I Built An AI System That Writes Viral Blog Posts INSTA… — Transcript

In today's video, I'm going to show you step by step how I built a full AI-powered blog generation system inside of Naden that can write highly effective, well-researched blogs by a team of AI agents. And in case this is your first time here, my name is Mikuel, and over the past 12 months, I've helped over 40 businesses implement these sorts of AI solutions and taught over 17,000 people in the process. All starting with zero technical knowledge. So, by the end of the video, you should be able to see exactly how all these agents connect, how the system works, plus I'll give you the whole blueprint for free. With that being said, let's dive in. All right.

time here, my name is Mikuel and over the past 12 months, I've helped over 40 businesses implement these sort of AI solutions and taught over 17,000 people in the process. All starting with zero technical knowledge. So, by the end of

So, this right here is the system. I'm going to run it from scratch and show you exactly how it works before we go step by step and show you how I built it. I'm going to press execute workflow.

So, this right here is the system. I'm going to run it from scratch and show you exactly how it works before we uh go step by step and show you how I built it. I'm going to press execute workflow.

I'm going to put the post topic. Let's do AI in finance, and let's do investment bankers.

So, the posttopic for the blog is AM Finance. Then we do investment bankers. Let me just finance. There you go. And tone of voice will be direct.

So, the post topic for the blog is AI in finance. Then we do investment bankers. Let me just finance. There you go. And tone of voice will be direct.

doing. Then what we do is we generate an outline using an AI step which will then be split into four different parts cuz the outline has four different parts of uh of the blog. We'll have a research agent which is this one right here which

So these are the three different inputs that we give it. We press submit. What this will now do is it will send the information to the first AI agent, which is a newsletter expert, which basically does research on the topic that we're doing. Then what we do is we generate an outline using an AI step, which will then be split into four different parts because the outline has four different parts of the blog. We'll have a research agent, which is this one right here, which will do more research on the actual topic that is written on that first paragraph or first section of the blog, and second and third and fourth as well. Then what we do here is, using the outline, which is split up into four different parts, we send it to the research agent, which is this one right here, which is now doing research on each part of the blog using Perplexity, and it's actually writing the snippet for that specific part. Then what we do here is we merge the output from here and here as well. We aggregate them, which means we put them all into one place before sending it to the editor AI agent, which in this case is an AI agent that will edit the actual newsletter. So that's going to be the final version, and we use Claude here because Claude is a better model to write content. Then we use just an AI step to generate the title, and then we send the actual newsletter to our inbox. If I go to my inbox, I can see that I have "Generative AI Transforming Investment Bank Operations," and this is the full blog that we use. And a good thing about this is that it actually references which sources. If you go here to two, it references the source that we're using, right? Also, you can see down here first, second, third, fourth, fifth, and sixth, and this right here is just a prompting thing. Um, and now you have a fully optimized blog which gives us highly in-depth research from the top sources, from top credible sources that we can use. And a good thing about this is you can have this running daily because right now we have a form, but you can replace this with a Google Sheet that will have different topics and then an on-time schedule, which means that it will run every week, every day, and so on. And you'll have a fully optimized blog in your inbox. So with that said, let's see exactly how this works and let's go step by step. All right, so let's go through the first step, which is the form input. So we added a form. Again, this can be any kind of input, right? The form is just the easiest because we can just simply execute step and we can have access to this form right here, which is a very, very simple form that we just create natively within NN. And for the forms itself on NN, we have the test URL and we have the production URL. Now bear in mind the test URL is just the URL that you use when you're testing. Okay, hence the name test URL. When you're going to production, which means that you are turning this on, you're activating this, that's when you use the production URL. So, in theory, what this would look like is you wouldn't have to go inside here and press execute step and run the whole thing. You would just activate it. You would use a production URL, and that's the thing that you paste into your browser when you're doing whatever you're doing, right? As you can see here, this is not activated. So, this is deactivated. Um, so we can't see it.

different parts. We send it to the research agent which is this one right here which is now doing research on each part of the blog using perplexity and it's actually writing the snippet for that specific part. Then what we do here

Then, we have authentication right here. Authentication can be used to basically add a password between the person that wants to fill out the form and actually filling out the form because if you don't want anybody or if you don't want everybody to have access to the form, then you want a password.

will edit the actual newsletter. So that's that's going to be the final version and we use cloud here because claude is a better model to write content. Then we use just an AI step to generate the title and then we send the

Well, in this case, we can just leave it blank because this URL is pretty unique.

actually references which sources. If you go here to two it references the source that we're using right also you can see down here first second third fourth fifth and sixth and this right here is just a prompting thing. Um, and

