So What? How to Write a Book with Generative AI

Well, hey everyone, happy Thursday. Welcome to So What the Marketing Analytics and Insights Live Show. I'm Katie, joined by Chris and John. Gang's all here. And John is in his own home.

I know, back finally out of the jerkshares. Uh out of the jerk shares. Well, this week we are using generative AI to write a book. So we, in the matter of however long this live stream is, are gonna write a book, apparently. Uh so Chris, where the heck are we starting with this book that we're writing?

Well, we we're gonna start it in a couple of different places. But the first place to start is do you want the book to be copyrighted or not? Um generally most people do. But the reason I ask is because if we want any work to be copyrighted, it has to start with a human originated work. But if you start with like a prompt, say, like, oh, well, write a book about B2B marketing, right?

And it capably splits one. You can't copyright that because it is not a human-origin originated work. And under in all but five jurisdictions on the planet, um, that would not be copyrightable. So we do generally speaking, if you're gonna go to the effort of creating a book and you're gonna go to the effort of marketing a book, you probably want it to be copyrighted as no one can just copy and paste your entire thing and you have no recourse. So that's the first part.

So we need material that's human originated in order to write a book that will be copyrightable. And it just so happens that we have this thing called the AI ready strategist, which Katie, you want to take a few minutes to uh explain what it is that we're looking at here. Absolutely. So if you haven't heard, um we are in the process of launching a brand spanking new course, the AI Ready Strategist. If you downloaded our um AI marketing ready uh strategy kit from a couple of months back, this course is based on that kit with all of the frameworks and templates and checklists and workbooks, expanded out, plus a heck of a lot more content and useful assets that have been created in addition to what lives in the downloadable kit.

So the kit is a good start. This course really digs into the details of how to use those assets in a way that's gonna set you up for success. And so we called it the AI ready strategist, but it's really the technology ready strategist because AI in the the way that the course is done designed, AI is the tech focus right now. Tomorrow, it could be something else. The course is foundational, so swap in any flavor of the month shiny object technology, and everything you learn in this course will still hold true.

Just swap out AI for Bitcoin or Bitcoin for NFT or NFT for CRM. It doesn't matter. And that's what I really love about putting this kind of a course together is that it's going to stand the test of time and it's going to outlast any trend that comes along in technology. And so to be clear, in this course, we're not teaching you how to use generative AI. We're getting you ready from an organizational perspective to use generative AI because as you'll learn in the course, the number one failure point of these integrations and digital transformations is people.

Number two is process. We're gonna go through all of that in the course. So right now it's what August 28th. That code is good through midnight tomorrow night, uh Eastern Standard Time. And then the course goes live September 2nd.

So go ahead and get it. I would recommend pre-registering and getting that discount, that strategy 2025, and then it will be available to use, to love, to enjoy, to rewatch. All the downloads are there. Trust insights.ai slash AI strategy course. How'd I do Chris?

That that was a good commercial plug. So here's where we're gonna start the book writing process. You already did all the hard work of creating the course content. And in every module, there are, of course, transcripts, because you should be creating transcripts. So our first step towards creating the book is to get all of the raw data.

So all the transcripts uh we download. Now they're PDFs, and PDFs are compatible with generative AI systems, but they're not ideal because they're bulky files. So what we'd want to do is turn those PDFs into just plain text files, the smallest, lightest weight thing. So you uh there's any number of ways to do this. I use a free utility called PDF to text.

It's it runs on the command line. You can set it up as a batch job and just turn all the transcripts into text files out of away from PDS. This preserves the letters and numbers and the words part, but it it gets rid of all the formatting and weird stuff inside of PDFI. So that's ingredient number one. Ingredient number two that we need, we need a book outline.

Now, again, in this case, because you already wrote and built the entire course, we basically have the book outline, right? It's the course outline. It's just that will be instead of lessons, there'll be chapters. So that is is pretty straightforward. If you had say a bunch of disconnected materials, you could have generative AI create an outline.

You could say, hey, here's everything I've written about. Help me organize this into some kind of coherent outline so that I, you know, so that it's not just a disconnected bunch of mad ramblings from an old person sitting on their porch. We don't have to do that this time, but that would be step number two is to say we need we need structure, we need logic, we need order. So we've got that. Well, and uh, if we back up a second, so that's where the AI kit started is I have all of these things.

