Make Notion Your AI's Brain (Notion's Official Setup)

0:00People are using Notion as their second brain because it works so well, but there's a problem. You see, it works well because it's in the cloud, which means you don't have to maintain it, you don't have to worry about your computer being up, it just works. You can easily read files because Notion is designed to be pretty and accessible. You can share files with other people, so if you have something in your second brain that you need a teammate or a spouse or a friend to see, you can share it easily. And you can always export out of Notion if you're not happy with them and want to

0:26use a competitor. All those reasons make Notion a great place for your second brain. But the problem is that your agent forgets to save data into Notion. And sometimes it forgets that your important data is in Notion, and so it looks all around, it says it doesn't have it. So, how do you ensure that all your data lives in Notion and that all your agents know where to get it? Well, today we have a solution. The person at Notion who's in charge of making it more agentic is joining me to show you how to set it up the right way. We're going to get into it, and I've got links below to

0:59guide that will make it easy for you to set this up. Let's get into it. Presented by Zapier, the AI automation company. How do you set it up? What are we looking at here? This is a fresh Notion. I just signed up. Um, and then I've got a Claude. So, I'm over here, I've connected my Notion. This is just in the basic, like if you go to customize and connectors, uh, Notion, you can see it's connected

1:20there. Um, so let's go back to a new chat. And we've created this prompt. developers.notion.com/prompts/setup But I I never wanted to go there because it's a developer tool, and I'm not looking to code something. What is this supposed to be? This is This is incredible alpha because you're so early, it's like hidden in

1:38this developer document thing. Okay. Um, [laughter] though it's going to it's going to become a much more, uh, mainline part of the flow, uh, very soon. But but this is the way to get it for now. So, uh, you go to you go to back to Claude and you say something like, um, actually I think we have like a little It says this page contains instructions for your agent to use the Notion MCP as effectively as possible. To get started, share this page with your agent. So, I'm going to say can you help

2:05me set this up? As simple as that. And so, it's going to read the page and it's this whole set of instructions on how to use a Notion as basically like the data backpack, right? So, no matter what agent you use, whether it's, you know, Muse or Crock-Pot or Instinct or ChatGPT or Claude, they can

2:26all talk to this. Yes, I can read it. Amazing. It is Notion's official setup prompt. So, we have like versions and when it was last published. Okay. It's written for an AI agent to follow. Goal is to make Notion the shared memory and workspace that all of your agents use. Why it exists. Agents like me lose context between chats and between products. Notion's argument is that your memories, decisions, documents live in

2:45Notion. You own them. What the setup would involve. We can go into that. And then, uh every one of these changes requires your approval first. So, a very important thing about this prompt is we're very careful to not try and prompt inject or get an agent to do anything that you don't want it to do. It's purely like tell the user this is what we're offering and do whatever they're asking you they actually want you to do, but don't, you know, do anything without I see. So, it is explicitly now getting a new a prompt that tells it from now on, I want you to go check Notion before

3:17you do anything. Also, you're telling it upload all your memories that you have of me into my Notion account so that I have that in Notion, too. And now you've got this collection of information that is the single source of truth and what Claude is telling you is, "Hey, you know what? This could overlap. You might have two different sources of truth." And you're saying that's okay. So, let's go through

3:35it. Let's do it. Yeah, amazing. Okay. Okay, great. This all sounds good. I would like to use Notion as my source of context. What's first? If it's going to have to do searches on Notion it, in addition to its own memory, is every result that I get now from Claude going to take a lot longer? Um when you ask it something like, "What time is the baseball game tomorrow?" it knows I'm not going to need to go to your

4:04notion for that. Um if you ask something like uh "Hey, what did we decide last week about XYZ project?" it's going to you know, search your notion. Um so, for simple things where it clearly doesn't need notion, it won't change at all. Uh and then for things where it's on the margin, potentially it could get a little bit slower, and that's just sort of the trade-off of like having a sort of centralized data backpack. So, yeah, that's that that part of it is a little bit unavoidable, but from my experience, most of the things I ask agents to do, I treat as basically async anyway. I fire off a prompt, and then

4:41I'm doing something else for at least like, you know, a minute or two until I come back and it's done. What's the difference between that and the way that we're doing it now, and the way that I've been doing it, which is just integrating notion and telling the agent to store data in notion and go

4:54grab it from notion? The big difference is this is just more reliable to actually use notion without you having to specifically prompt it every single time you want it to use notion to use notion. Mhm. That's just really the biggest difference. Okay. Because I'm now having a first-level instruction before my before my prompt. Okay.

