You type a simple question into the chat window. It just spits back a generic robotic wall of text. Yeah, and it's practically unusable for your actual daily tasks. So you tweak the prompt, you try it again. But the result is exactly the same frustrating garbage. Welcome to the deep dive. That is a deeply relatable frustration for all of us. It really is. Today, our mission is unpacking a very popular mindset. People constantly tell you to learn prompt engineering. Right.
They say to study frameworks and memorize these arbitrary rules. We're going to explore why that is pointless. It is honestly a massive waste of your time. A total waste. Instead, we're outlining a totally different approach today. We will build a custom prompt optimizer from scratch. You can build it using totally free Google tools. And it really only takes about 10 minutes to set up. I have a vulnerable admission to make right here. I still wrestle with getting generic robotic
answers myself. Oh, yeah. Yeah, I just throw my messy thoughts at the screen. I expect the machine to magically understand my intent. But it almost never actually does what you want. Well, you're definitely not alone in that specific struggle. It was a very common friction point for everyone. So our roadmap for the discussion today is clear. We're going to explore why AI constantly guesses. Then we will build this optimizer in two steps. Finally, we'll plug it into your
daily workflows. Let's start with that guessing game you just mentioned. Before we can fix the AI, we must understand it. We need to know why it gives us terrible answers. Especially when we just type naturally, you know. Right. The core issue is that AI isn't your friend. Think about talking to a very close personal friend. Okay. You might say you want to lose some weight, but you also admit you absolutely hate the gym. Right. A good friend knows your personality very
well. They might suggest swimming or... taking a long walk. Exactly. They tailor the advice to fit your specific lifestyle. But the AI does not know your daily habits. It doesn't know what you actually like doing, so it just tries to guess what you want. You throw a messy idea at the chat window. And it gives you a boring 10 -page workout plan. It's a rigid plan you're never going to use. It is incredibly frustrating to deal with that output. The source material
introduces a brilliant core analogy here. Yeah, I love this part. It calls the prompt optimizer a cleaning machine. It's essentially a cleaning machine for your brain. I really love that specific framing for this tool. Instead of stressing to write a mathematically perfect instruction, you just throw in your messy half -formed human thoughts. The system automatically cleans them up for you. It turns a simple sentence into a highly structured prompt. It fills in the missing pieces of your
logic. Right. And the AI then knows exactly what you actually need. It stops making wild guesses about your underlying intent. Yeah. Beat. But that brings us to the dreaded fluff problem. Oh, the fluff problem is absolutely everywhere in AI today. The model thinks talking a lot means being helpful. It adds these incredibly long, useless introductory paragraphs. It gives you extra advice you never even asked for. It equates word count with actual value and intelligence.
It really does operate on that exact flawed logic. When you give a short prompt, you leave empty space. And the AI desperately has to fill those empty gaps. It starts guessing what your true goals might be. Let me ask you this. Why does the AI always generate useless fluff when I ask a short question? Well, it tries filling the empty space by guessing your goals. It just plays it safe. It assumes high word counts mean it's being helpful. Right. It talks too much because
it thinks length equals value. Exactly. The optimizer acts like a professional editor sitting beside you. OK. It finds the structural holes in your messy story. It fills them in automatically before the AI answers. You avoid doing any of the heavy lifting yourself. Two sec silence. So we know the AI desperately needs an editor. Definitely. Needs a rigid framework to stop the endless guessing. Let's walk through exactly how to build that editor. You just need a completely free Google
account. OK. The setup is way easier than it might sound. You don't need to be a complex tech wizard. Good to know. You need Notebook LM. Gemini, and two simple text files. That is your entire toolkit for this entire build. Just to be clear on the technical requirements here, we're just using basic text files for this setup. That's it. Let's walk through the contents of file one. This file contains your core system instructions, right? Yeah. This is where you establish the
permanent ground rules. You open a blank document on your computer screen. You tell the system exactly how it should behave. You tell the AI it's a world -class prompt engineer. That is the exact phrase to use right there. You tell it to read the user's messy input. Then you ask it to create a JSON schema. Which is just a very organized list that keeps the AI boxed in. That is a perfect way to describe the mechanism. It forces the AI to answer specific targeted points.
It stops the endless wandering through random topics completely. It cuts out the useless conversational fluff right away. File 1 also asks for a brief thinking guide. We'll dig into the mechanics of that shortly. You keep the overall tone professional, clear, and helpful. Everything outputs neatly inside a single formatted code block. You save that text file to your local computer. Name it prompt -optimizer -system -instructions for clarity. Next, we move on to building file number two.