Can you help me put them together in a logical way? Which is how the kit came about. So AI didn't write the downloadable paper, the kit that uh everybody got. I did. But AI helped me put it together in a way that made sense that I couldn't necessarily see.

And then we took that. I had AI say, okay, this is great, but let's turn it into a course. And it gave me the outline, but it wasn't all like AI just did it for me. I still had to connect the pieces. So I just want to make clear, you know, to people to, as you're saying, like let AI do it.

You still need a lot of human intervention. The amount of time I spent working on this and rewriting it versus having AI put something together, it's out of proportion. It was more me than AI, but AI helped me connect the pieces in a logical order that I was struggling to see. Exactly. So that's ingredient number two, an outline.

Ingredient number three is there may be things that supporting information that might not be in the transcripts, but would still be relevant. One of the things that Katie does really well is uh she's an extremely good storyteller. And if you're not subscribed to it, the trust insights newsletter uh available at TrustInsights.ai slash newsletter, every week Katie writes what's the we call the cold open, that first part of the newsletter, the first half of it, actually more than half because it tends to be uh about two-thirds of the the original content in there. And so what we would want to do is say, well, let's use that as a supporting data source because there may be things that she may have written about maybe in February of 2024. She might not remember that she wrote about it, that but it's relevant.

But we would want to include things like those stories. So our third ingredient would be to take the newsletters that we have, sew them all up, and put them into one great big honk and markdown file. So markdown is the plain text format that generative AI is very capable of reading. It's a really good format. We do this every December.

Uh, what's called Letters from the Coroner Office, which is Katie's annual book of all the newsletter entries she's written for that year. However, it's also a great way to reuse this for other things. So this is going to be our next data source is having uh all of the newsletters from the last 18 months or so, since January 2024 to today. So that's the ingredients. Next, if we continue the analogy of the kitchen, we need appliances, right?

We need a ways to process the stuff. The two hardest tasks when writing a book are writing and editing, right? Those are both things that that require a lot of time, a lot of thinking, and uh we would probably want to say, let's get some help specific to that. How would we do this? Well, the easiest way to do this would be we'll get fancy agentic AI.

Agentic AI is essentially saying, let's create little AI agents, little self-contained apps or programs that can run autonomously and perform a specific task, like writing a chapter of a book, uh, or maybe editing a chapter of a book. So that's we're gonna need that as one of the appliances. We're also going to need Katie's writing style. Katie's writing style is very distinctive, and so we have a prompt that we've used, and we've shared this before in the Analytics for Marketer Slack group, but it's this nice, really long writing style prompt that helps an AI deconstruct a writing style, quantify it, and codify it into rules. So our first step would be to take perhaps all those newsletters, because the transcripts are the spoken word, not the written word.

Take all the newsletters and build a writing style for that. So what we're gonna do, what I'm gonna do is I'm gonna get um, oh, I forgot one other thing. Uh the the actual system. For today, we're gonna use Anthropic's clawed code. So clawed code is a coding environment.

It's meant for writing code, but you can use it for any language model task. You do not have to use this. You can take just the regular prompts themselves, and as a human being, use the regular clawed desktop or chat GPT or Gemini or Copilot. That part doesn't matter. And it is a tool called Serena.

And Serena is a coding MCP. Now, this is starting to get into the technical realm, but it uh what this does is it's like a librarian. When you add it to a code repository, or in this case a book repository, it's gonna read through all the files, it's gonna read through your intent on what you want to do, it's gonna try and figure out oh, this is we're we're we're we're gonna you're doing this, you have these names, you have these variables, these are the instructions and directives that have to keep in mind all the time. Serena is super useful for keeping AI on the rails because it remembers things from conversation to conversation that if you don't use it, it just tends to forget. Um so that's the last appliance.

So we have Claude Code, we have Serena, and we're gonna build the writing style and then start the writing process. While I get the setup, Katie, once you and John debate why you would even do this. Well, you know, uh John, you know, I remember earlier on in the Life of Trust Insights, you wrote a book. And I this was before generative AI was available to general consumers. And so the AI technology has existed for a long time, but the way that it exists that we know today has only been around for a couple of years.