5:14Yeah. Exactly. I'm finished at 10. So, finish the discovery step. Your workspace It's my workspace, and it's empty. This is what we expected. It's a brand new workspace. So, one thing to check earlier in some earlier chats here, uh we've got all these like things. None of that is in the workspace that's connected now. That's so funny. So, this is I disconnected and reconnected so we

5:34could get like a fresh start. Mhm. [clears throat] It's got a whole bunch of uh stuff that it remembered from Claude's memory. So, this is actually sort of a benefit cuz it could import what Claude remembers about me to my notion, even though I disconnected a uh I I connected a separate notion. So to start, it's basically like I propose we import a bunch of useful configs in your notion, import your two saved memories into notion memory, create a private documents database, and then import your two important surviving memo drafts into the documents. Okay, so what I create? The guide says your agent operating guide will live

6:08inside a private database, not as a loose page. This is actually just an important general benefit of notion. Um it's really effective if you create databases and have those databases relate to each other. So when you have something like a documents database, it just helps agents find things faster because they can query on structured data rather than just searching for

6:28anything that may contain some keyword. I do that all the time and I think it's a pain to create a database, but now I just tell the agent go create a database for me. It makes it easier for me to read it. It makes it easier for me and the agents to find it. It's a 100% a

6:42win. Exactly. And one one of the things we're working on is trying to develop sort of like evals to sort of see to what extent does it benefit and then that we can use to optimize to make it make sure it gives you the most benefit. Um okay, cool. So we're going to create this agent operating guide page. Let's create

6:59the operating guide. And so what is the operating guide? Basically what it will do is serve as the central main instruction point for all of your agents that you may want to use. So once I set this up here, I can take it over to Grok Bot or ChatGPT or whatever, and the way that I'm used to my agent behaving, it will just work. And it gives me a place to modify my agent's behavior. So for instance, what was the thing you were saying earlier about um Sometimes it doesn't go into notion and it just will guess it or sometimes it'll go into its own memories, Grok Bot

7:33might, when what I want it to do is always go into notion. And if it's failing in notion, meaning the data is not there, then tell me and we'll put it in there. Yeah, perfect. So okay. So, basically, we create this custom prompt that's like use Notion as my durable shared memory, etc. So, I hit copy here. It links to my

7:52specific operating guide. Uh-huh. And so, now I have that saved in a central place. And when you paste this, which it shows you where to paste it, settings profile, so I go settings. Oh, you're going right into the settings of Claude, and you're saying before you act, this is what you should know. Yeah. Exactly. So, this is where the key thing happens, right? Um

8:19I've got my agent operating guide here. Once per session, before work that depends on me or my projects or before writing to Notion, read that guide. So, this is the way that it doesn't slow it down for stuff that it clearly doesn't need Notion. Um search Notion before answering questions that depend on personal project context. If search finds nothing, check the canonical memory destination. So, it's like basically trying to get your agent to treat Notion as the context store, and also get your agent to be really good at

8:47using Notion. Okay, what's next? What's this? So, the daily loop is sort of like a maintenance routine. The idea is like it would really slow it down for it to like on every single chat trying like dump memory into Notion. Mhm. And so, if we can have it every day review all the chats you did with your agent and then like update Notion as necessary to keep your Notion context

9:10clean and up-to-date. The daily loop is a really helpful thing. Some other agents, like if you're in Grok Bot, it'll just schedule it. So, here's the setup report. We've got my workspace. We created the documents database. We created the agent operating guide. We set up the custom instructions. We're not We're not doing imports. Daily routine. I'm going to create a new chat really quick and say um let's sort of like test it out. Uh I'm interested in a new car, uh seven-seater SUV, uh with room for two car seats,

9:45uh in the um middle or third row even. Uh what are a good range of options? And ideally what it'll do See, it says checking required setup steps before starting the task. Loading tools. We're looking at Notion. Great, wonderful. So, I didn't say anything about Notion, but it knows Okay, this is the kind of thing where uh maybe this user has said something about this in Notion before, or maybe,

10:14you know. Mhm. This is basically the idea of just like having the agent behave this way by default. The main point of leverage you have is these persistent uh custom instructions. And so, this is it did the three-row SUV search. Uh-huh. Created a document, and then it created a short list database. Yeah. So, now when you ask a question about an SUV that you should get in a new chat,

10:40Yeah. it should know that this is a search that you had done before. Let me be like, can you add photos? There it is. Look at that. Yeah. That's the stuff. Right? Amazing. And some of these don't seem to have loaded. So, bottom line, everything gets saved in Notion as a single source of truth. Anytime there's an important information, it gets added into Notion as a database. Before it pulls information about you that would possibly be a Notion, it knows to search Notion, and then bring it back in as an answer. This is the goal that I'm trying to that I'm trying to achieve. I've just

11:17been doing it by connecting Notion in and trusting it to work. Your insight, or the insight that I've gotten from you today, is make sure to tell it up front in its system settings when there's information that possibly could be in Notion and you've got it all written out, go into Notion first and then get it. When there's anything that I need to put into Notion, make sure on a repeated

11:36basis to go and put it into Notion. That's the answer. Exactly. That's the key thing. This is the prompt that will help you set that up, but basically where you will land if you do that is it will give you some stuff to paste here. And then it Okay. the agent sort of just uses Notion as

11:53your source of truth. Okay. All right. This is great. I love it. And if you like this, I've got another interview for you where you can see how to get all your in all your company data into Notion in an organized way, and there's a link right there. See you in there.