This is your very important example pairs file. Ground rules are great for setting the baseline behavior, but the AI also needs to see tangible, good examples. This second file might actually be even more important. It shows the AI the exact style you want. It operates in a very simple before and after logic. Let's look at a specific example from our source. The before state is a messy, raw user thought. Like you might just type, I want to learn to code. Right. That is
a very broad, messy, and unstructured goal. The after state is what the optimizer builds. It becomes a highly structured, detailed JSON plan. It specifies learning Python specifically for an absolute beginner. It outlines a manageable weekly plan for the user. It also adds that crucial thinking guide we mentioned. It tells the AI to focus on underlying logic first. It explicitly says to avoid using complex technical jargon. It provides small daily wins for ongoing user
motivation. You save that second file to your computer too. Name it prompt. Optimizer -example pairs to stay organized. Now you have your two core structural files ready. You open a brand new notebook in Google's Notebook LM. You upload both of these text files as sources. Then you do something very important for the system. You click the Convoke to Source button for both files. This tells the system these files are the absolute truth. It must follow them without any deviation
at all. Whoa! Imagine turning a blank text file into a professional engineering brain in seconds. It really feels like unlocking a weird new superpower. Totally. The instructions mention a chain of thought process. What exactly is that doing behind the scenes? It acts as a mandatory guide for the model. It forces the AI to explicitly plan its logic. Oh, okay. It makes it think through the problem step by step. It does this before
it starts writing the final answer. Got it. It gives the AI a clear thinking path to follow. Right. And that eliminates the guessing game completely. The AI has a concrete plan before it ever speaks. Two -sec silence. We have built the foundational brain in Notebook LM. Before we connect it to our chat window to see the magic happen, let's take a quick break. Sponsor. And we are back. We just built our custom prompt optimizer in Notebook LM. Now we need to connect
it to our daily workflow. This is where the real practical magic kicks in. Linking the two distinct tools is actually quite simple. It really only takes a few very quick clicks. First, you open the standard Gemini website interface. You start a brand new chat session right there. Look closely at the prompt box at the bottom. You will see a little plus button sitting there. You click that plus button to open the menu. A list of
integration options appears on your screen. You select Notebook LM as your primary data source. Then you choose the specific notebook you just built. You confirm the source is active in your chat window. From that exact second on, Geminite checks your instruction files. It checks them every single time you send a message. You do not have to keep uploading the text files. You never have to re -explain the basic ground rules. The notebook stays permanently connected to the
chat session. By attaching that notebook as the primary source document. You're giving Gemini a rigid set of boundary conditions. It cannot just pull from its vast, messy training data. It has to filter everything through the brain you built. That brings us directly to the two -step workflow rule. This is the core workflow for your daily tasks. Step one. Type a messy idea into your new notebook LM. This quickly gives you the highly structured smart prompt.
Step two is just as simple as the first. You copy and paste that smart prompt into Gemini. And Gemini then gets the actual real work done perfectly. Let's look at some real -world examples from the source material. These show exactly how powerful this two -step workflow is. Let's start with a very relatable daily fitness example. I think everyone can relate to this specific scenario. You are completely exhausted after a long day of work. Your daily schedule is an
absolute total disaster right now. You do not have the energy to write perfectly. You don't need a mathematically perfect instruction here. You just type out your messy, chaotic daily reality. You type, I am tired and want to be fit, but I have zero time and I absolutely hate running. Without the optimizer, a standard AI just says, try harder. It gives you a generic 5K marathon training plan. But with the optimizer, it builds a tailored structure. It creates a detailed JSON
list just for you. It understands the actual underlying problem you are facing. It specifies 15 minutes of gentle, low -impact exercise. The thinking guide tells the AI to focus on consistency. It strictly deprioritizes intense exhausting daily workouts completely. You just copy that professional clean code block output. You paste it directly into your Gemini chat window. You get a totally realistic 15 minute daily workout. It fits your messy chaotic life absolutely perfectly.