And your book predates that. And so I guess asking you as someone who has written a book the uh quote unquote old-fashioned way, um, you know, what would you would you do something like this as opposed to the way you did it before? Yeah, absolutely. I mean, this is just so much pain to anybody that's gone through writing a book. And it's even it's funny.

I was talking last week with Ashley Faust, who just had a book, Human Centered Marketing came out, and it was the same type of thing as my book. It was, hey, let's look at everything that's worked over the past five years and you know, just write up all the tactics. So you kind of have a book of like, okay, if somebody throws you this job today, here's how you do it. But yeah, it would be so much faster, easier, cheaper now, because you know, I had to do full brain dump on everything. Whereas, you know, you can say, hey, create a chapter on what to do at a trade show, and you at least will get, you know, 50 or 60% of all the good stuff out there, everything that's just out in the in public.

And then you can just go through and add and edit and you know, just kind of redirect the copy. You don't have to do it all from scratch. And you know, staring at that empty page, that's the most painful part of getting the book off the ground. And so being able to, yeah, just generate a whole bunch of copy that's at least close to where you want to go, is a huge time savings. And then, you know, being able to do stuff like have it change the voice a little bit without having to rewrite the book.

I mean, I I was talking to a screenwriter who was talking about how he wrote a whole movie and the production company came back and said, okay, the lead needs to be a woman. And he was like, Oh, all right, no problem. Let me just throw it into AI, you know, grind the whole thing out, and 90% of the work is done. So yeah, it's definitely uh a much better time to be an author than ever before because it's a whole lot less painful. I've struggled a little bit with the gosh, how do I say it?

I guess like the morality of it of did I write the book or did AI write the book? And you know, and this is something that Chris and I have debated on podcasts and in other conversations. And I think if I were going in and saying, hey, generative AI, how do I improve the process operations of a business and took what it wrote, that would be to me, I'm not really writing a book. But where I'm mostly okay with this version of it is because it is all my writing. It is everything that I've personally, the human has done.

I'm just asking the machine to like clean up my rambling into something a little bit more coherent and usable. Exactly. So part of the kickoff process for this, again, using a coding tool is this onboarding where the onboarding tells the agent, hey, you have to read all the background data that we provided, learn what's in it, learn all the pieces, learn what each file is for, and then store that in your own internal agent memory so that it remembers, oh, this is a book project. This is not a coding project, that we're not using Python, we're using Markdown. Um, these are the files that are available to me.

This is the layout, the folders. Because in here we have uh, and this is part of the process. We have a clear set of folders. We have things like writing style, we have the documents, which is in a coding project where you put things like a requirements document. And even in a project like this, I didn't do it for this project because KD already did that for the course itself.

Um, but if I was doing this from scratch, I would absolutely have a requirements document that says like, here's who this book is for, here's the audience it serves, here's the things that have to be in it. Um, and the course transcripts and stuff like that. So let's go ahead now and start our first prompt. Our first prompt is to say using the instructions for how to use a writing to create a writing style, you're gonna read some of the background newsletters that we provided and extract out what Katie's writing style is, fill out the template, because there's a template in our writing style instructions, so that we end up with a a solid analysis of what Katie's writing style is because and then put it in the writing style folder. Now, while it's doing that, I also forgot I need to put an output folder to actually put the chapters of the book because it wouldn't be terribly helpful to just have it leaving files all over the place.

The components of a writing style, just to review, are things like clarity, conciseness, and not saying that these are how you should write, it's to measure how concise is the author. Some authors very verbose. But you want to know that so you can mimic it properly. Uh, purpose and goal, sentence structure, diction and word choices, specificity, figurative language, and so on and so forth. And so these are all the things that we would want to diagnose and extract from existing writing.

Where almost every writing style prompt in AI goes completely wrong is here. The frequency. So that says frequency one to 10. Whenever somebody squinting, I was like, What whenever somebody says, you know, analyze my writing style and gendered AI comes back with, oh, well, you're a storyteller in the first person and you use whelp a lot in your language or whatever, it doesn't quantify it, which means that when you then go use that writing style and tell it to imitate it, what you often end up with is a caricature of that person's writing, because it tries to use those writing style elements all the time and starts every other sentence with well. I'm like, that that's not me.

