Welcome to the deep dive. We are going on a very specific mission today. But before we get into the mechanics of it, I want you to picture something. Oh, I know the exact scenario. You finally sit down to work on a major project, you open up a totally fresh chat window, and immediately you just feel this agonizing sense of exhaustion. Yeah. Because you suddenly realize you have to upload the exact same PDF. You have to explain
your entire brand voice. Again, every single session basically starts from absolute zero. It is a profound friction. I think we are constantly introducing ourselves to our own tools. We totally are. It's exhausting. Today, we are looking at a brilliant guide on curing AI amnesia. The goal here is simple. We're going to build a long -term memory system for your work. And the best part is we'll use two completely free tools for this,
Cloud Projects and Notebook LM. There is no coding involved, there are no complicated scripts, and it takes under 20 minutes to actually set up. Yeah, and once you finally build this system out, you'll never have to brief your AI from scratch again. It honestly changes everything about how you work day to day. Let's explore the hidden cost of this amnesia first. Before we can actually build the solution, we have to understand it. Why is the current way we use
AI quietly draining our productivity? Well, it's sneaky. It feels like a minor annoyance in the moment. But it actually scales terribly across a work week. Right. Every time you start a new chat, you're wasting valuable time. You're wasting your own mental energy just doing setup work. The source material uses a really great analogy here. Claude's memory is basically like a dry erase whiteboard. The exact moment you close that chat tad, it's wiped completely clean. Exactly
that. It just vanishes into thin air. Now Claude does try to synthesize your conversations every 24 hours, right? It does, yeah. But that synthesis only applies to new standalone chats. It's very broad. So if you had this incredibly deep work session on Tuesday, By Friday morning, it is entirely forgotten. You are completely back to square one. Yep. You're re -explaining all the core concepts before you can do any real work. And then there is the token tax. Oh yes. This
is the part almost nobody talks about. Let's define that quickly for everyone. Tokens are tiny pieces of words AI uses to charge you. Perfect definition. So, I mean, think about pasting a 5 ,000 -word brand guide. Right. If you drop that into every single new session you open, you are repeatedly paying for that exact same data. Over and over again. Exactly. Across a busy month, that financial cost really adds up. It is deeply inefficient, but I think there is
also an intangible loss here. The text describes it as losing the feel of your project. Oh, this is so incredibly real. When you work with an AI for a solid hour, it actually starts to learn your unique rhythm. It picks up on your tone. It learns exactly what you're going to push back on. It starts to feel like an actual creative partner. And then you just close the tab. It's gone. That shared feeling completely resets. The very next morning, it is a total stranger
again. You lose that incredibly valuable shared context. It's tragic, honestly. I actually have to admit something here. I still wrestle with this terrible habit myself. I catch myself pasting the exact same context brief three times a week, just out of sheer laziness. It's such a ridiculously common trap. We just default to the easiest immediate action in the moment. Even if we know it is going to cost us more time later. Exactly. So why is this specific cycle so incredibly hard for us
to break? Well, because the AI sounds so confident. It's conversational every single time. So we naturally project human traits onto it. We subconsciously assume it remembers us, because the human colleague would. And when it obviously doesn't, we just panic -paste our documents to catch it up. We project human memory onto a machine that sounds confident. Yeah, that's it entirely. Which leads us directly to the actual fix. To stop starting from zero, we need to combine two very different
tools. Wait, I want to push back on that a bit. Isn't the whole point of these massive AI models that they do everything? That's the marketing pitch, sure. So why would I complicate my life by using two different tools? Because people fail. when they expect one tool to do absolutely everything. You can't use a hammer to turn a screw. Ah, I see. Claude Projects and Notebook LM are built for entirely different jobs. When you finally put them together, the entire system
just clicks. Okay, let's look at that first tool then. Claude Projects. This basically acts as your persistent, everyday workspace. Right. You upload a file once and it stays there forever. It uses AI to constantly remember your unique context. And just to clarify, Rag, for a second, it is a way for AI to pull your exact files instantly. Yeah. Great way to put it. So inside Claude projects, the AI can constantly read those files. Right. But it can also still search the broader open
web. It still has its vast general knowledge. Yeah. Your specific files just add an extra layer of personal context. But Notebook LM works in the exact opposite way, doesn't it? It acts strictly as the archive. This is the massive difference. Notebook LM only pulls from the specific documents you provide it. Interesting. It has incredibly strict data adherence. There is absolutely no
external web knowledge allowed in there. So it is kind of like having a brilliant creative director in Claude and then an incredibly strict literal archivist in Notebook LM. They're doing totally different jobs. That's a really great way to visualize it. They're a team. Now here's why we actually do this workflow manually instead of using a script. You mean advanced things like MCP. Right. Which is basically code that connects apps together automatically in the background.