Let's look at a completely different technical coding example. Let's say you want to learn Tailwind CSS programming. If you just ask AI to teach you, it fails. The topic is way too broad for a simple prompt. The standard AI answer will be incredibly confusing. It will overwhelm you with advanced technical details immediately. But you put that messy idea into Notebook LM first. The system realizes you need a clear step -by -step path. It actively stops you from feeling
totally overwhelmed right away. It systematically organizes the topic into core foundation. concepts. It creates small, manageable daily coding challenges for you. The thinking guide forces the AI to explain the underlying why. Like why use Tailwind CSS instead of older CSS files. You take that beautifully optimized prompt from the screen. You paste it into your preferred AI chat tool. A clear, professional step -by -step lesson plan appears immediately. The final practical example
is about growing a business. Growing a weekly newsletter or getting more Twitter followers? Do not just ask the AI for a generic growth plan. You type your raw, unpolished goal into Notebook LM directly. The optimizer acts exactly like a professional marketing consultant. It generates a strategy tailored for your specific brand niche. It builds a rigid content structure for your specific niche. It clearly defines your core daily content pillars. It maps out a realistic
daily social engagement strategy. The underlying thinking guide is absolutely crucial right here. It tells the AI to focus on providing real value. It strictly avoids annoying your followers with constant ads. You copy the resulting code block from the optimizer. You paste it back into your main Gemini window. You get a brilliant 30 -day content calendar with great headlines. The real secret is knowing which tool to use when. Notebook LM is just the filter that cleans your messy
thoughts. Gemini is the engine that does the actual heavy lifting. It is exactly like having a dedicated professional coach. It stops embarrassing mistakes before you even start typing. So I generate this smart prompt using Notebook LM. Am I locked into only using Gemini for the final result? Not at all. You can copy that optimized structure. You can paste it right into ChatGPT, Claude, or any other model you prefer. So the cleaned up prompt works perfectly in ChatGPT or Claude
too. Absolutely. The prompt structure is already thoroughly cleaned up. It will work beautifully on literally any modern AI model. Two -sec silence. This basic workflow clearly saves hours of daily frustration. Now let's explore leveling up this entire system. We can tweak it to fit perfectly with highly unique tasks. Small incremental changes make a truly massive difference here. You do not have to rebuild everything from scratch, ever. You just push a few buttons to alter the
overall vibe. Let's discuss changing the AI's underlying conversational tone. You can easily change the personality of your custom optimizer. You just swap out one single line in the core instructions. You might want a patient and clear teacher for learning. That tone is great for grasping new, complex topics. But for daily business tasks, that tone might feel too soft. Right. For serious business, you want something much sharper. You chained it to a direct and data
-driven consultant. You want crisp, highly actionable insights without the fluff. For purely creative projects, you try an inspiring and imaginative coach. That one single line completely shifts the semantic space. It changes how the AI responds to your messy thoughts. The source material also introduces gems inside of Gemini. If you use Gemini constantly, attaching the notebook gets tedious. Doing it every single time causes unnecessary daily friction. The gems feature fixes that annoying
repetitive hassle completely. Think of a gem as a little dedicated shortcut app. It lives permanently right there in your left sidebar. It saves you so many unnecessary manual clicks daily. You don't have to attach the notebook every single time. You just click your custom optimizer gem and start working. You start typing your messy idea into the box immediately. It keeps everything you need perfectly organized in one place. It is a massive time saver for
writing emails or planning dinner. You can also make the AI significantly smarter over time. You do this by adding more example pairs to your logic file. This is where the real personalization actually starts happening. The more before and after stories you show the system. The smarter and more refined its outputs eventually become. It gradually learns your highly specific personal preferences and quirks. It is exactly like teaching
a friend how you like your coffee. Once they see you make it a few times, they stop asking. They just intuitively know exactly what you want. If the AI consistently gives you way too much text, you just add very short punchy examples to the file. If it misses crucial details, show it a JSON with many fields. Every single new pair makes the system feel more like you. This system works incredibly beautifully for technical
coding as well. It makes absolutely sure you include important technical details, things like web accessibility, proper meta tags, and clean structure. It actively prevents you from getting sloppy buggy code back. Even if you think your prompt is already pretty good, you should still run it through the optimizer system anyway. The source notes it improves good prompts by 20 to 30 percent. It consistently finds a reliable
way to add a clearer thinking path. It actively catches the small technical details we naturally forget. What if I want to type my messy ideas in a language other than English? Will this system still work? Yes. Gemini is excellent at translating, so you can type in your native language and the system will still automatically optimize the logic. It takes your native language and still optimizes the prompt perfectly. It seamlessly handles the complex translation and the logic
cleanup simultaneously. It is an incredibly robust system for international users. Beat! We have covered a massive amount of ground today. Let's zoom out and recap the really big overarching idea. The biggest takeaway is shifting your fundamental mental framework. You do not need to become a professional prompt engineer. You do not need to memorize complex, rigid conversational frameworks. Your instruction text files are the absolute undeniable truth. Notebook LM is simply the filter
for your messy human thoughts. Gemini simply takes the required action. based on that clean logic. You just need 10 uninterrupted minutes to set up the brain. Then you just let the automated filter do its daily job. It completely changes the dynamic of how you work. You stop fighting the machine to get what you want. You start actually collaborating with it on a structural level. I strongly encourage everyone listening to try this out. Do not wait for a massive, overwhelming
project to begin. Start with something very small and incredibly low stakes today. Plan a simple dinner menu for the upcoming week. Plan a very simple weekend trip to a nearby city. Type in your messiest, least organized thoughts about the topic. Just sit back and watch how the system cleans it up. See the massive difference in the final structured output. Once you see the power, you're never going to go back to guessing. It really is a fundamental paradigm shift in daily
computing. I want to leave you with one final thought to mull over today. If two free tools can take our messiest half -formed thoughts and translate them perfectly into structured machine logic, how long until we stop typing instructions altogether and just let AI listen to our messy conversations to run our lives for us?