And well, it it's funny because earlier versions of my writing style that we've done did turn out to be a caricature. Um, the Katie artifact in Claude feels very much like a caricature of me. Um, but it's a good starting place at least. Um, you know, when we get to the point where we have a little bit of free time, you know, we can update all that stuff. But for now, it's a it's good enough because I know what I would and wouldn't say.

So I can take that and edit it very easily, whereas someone who isn't me might be like, oh yeah, that totally sounds like her. Let's just run with that. And it's not quite hitting the mark. Mm-hmm. Exactly.

So Claude has gone through and evaluated your writing and provided the descriptive part so that it knows it's right, but it also provides the quantification. So for example, narrative style if applicable, the frequency for this particular style is an eight. Uh some other things in here, let's take a look a use of sources. So you typically in your newsletters you don't cite a lot of sources because it's you are the source. Like you are the authority on this stuff.

So you don't tend to cite a lot of sources. So it should know that. You don't write a ton with imagery. I said that's very narcissistic of me. But that's your style from the newsletter's nothing wrong with it.

Everyone's got their own style. So we've got the writing style now our next step is we have to actually build these two agents the agents that I talk about. Now in Claude code in particular you build the agents right inside the interface. So I'm going to type in the word command slash agent so we're going to create a new brand new agent. You can choose from personal or project personal means it applies to everything that you use with Claude code and project is specific to a project which is what we want to do.

With Claude you can either give it your complete prompt yourself or you can give the start of a prompt and have it edit it. Now I have found personally that having it edit uh my prompt comes comes with a better prompt. So I'm going to say gent with Claude and I'm going to give it the agent description. It's called chapter writer. Chapter writer is given an assignment, and then uh it consults the course transcripts to see what the source information is.

It will look at the writing style to make sure that it's it knows how you write properly. It will retrieve background stuff from the old issues of the newsletters if it doesn't have enough information to write with, and then it will write to the chapters folder. So I'm gonna say, go ahead and now make me this agent that is uh and it's just gonna create this thing, and it'll become available as essentially a little mini app inside Claude, um, which is super super useful because uh what these things do is they run inside their own little context box so they don't soak up main system memory, and you can do other things while they're running, which is super cool. So we're gonna specify inherit from parent, and we'll give this the color blue, and our first agent is done. Now it's done.

You're making it look easy. I I this is a cooking show. Um the second agent is going to be called a developmental editor. So there's two fundamental different kinds of editors. Um, there is like a copy editor, like fixing typos and stuff like that, and then there's a developmental editor who goes and reads things like does this make sense?

Like, is this logical? Do we forget things? Is this serve the purpose of the audience? And very often a developmental editor will come back to you with your book all marked up in red, saying, like, you need this, this, this, and this, and your book is terrible and and stuff. And they don't, it's not so much like you know, this is a should be a semicolon.

It's like you completely forgot things or you you lost the plot in this chapter. Try again. Um, we're gonna create a second agent called in a developmental editor, and it's going to reference that gigantic prompt because you can do that with agents, you can give them extensive programming elsewhere, and this will be the follow-on agent that will go on after Claude writes a chapter, then the developmental editor will come in behind it and fix it up and say, like, yeah, you know, this could have been better and stuff like that. John, I saw you nodding along. I'm guessing that was part of the process of writing your book was having a developmental editor basically rip your ego to shreds.

Well, it wasn't mine. I had heard a story about Brad Meltzer, the famous author about how his editor in one chapter wrote in the sidebar, uh, this chapter is so boring, I want to kill myself. And even as a best selling author, he was like, That scarred me for a long time. I went. But that yeah, that's the role.

It's and it is I I have the same thing. I have an editor that I work with that you know continually beats the crap out of me. And uh, yeah, it's humbling, but that's the only way to get to great writing. So, yeah, another thing, if you can have you know, AI do some of that beating in the privacy of your home, that will share you a uh, you know, some embarrassing time among your peers. I feel like in some ways, like all of my feelings are gonna be spared from seeing how poorly it's written the first time.

Because one, I've already done the writing, so it's gonna basically repurpose everything I've done, and then it's gonna edit it. So I'm all I'm gonna see is a really nice output that I can then go through as a human and edit. Uh, but I'm just sort of like, oh man, my feelings really get spared here, don't they? Exactly. So we now have the writing style, we have the course transcripts, we have the background information on the newsletters.

