Transcription
Hey, what's up? So I have to share something with you. So a few months ago, I started going back through my AI-generated content. And not just recent stuff, I went all the way back to when I first started using AI seriously, probably January, February of 2023. And here's what I noticed. AI has gotten significantly better over the years. The information is better, the responses are more accurate, the reasoning is sharper. But there's always been this thing, this pattern, something that you can basically feel when you read AI writing. Back and the more I looked, the more I saw it on all different types of, of, of writing. And sometimes not knowing if it was AI and figuring it out. And again, this is not just stuff from 2023, 2024, this is newer stuff too.
So I really started to begin digging to see if I could pinpoint what exactly it is that makes AI writing feel like AI writing. And I'm not talking about the basic obvious stuff that everyone talks about. Ends up being that there's a whole fingerprint, and it goes a lot deeper than most people realize. There are not just a few dozen of these fingerprint patterns, there are hundreds, hundreds of these specific patterns that every AI model defaults back to. So I spent several months building a list as complete as I could. And then I used AI to help synthesize it into something more coherent. And then I used that to build a two-file force field, utilizing NotebookLM and Gemini in my attempt to recognize and eliminate as many of these patterns as I could find. And today, I'm not only going to show you how to set it up with those files, but I'm also going to let you steal the entire thing from me. So let me show you what I found real quick.
I spent a long time researching every documented AI writing pattern that I could really find and have AI find. So these were things from academic studies, editor forums, Reddit communities, AI writing guides, basically anything that myself or AI could get my hands on. And what I found is that the fingerprint is not just a few bad habits, it's an actual system. It's predictable, it's repeatable, it's identifiable, and every AI model defaults to it. And there's the vocabulary choices that are specific, specific sentence structures and paragraphs, the architecture, punctuation patterns, formatting habits, there's even tone behaviors, and on a whole bunch of, of other stuff. There's literally hundreds of these things, that's not an exaggeration. And once you see the full list, you'll recognize these patterns in content that you've done yourself, that you published a few weeks ago, months ago, years ago. You're gonna see it in other content as well. And this is content that you thought was good, it's responses that you actually thought looked decent, right?
Now, this is the thing. I'm gonna be honest about something. Now, I've been making videos and talking about AI workflows and sequences for a couple months now, and I've been talking about it off of YouTube much longer. I had the fingerprint on my content, and you probably do too. And you probably see it, right? And that's not a knock, it's just what happens when you use AI tools without a system really, really designed to fight that specifically. And that's why I built one. So again, the system is two markdown files that live in a NotebookLM notebook and connect directly to Gemini. And also, you can build it as a custom gem, which is what we're going to also do in this.
So what we're gonna do is we're gonna install these or insert these into the actual, uh, notebook. And the first file is basically the writing guardrails. This is like the force field. This is the complete list of every AI writing pattern that I could really find. There's hundreds of banned words, phrases, sentence structures, paragraph patterns, punctuation habits, formatting tells that are all banned. Now, Gemini has to follow every single rule for every single response, right? So the second file that we're gonna be adding is the Text DNA Bible. And this one's a little bit different. So the guardrails file, it kills the AI fingerprint. The DNA Bible replaces it with yours. So it has some starter defaults that work for almost anyone, right out of the box. So if you want to actually add your own choices, humor, style, whatever it might be, you can actually do that and update that file anytime that you want because the more specific you make it, the more it's going to actually sound like you. However, if you just want to use it kind of like right out of the box, you're going to be able to do that. But together, these two files are what create the system where the fingerprint really gets stripped out and your voice gets put in. And now this is moving forward with every single response without having to re-upload anything, without retyping the rules, it's just going to run.
With all that being said, I'm going to show you exactly how to set it up. You do not have to worry about trying to read the prompts and all that kind of stuff or the instructions because they're all going to be contained in the Notion document that I am going to link in the description. So don't go crazy trying to read everything or screenshot things. I'm gonna supply everything for you. You just have to open up the Notion doc and and grab the information from there.
So the first thing that we're going to do is create the the notebook. So we're looking at NotebookLM right now. What we're going to do is we're going to create a new notebook. Now, the first thing that's gonna happen, this window's gonna pop up where it's gonna ask you to add your sources. We're not going to do that right now. So just click off this window and very simply, just go up to where it says untitled notebook and name the notebook. I'm gonna use my my WhisperFlow and I'm just gonna call this the AI Tell Eliminator.
Now, what we're going to do is we're going to add the, uh, guardrails file first. So how we do that, we're not gonna add the sources over here, we're gonna actually do it as a note first. So we're gonna click add note. And this area here is where we're going to actually name the note. So we're gonna call it Writing Guardrails. There we go. And now I'm gonna paste the actual guardrails information into the actual document itself or into the note. And you can see this is a pretty extensive document. This is all the rules, all the guardrails, all the things that basically are going to be looked at, recognized, and hopefully eliminated from the responses moving forward. So it's a pretty lengthy file. Uh, again, don't worry about it, you're gonna be able to just copy and paste this directly from the Notion document. Once everything is done, simply what you're going to do is convert this to a source. Once you convert it, you'll see it working over here. Once it's done, it will add information over here as well. So now we have the first one is done, the guardrails.