Yeah, exactly. Advanced users absolutely love setting up those automated complex scripts. But they require you to do complicated terminal work. You constantly have to authenticate tokens through Google. That sounds fragile. It is. For most normal people, those setups just break without any warning. a random library updates overnight, and boom, your entire workflow is totally dead right before a major deadline. So this manual method is just much more resilient. It is completely
free. There is zero coding required. Not at all. You can even do it on your phone. It stays perfectly stable. It actually stays working. That's the entire point of building it this way. I want to linger on Notebook LM for just a second, though. Why are strict boundaries sometimes better than having endless web knowledge? Because the open web is totally full of contradictions. It's full
of generic advice. That is very true. When you're making critical business decisions based on your own specific client data, you want the AI locked in a room with only your proven facts. Endless internet knowledge dilutes the truth when you need factual boundaries. Yeah, exactly. So let's get hands -on and actually build out this creative director's workspace first. Let's do it. Setting up Claude projects. The guide is incredibly clear
about the pacing here. You do not need to do absolutely everything in one single sitting. You start simple. People always overcomplicate this step. You just go into Claude. You click on new project. Right. You give it a highly specific name, something like newsletter Q2 or client strategy redesign. Inside that project, you will see a file section. This acts as your new long term memory. And the golden rule here is. Exactly three files to start. Right. Do not upload 20
chaotic files on day one. Just three. First, you need your context. .md file. This explains who you are, your core goals, your specific tone. Keep it super lean, under 500 words. Second, you upload your formal brand guidelines. Third, and most importantly, your decisions log. The decisions log. This is basically a running document of choices you have made and the reasons why. Yeah. Once those three are finally uploaded, you set up the project instructions. These are
basically standing rules. Claude has to follow them before you even type a single word. You literally tell it, check my decisions log before answering. always match my brand guidelines. You essentially hand it a custom instruction manual for your own brain. But I know people assume AI just naturally learns over time. Oh, they do. Without this next manual step, all that brilliant brainstorming just vanishes into the
ether. It vanishes completely. You absolutely cannot just close the tab when you're done working. Right. At the end of every single session, you have to run the wrap -up habit. The prompt they suggest for this is actually really simple. Summarize what we worked on, decisions made, and next steps. Keep it under 200 words. You just copy that short summary, you paste it directly into your decisions log document, and you re -upload it to the project.
The whole thing takes two minutes, Tops. Literally two minutes. So why is this specific decisions log the Absort secret weapon here? Well, human memory is honestly terrible. I mean, we forget what we had for breakfast, let alone why we discarded a marketing angle three weeks ago. Yeah. If the AI doesn't have a log of that specific choice, it's just going to suggest the exact same bad angle next month. It records the logic of your
choices to prevent recurring arguments. Yeah, it creates a perfectly linear history of your own logic. We are going to pause for a brief moment. When we come back, we will explore adding the heavy research layer, mid -roll sponsor read placeholder. Welcome back to the deep dive. We have successfully built the active workspace using Claude, but Claude naturally hits upload limits very fast. It really does, even if you're
on the expensive paid plans. You upload a few massive research PDFs and you immediately run out of contextual space. So where do we actually put those massive industry reports? Where do we store our giant competitor analyses? This brings us to the second tool. Notebook LM. This serves as your heavy research layer. It comfortably handles up to 50 different sources per notebook. Right. You can upload dense PDFs, long Google
Docs, or endless website links. You can even drop in YouTube transcripts and huge audio files. And you just organize them by topic. Put your audience research in one notebook, competitor landscape goes in another, industry trends in a third. The actual workflow here is fascinating. You do not ask Claude first. Right. Never ask Claude first for hard research. You go directly to Notebook LM. You ask a highly specific, targeted question. Something like, what are the top three
audience patterns found in these sources? And Notebook LM gives you a grounded, perfectly cited answer. It is only using your uploaded files. Exactly. You then take that clean answer, copy it, and paste it back into your Claude project. You basically tell Claude, here is my verified research. Based on this exact data, write out the new strategy. Whoa. Imagine having your entire project history perfectly recalled in an instant without a single hallucination. It is a superpower.