Um, we have the two-agent setup. We have the Serena memory setup. Our next step is to kick off the writing process itself. And so we're going to give it a great big honk and beefy prompt. In fact, let me put this on screen and make it larger so that we can see it because it's really hard to read.

And it goes like this. Using the chapter writer, read the outline, and then with the Serena MCP, if needed, write each subsection of the outline, starting with section 1.1 as a chapter, with your chapters outputted to the chapters folder. For example, lesson 3.1 becomes chapter 3.1. Each subsection from the outline is a lesson, and each lesson is its own chapter. Rely on the course transcripts and the appropriate lesson number as the primary sources for your information.

When you're writing, if you need supplementary information, use the data in the background folder, extracting paragraphs with Serena MCP. And this is important because we want to pull not just a word or a phrase, but the whole paragraph for the extra context. Adhere strictly to the writing style in writing style unless the style conflicts with this prompt. This prompt has supremacy. And I'll explain why that's important in a second.

Use verbatim text as much as practical. That's really important because we want to keep Katie's voice as intact as possible. Remember, this is a book. So only have one list per chapter because otherwise AI loves making lists. At the end of the chapter, the only chap list in the is at the end of the chapter named So What?

Otherwise, the rest of the chapter should be formatted as prose appropriate for a book. You must write prose for everything except the key takeaways. Because this is a book and not a course, you will need to remove references in the transcripts to the course. You also need to remove the pleasantries that are course specific, such as welcome back, and in the next lesson. Um, this will be self-contained, non-fiction business book should have no references to the course or other content documents.

Write each chapter from the subsections, the outline now with the chapter writer agent answering MCP, beginning with lesson 1.1. Proceed through all the chapters and uh lessons uninterrupted. So we're basically gonna say, you are now gonna go off and you're gonna do this thing. So let's go ahead and plop this in the chat. And now it is going to begin what is approximately a two-hour process where it's going to read, write, and go on, and etc.

And just keep reading and writing, keep reading and writing throughout time. So, like a good cooking show, uh, we have the fully baked uh process where it has gone through, and you can see, let me get down to the relevant section here. You can see, you know, write this chapter, write this chapter, write this chapter, and it just goes through every chapter takes between three to seven minutes for it to process. And what comes out in the end is a list of chapters. So let's open up actually.

Let's uh let me open up one of these so we can see. As you like to say, Chris, I feel like two hours, like three to seven minutes per chapter, is a reassuringly long time for AI to be processing. Because I feel like if it came back within 30 seconds, yes, that's really fast, but then it says to me, did it even really pay attention to the background material? We gave it a lot of information. Did it just skip over it and say, I'm just gonna write whatever I feel like?

Whereas if you're saying it's gonna take two hours to process through everything we gave it, I'm like, oh, that actually makes me feel more reassured that it's really going to take a look at what was previously created and put it together in a more thoughtful way. Exactly. So here's an example. Uh, let me make this full screen here and make it uh easier to see. Four point introduction to the six C's of data quality.

When you read this and you look at this, this is A, it's fairly close to the course transcript, right? Uh all the uh speech disfluencies, uh repeated words and things like that, uh uh tie-downs on the end of sentences, uh using like okay or write or you know, all that stuff has been extracted out. And so what you end up with is a pretty decent section of a chapter, right? And this looks like this looks like what you would want to see in a book. And here's the so what the list of takeaways from this section.

More data isn't always better. Well, yeah, I mean, that's that's kind of a given. So now the first draft is done. What do we do? We now go back in and we fire up the developmental editor.

And the developmental editor is given this prompt. Using the developmental editor in the MCP, you're gonna evaluate each chapter in the chapter's full for the following. Is the writing fully aligned with the writing style? Is the chapter aligned with the outline? Like, I did you make things up?

Uh, is it aligned with the sources in the transcripts? If you need additional content such as stories and personal anecdotes that are hallmarks of KD Rivera's writing style, are they sourced from the newsletters? Is the content truthful and factually correct based on the sources, avoiding hallucination and avoiding fabrication? Did you make things up? Does the content serve the needs of the business nonfiction reader who wants to understand AI strategy at a strategic level?

Is the content written for appropriate for a book with references to courses, downloads, transcripts, etc. removed? Is it not part of the book? Does the content have the required so what list of takeaways at the end? Is the content written well written as form well formatted prose with no other lists besides so what?

