time, yes, you've still got a really well-trained AI, but it's never it's not going to be quite as good as if you you create that first draft yourself, but what it means is you're not kind of paralyzed by trying to get it perfect. You're not editing as you go. You can it can be more of a sort of stream of consciousness and you have just your little bullet points to tell you what to talk about next or what story to tell next, etc. So, uh, it's a little bit more work than just having the AI do it for you, but you'll get it's it's how you get the best possible results because it's still coming from you. Also, again, if you are creating a book and you're going to put it on Amazon, Amazon distinguishes between AI generated versus AI assisted. It knows that the gene is out of the bottle. AI is not going away. So, it it accepts the people are going to be using AI. But what it doesn't want really is people just to have AI write the book for them. It's obviously it's all about user experience. So if you've got anything in your book, even if it's just a paragraph, um Amazon asks you to declare if you've got any any AI generated content. Now you can argue whether you actually need need to declare it. You can argue how good or bad the AI um you know uh what are the what what's the word I'm looking for? the the little programs that spot whether it's AI written sometimes they get it wrong. Um but ideally we want to stay away from that. Um if if you create the first draft whether you write it or type it or speak it doesn't matter. Uh and then you AI polishes it up for you. It doesn't matter how much AI then improves it. It's still considered AI assisted because you created that initial draft which is and and Amazon is happy with that. So AI can help you with all the deep research. It can help you structure it before you even write a single word. It can um it can take your raw output and turn it into written pros and really polish it up for you. Uh and all of that is okay. Amazon's cool with that. Um what it's not cool with is is you going in writing a prompt saying, "Hey, write me a chapter about XY Z." Um so again, if you if you speak the book, you're going to or if you speak the passage, the first raw draft, uh it's going to be more authentic. And if you are publishing on Amazon or a similar platform that has a similar rule, you're not going to get caught out by that rule.

Knowing you're a ninetime bestselling author, I actually was going to ask you your opinion on putting out books written by AI, but I feel like you've kind of said AI assisted is okay, but not the whole thing.

Yeah. Yeah. And say sometimes you just want to get something out of the door really quickly, and it's a it is it's it's not really an issue if the AI does it. I I would always always I've never uh sent anything out that AI's written that I haven't gone through and edited and proof read and um you know made tweaks to um you should always proof it. You never know if it's going to make a really horrible mistake. But um yeah, if it's just a a fiveminute email and you just need to get it out the door, fine. But if you're writing a book and it's got your name on it and it's, you know, it's your kind of like your your ultimate business card, you probably want it to be coming from you, not coming from the AI. So that's not that's a situation where I would I wouldn't be using the AI to do the first draft for me. But because it's got so much uh depth of kind of understanding of who you are, who you're writing to, etc. It can do a fantastic job of taking what you give it, that raw first attempt, and turning it into something that's really polished. Um, and yeah, you you've always got to review it. And I I'd still recommend having a final human edit slashproof read, but it's a much more light touch thing because the AI's done most of the work before you actually hand it over.

I love that. Thank you. I actually have written six books myself. They were fiction books. And I remember my editor saying to me, she's like, "Yeah, you say the same thing a lot. I took out about 10% of your actual writing just because you repeat yourself." So yeah, I definitely think that getting something like AI to clean it up a bit and then also getting it edited, like getting an editor to just look it over would definitely be like

the best way forward there.

AI is really good at that kind of stuff. It can pick out redundancies. It can pick out if the pacing's dropped off, which is applicable both to fiction and non-fiction. You still want to worry about, you know, keeping people engaged. Um, you know, it can check for continuity very very quickly. So, it's it's a really powerful editing tool as well as powerful writing tool.

Absolutely love that. And today has been amazing, John. It's been really really nice to hear from you and hear all of your like tips. I feel like we were coming into this thinking we were just going to be learning about how to make our writing good, but there's just so many pro tips in here that are just phenomenal for everything in relation to creating a chatbot. uh like a bot on chat GBT basically for anyone doing any sort of writing whether that is um a newsletter or an email or whether that is a book chapter. So loved this. Thank you. And where can people find you?

Uh well my books are on Amazon uh as you would expect. And um you people can find me at johntai.com. That's probably the best place to go.

Amazing. Thank you so much and thank you for being here. Thank