You get the strict factual recall of Notebook LM seamlessly combined with the creative reasoning of Claude. It's the best of both worlds. A quick note on data safety here. People always ask if their private data is actually safe. Notebook LM is officially a Google product. It follows standard Google workspace privacy terms. So it's perfectly fine for general market research and creative drafts. But you still need to be careful.
Right. Don't go uploading highly sensitive legal or financial data unless you explicitly clear it with your IT team first. Let's talk about prompt drift for a second. How exactly does this dual step process prevent the AI from slowly drifting completely off topic? Well, when you just chat loosely with an AI for hours, it naturally starts losing the plot. It gets overly creative,
just starts filling in the blanks. By forcing the AI to start from notebook elements hard facts every single session, you constantly reset its compass back to reality. Anchoring Claude to factual sources stops it from making up creative data. Precisely. It keeps the final output highly rigorous. Let's look at how this actually applies in the real world. Real use cases. Because this definitely is not just a system for writers. Not at all. Think about digital marketers for
a second. You create a totally separate cloud project for each individual client. Okay. You upload their specific brand voice. You upload all their past campaign results. So every single time you open up that client's project, Claude already knows them intimately. Right. There's no tedious re -briefing. You just jump in and ask, what specific content format worked best for them last quarter? Educators are using this heavily too. You upload your massive core site
line. You upload anonymous student feedback. You can easily ask Claude to rewrite a confusing lesson based entirely on your specific teaching philosophy. And for solo founders, this setup is just brilliant. They call it creating a business brain project. This is honestly my favorite application. You upload your big yearly goals, your external communication guidelines, your internal pricing rationale. Yeah. You can actively use it to fight against scope creep. A client wants a brand new
service, right? Right. You ask, based on my past decisions about scope, what should I actually propose here? It essentially acts as a perfectly objective business partner. It forcefully holds you to your own established rules. It totally does. But obviously, any system has its traps. The guide highlights four major mistakes that will completely break this setup if you aren't careful. We definitely have to cover these. Mistake number one is uploading way too much data all
at once. People always want to just dump 30 chaotic documents in on day one. Yeah, you have to start with three to five. Mistake number two is never bothering to update your core files. Your business context changes. Your goals shift. You really need to review those core foundation files every four weeks. It's a really small habit, but it makes a massive difference. Mistake number three is skipping the wrap up. We said it before, but
it absolutely bears repeating here. Without that two minute summary step at the end of a session, you are not building a memory. You are just having totally isolated, forgettable chats. Yep. And mistake four is using Claude projects to hoard all your heavy research. Right. Do not treat Claude like a giant filing cabinet. That is exactly what Notebook LM's job is. You have to keep the tools in their respective lanes. Keep them separated. I am actually curious about that very first mistake.
Why does giving an AI more files actually make it less smart? Shouldn't more data be better? You'd think so, but no. It's this phenomenon called context dilution. OK. When you give it 50 vaguely related files, the AI's attention mechanism really struggles to weigh what is actually important. It might pull a totally mediocre point from an outdated file instead of the critical, nuanced point from yesterday's file. Too much background noise hides the critical signals the
AI needs. Yeah. Keep the active workspace lean. Keep the archive layer deep. Let's recap the big idea we were exploring here. We have all been paying for this incredibly powerful evolving partner, but we have been treating it like a basic, completely amnesiac question answering machine. We've seriously been limiting its true potential just because of our own lazy habits. The roadmap to fix this is genuinely simple. Week one, build the Claude Foundation. Right.
Right. Your short context file. Upload it. Start practicing that two minute wrap up habit. Week two, you add the heavy research layer. Set up your notebook LM, add your massive reference sources, but strictly keep the two tools separated. Absolutely crucial. This fundamentally transforms the tool into a true creative partner. It finally stops being a complete stranger every morning. Because it already knows where you are. It knows exactly what you are building toward. So here's
the immediate challenge for you. Okay. In your very next work session, don't just close the window. Force yourself to ask for that 200 -word summary. Start your decisions log today. Stop starting from zero. It is a relatively small shift with incredibly profound leverage, but
it really does make you think. If you successfully build an AI partner that perfectly remembers every decision, every pivot, and every preference you have ever had, five years from now, where exactly does your original thinking end and the AI's memory begin?