So this agent is given these instructions, and it now goes through and says, okay, great. I'm now going to use the developmental area and inspect each chapter. And it when it went through, it modified about half the chapters. It found things that the first draft didn't really do as well as it could have, and it fixed them up. So we should what we have now is uh let's close that's the wrong folder.

Um here's the Katie book folder. So we now have the 20 uh 28 chapters in total, all nicely cleaned up and in markdown format, ready to be used. Now, this is good. Like this is super helpful. However, in this current format, it's not exactly a book, right?

So we need to now go to the next step, which is to convert it. Here's the wonderful thing about the markdown format. The markdown format, besides being text, is easily converted to other formats. And so what the most popular format for book writing is Microsoft Word. So if I issue a command that says for every markdown folder in the markdown file in this folder, convert it to a word doc.

Now it has gone through and turned that ugly markdown into a nicely formatted Word doc. So here we're this is already looking better. And now we get we'll take we can take those Word docs, close this project, into I'm using a package called Scrivener. We'll call this AI ready strategist. Ron on this because obviously we're not gonna do the full book publication right now.

Oops. Oh, actually, let's do EPUB. Switch to EPUB. This is an ebook. We're gonna give it the section layout to assign it to a new page section.

Hit compile and go. Now, what it's going to do is it's going to take all of these chapters and hopefully successfully. I think it's still working. And here we are in Apple Books, right on my desktop is the new book. Nicely formatted, ebook ready.

And so we now have a lovely textbook to go with the course, or as it's as a standalone, in Katie's words, from all the hard work that Katie did in approximately three hours. With you know, 20 years of experience of writing behind it. Yes, 20 years of experience. A month of recording the course. Uh-huh.

Plus two months of planning out all the slides and the content for the course. Uh-huh stuff. But what this emphasizes is that if you've already got existing content laying around, use it. Take the content you have, use generative AI to essentially clean it up and reformat it and help repackage it and turn into something that you can extend the life of your content. So a real simple thing.

Let's say you have a podcast, um, like the Marketing Over Coffee podcast, and you have interviewed 400 guests over the years. Uh, you could take all the transcripts from all 400 guest episodes and interviews and call it, you know, I don't know, the espresso edition, and you know, marketing lessons learned from 400 guests on the marketing over coffee podcast. Following this exact same process of getting the transcripts, uh distilling it down, and turning it into a book. You could have probably, I would I would actually say probably five or six different books because the guests have been in different categories sales and marketing and biz dev and customer service and strategy. You already did all the hard work, the hard work is done.

Now it's just having machines essentially do the paperwork, just a matter of branding and out. Marketing over espresso, espresso marketing. Well, we get burned though. The one thing it would have to go through in the topicality of it, you know, it's like nobody wants to hear all those MySpace interviews from 1997, you know. Those aren't I don't know.

There is something to be said for nostalgia. This is true. It could be uh you know, the 500 marketing lessons you no longer need or something like that. Remember to put your bet your best prospects in your top eight. But I think you know, I think this is really cool.

Obviously, for those watching, like what Chris produced today is not going to be available with the course on Tuesday. Uh, that said, there may be a version of it coming down the few in the future, considering I have yet to actually read any of it. Um pro tip make sure that there is always human intervention with your generative AI and that you don't just take the first output and go, okay, good enough. Exactly. But it repurposes the hard work you already did and extends the life of it.

And, you know, like you said, if it's available eventually as its own standalone book, then the book becomes, oh, by the way, you can take the course and get the certificate and get access to the community and get all the downloads and worksheets and things. So the book could be sold at like $29.99 or $39.99, and then use that as a business driver to drive registrations for the course. And that's that's the hard part. And again, even not just the months you've spent working on the course, but almost two years of newsletters that you've written too. Yeah, I would agree with that.

And you know, it's there's and this is we, you know, without getting too off topic, but this is something that we talked about a few weeks ago. AI is going to replace a lot of the processes and summarization and you know, the things that you just showed. It doesn't replace the critical thinking. And it becomes even more crucial for you as the human to come up with those original ideas and do the critical thinking about them and figure out the different innovative ways to use the ideas that you have. AI is only going to take it so far.