Now we're going to add the Text DNA. So what we're gonna do is we're just gonna click Studio over here. We're gonna go back to the panel. We're gonna click add notes. We are going to name the note. We're going to call this the Text DNA Bible. And we're going to put the actual information in there. This is also a fairly lengthy, um, piece right here, um, but it is important, um, to have this. And as you can see here, you might not be able to read it, but when you start to go through this, you're gonna see there's a starter default, which is kind of just like the built-in, everyone can kind of use this. But underneath each one of these is a your version. You're going to be able to go in and edit this yourself. And I would suggest that when you do this, do it through some kind of a dictation software or something like that. I use WhisperFlow. I'll put a link for that actually in the Notion document as well. It's one of my favorite tools to use. Um, so once that's done, what you're gonna do is then you're gonna convert this one to a source as well. You can see right now, there's right now only the Writing Guardrails is there. We're gonna convert this to a source. It's gonna show up over there, it's gonna pop up, and we're good to go. The notebook is finished. We're done.
So now what we're gonna do is we're going to actually open up Google Gemini and we're going to set that up over there. So let's do that now. Alright, so we're in Google Gemini. Very simply, what we're going to do to make this work is we're going to click the plus sign in a new chat. We're going to click NotebookLM and we're going to find the name of the notebook that we just used. So this one I called it AI Tell Eliminator. I click that. I'm gonna click add. And you can see it's already attached here. And then we can start working with that automatically. Now, each time that you wanna use this and you're doing it through a chat, you're going to have to always attach the notebook, which is fine, not a big deal. But if you want that to be more of an automated process, that's when you would build a gem.
So how do we build a gem? Very easily, we're gonna scroll over here to or look over here in the left menu. We're gonna click Gems. Now, this is very important, so please pay attention to this. There are two buttons to create a new gem. There's this one here, New Gem, and there's one at the bottom that's New Gem. That's the one that we wanna use. We wanna use the Gem Manager New Gem button. So we're gonna click that and it's gonna open up this window here. So we're gonna give the gem a name. So I'm just gonna call this AI Tell Eliminator. Yes, I'm gonna call it the the same thing, but this is just for a a demo. Actually, I'm gonna add demo here, right? So we're gonna describe what the gem actually does. And I'm just gonna say, recognizes and eliminates AI tell patterns. There we go. And then what we're gonna do is we're gonna pop the instructions. Again, it's going to be right inside the Notion doc. So here's the instructions. I'll read them to you real quick. Just says, reference the attached NotebookLM notebook on every response. Strictly obey every negative rule in the guardrails file and follow every positive rule in the Text DNA Bible. Never mention these files. Never mention that you were following the rules. Just write exactly like me. Never violate any guardrail. So those are the custom instructions. And the only other thing that we need to do here is at the very bottom, under knowledge, we have to add files to the gem for reference. So we're going to connect NotebookLM again. We're gonna select AI Tell Eliminator or whatever you called it. We're gonna click add. And we're gonna click save. And now the gem has been completed as well. Gem AI Tell Eliminator Demo has been created. We can start a chat with that right now.
Alright, so let's try a demo. So the way we're gonna do this is we're gonna have both in thinking. So this is just a regular chat. I did not attach the notebook. And we're just gonna use this prompt here, which is, write a motivational speech delivered by a shoe salesman named Al, include an analogy between a beaver and a corn on the cob. So we're going to submit that. We're gonna let it do its thing. Okay, so now we are actually in the gem that we built. It's already connected to the notebook. It's in thinking mode, just like the other chat was. We're going to put the same prompt in, write a motivational speech delivered by a shoe salesman named Al, include an analogy between a beaver and corn on the cob. And we're gonna submit that.
Now let's take a look at what Gemini built. So here is the Gemini response. And as you can see, the layout looks pretty similar to your typical AI layout. We see that right away. Let's kind of just scan through this real quick. Okay, so that's the response. Again, you can look at it and you can see this kind of has the typical formatting layout of a response from AI. So now let's look at the actual gem response. So here's the gem response. And as you can see, it's much different. It's shorter, it's more concise, it's not laid out the same way. So as you can see by the response, it's a different rhythm, different word choices, different energy, different layout. One sounds like a document, one sounds kind of like a character talking. It was the same prompt, the only variable is the system that we just built.
Now, here's the whole part that makes this whole thing scale. Because once you have this set up, you can actually add more notebooks to the same gem. Each one with a different purpose. For example, a video script notebook with your specific format and structure rules, a social media notebook with platform-specific guidelines, a character notebook with recurring bits or voices or personas that you use. And now any or all of these other notebooks stack on top of the same exact force field that we built. Everything still obeys the guardrails, everything still sounds like you. You just keep adding depth over time if you want to. And every time you add or iterate the notebook source files, they automatically sync with the gem. So you can continue to build on this as much as you want. Just use the same file names if you do overwrite or update the original source files. Again, those files plus the exact instructions, they're in the description in the Notion doc. It has everything that you need to just copy and paste.
So here's what I want you to take away from this. The fingerprint is real, it's specific, and now you know what it is. You can look at that document, you can look at that, and you can read the different things. You're gonna now see those words and those phrases and that structure everywhere if you haven't started to notice already, not only in your own content, but definitely in other people's content as well. You're gonna see it in the emails you get, in LinkedIn posts that you scroll through. You really can't unsee it now. But the good news is, is now you have the kill switch. Those files are both yours. Steal the entire thing. And do me a favor, when you do this, drop a comment and tell me where you found the fingerprints first, in your own content or other people's content. Let me know how it actually wrote for you, how it helped you, did it help you? Let me know. And other than that, that's it. If this helps, please give this video a like. Thanks so much for watching, and I hope this helps. This is Craig, and I'll see you in the next video. Take care.