And you know, we get a lot of the sameness from generative AI. What makes it different and useful are the humans using it. Exactly. And what you also didn't see, uh, because I started this process this morning at 7 a.m. was the first two tries where the prompts themselves needed some improvement.

Uh, I looked at the first couple chapter drafts and saw, you know what, it it needs this, it needs this, it needs to be told this. And so, even in the the prompt case itself, you had I had to do some tuning and some hammering behind the scenes just to make the the bits and pieces work properly. I ran into an issue where one agent wasn't talking to another agent properly. However, that did only take a few hours to work itself out. But what I would encourage everyone to do is to look carefully at the inventory of content that you've already created.

Um, not too long ago, I was hanging out with a friend um at an event, and you know, she's like, I write on LinkedIn like two or three times a day. I just don't have time to write a book. I'm like, you already wrote the book. You just need to export your LinkedIn data, which we did, and it, you know, following a similar process, Claude was able to crank out 65,000 words of coherent text. Now it still needs editing, and Katie's book will still need editing, but that initial lift of, oh my God, I have to write a 65,000 word book is now I have to edit a book.

Right. Okay, I can tell I can I can stomach that. Well, I can say with confidence, I the thing that's missing from the course uh that I usually include in you know breakout sessions at events or in the newsletter is more of the storytelling and more of the anecdotes uh pulling from my own personal experience. I didn't want to distract students in the course by getting too far off topic. It was more of you came here to learn something, I'm here to teach you the thing.

In the book version, I would want to bring it back more of that storytelling of this is why I have this experience. This is why I know this, versus here's the thing, you came to learn the thing, here it is, go do the thing. And so I feel like that's going to be a big part of the editing process, at least for me, for a proper book, because to your story, um, I'm sorry, to your point, I'm a bit of a storyteller, and that's just not part of the current course. The course is the course, it's academic focused, whereas a book should be more narrative. Exactly.

The first word count, by the way, on the generated book that we just did is 65,900 words. 65? That's huge. I don't know what that means. I'll be honest.

So this is about 70,000. Ah, this is about 60. So it's between these two. Gotcha. So once I add in more of the storytelling and the anecdotes and the examples, it will be closer to Yeah.

Okay. And that's helpful because I feel like, well, and this is getting too much about me. If I feel like I'm going to do it, I want to do it right and have it be like a really good, not just to throw away, like, oh, hey, here's a compilation of everything I did. Which isn't to say that's how it comes out, but you know, I want to put my personal stamp on it. I don't, I don't want it to just be A, I put these things together, enjoy.

Exactly. So that's the process. I would recommend to folks. Give it a try, keeping in mind all the requirements. You're gonna need source data.

And that doesn't have to be an entire course. I mean I have written almost timeless my new book was written principally from voice memos from me foaming at the mouth for five hours on a drive to Albany. And in fact if you buy if you bought the deluxe edition you got a copy of the foaming at the mouth audio um oh yeah it's in the deluxe edition you can hear me swearing at people in traffic and stuff it's fantastic but I do that for two reasons one because it's entertaining but two it also provides lineage it provides um proof that the it was human originated. Yeah so you need you need the your source data you need ideally a generative AI system that has agents built into it that allow you to to process the stuff if you don't that's fine you can do it just with prompts it's just a lot more copy and pasting um you want to use generally speaking you want to use a reasoning model so we use for this book Claude Opus 4.1 but Gemini 2.5 chat GPT5 uh all those would be perfectly suitable um you want to have a good writing style prompt and you want to have good prompts to actually do the writing making sure that you do a test run on a a chapter or two before you kick it all the way off so that you can spot issues early and and fix them it's actually just like compiling software it really is I mean because the stuff that I've already written that's the requirements. So now it's just a matter of putting it all together with the technology, as you stated at the beginning of the live stream.

Exactly. Any final thoughts? Disney to cover. Yeah. Stay tuned for the book coming soon.

That's good for this week's show, folks. Thanks for tuning in, and we will see you all on the next one. Thanks for watching today. Be sure to subscribe to our show wherever you're watching it. For more resources and to learn more, check out the Trust Insights Podcast at TrustInsights.ai slash TI podcast.

And a weekly email newsletter at TrustInsights.ai slash newsletter. Got questions about what you saw in today's episode? Join our free Analytics for Marketers Slack group at TrustInsights.ai slash analytics for marketers. See you next time.