Then we have the form title. In this case, newsletter form, and then these are the fields. So the fields are what are we asking the user to fill out? In this case, we are asking the post topic, which is a text field, and the placeholder can be AI in finance. Placeholder just means this, right? So if you execute a step, this is a placeholder, which means what is a thing that they see, and typically you add this as inspiration for what they could add, right? Let's make this a required field actually. Target audience, text, required field, and then tone of voice drop down. So we have three different choices. So we have authoritative tone, humorous, and direct. These are the three different choices. So this will be the input. Again, when we press execute workflow, this will execute the actual workflow.

form, but you can replace this with a Google sheet that will have different topics and then a ontime schedule, which means that it will run every week, every day, and so on. And you'll have a fully optimized blog in your inbox. So with

So I can do AI in computer science, target audience 50-year-old men, tone of voice can be authoritative, and then we see here that this is the answer that we get, okay, and these are the variables that we then pull into the next step. Now I'm going to pin this so I don't have to rerun it again every single time. The next step we're doing is we're setting variables. Now honestly, you don't have to do this. I don't know why I did this, to be honest. You can go without this, but I just wanted to set the variable topic, target audience, and tone of voice for no reason, actually.

the easiest because we can just simply execute step and we can have access to this form right here, which is a very very simple form that we just create natively within NN. And for the forms itself on NN, we have the test URL and

So, we can actually delete this. But because this is connected to all these, then it makes sense to have this. I typically have this just to have things organized. And basically what this looks like is we're replicating these variables to just have topic, target audience, and tone of voice. And typically why you would do this is because on the form submission we also get the submitted at in form mode, which is extra variables that I don't want to see. But at the same time, you can go without this because I mean these fields are only two fields, right? There's not a crazy amount. But we usually use set variables when we have an input which is a ton of different variables. I think you use like API or any of these softwares, they give you like 100 variables, and then you basically want to extract the variables that you actually need, and that's where you use a set variable. In this case, it didn't make much sense why I used it, but you can still use it because we don't get charged by step, we get charged by execution. Then we have the first AI agent, which is a newsletter expert. We can use a tools agent. There's three different types of agents: OpenAI functions agent, which uses OpenAI to just do its thing.

that's when you use the production URL. So, in theory, what this would look like is you wouldn't have to go inside here and press execute step and run the whole thing. You would just activate it. You would use a production URL, and that's

Conversations agent, so you can talk to it. We leave this as tools agent because we are using tools with the agent. The source prompt, which is what is the input that we're getting. In this case, if I execute the previous nodes, I can see that the input is this. So we give it the prompt user message because every AI agent has two different types of prompts. Okay, it has a user prompt and it has a system prompt, also assistant, but the main ones are user and system.

Then, we have authentication right here. here and authentication can be used to basically add a password between the person that wants to fill out the form and actually filling out the form because if you don't want anybody or if

And the user prompt in this case would be what is.

Well, in this case we can just leave it blank um cuz this URL is pretty unique.

Then we have the form title in this case newsletter form and then these are the fields. So the fields is what are we asking the user to fill out? In this case we are asking the post topic which

is in text field and the placeholder can be PI and finance. Placeholder just means this, right? So if you execute a step, this is a placeholder, which means what is a thing that they see and typically you add this as inspiration

for what they could add, right? Let's make this a required field actually. Target audience, text required field and then tone of voice drop down. So we have three different choices. So we have authorative uh tone, humorous and direct. These are the three different

choices. So this will be the input. Again, when we press execute workflow, this will execute the actual workflow.

So I can do AI in computer science target audience 50 year old men tone of voice can be authoritative and then we see here that this is the answer that we get okay and this is the variables that

we then pull into the next step. Now I'm going to pin this so I don't have to rerun it again every single time. The next step we're doing is we're setting variables. Now honestly you don't have to do this. I don't know why I did this

to be honest. Um, you can go without this, but I just wanted to set the variable topic, target audience, and tone of voice for no reason actually.

So, we can actually delete this. Uh, but because this is connected to all these, then it makes sense to have this. I typically have this just to have things organized. Um, and basically what this looks like is we're replicating this

these variables to just have topic, target audience, and ton of voice. And typically why you would do this is because on the form submission we also get the submitted at in form mode which is extra variables that I don't want to

see. But at the same time you can go without this because I mean these these fields are only two fields right there's not a crazy amount. Um but we usually use set variables when we have an input which is a ton of different variables. I

think you use like API or any of these softwares, they give you like a 100 variables and then you basically want to extract the variables that you actually need and that's where you use a set variable. In this case, it didn't make

much sense of why I used it, but you can still use it cuz we don't get charged by step, we get charged by execution. Then we have the first AI agent, which is a newsletter expert. We can use a tools

agent. There's three different types of agents. OpenAI functions agent, which uses OpenAI to just do its thing.

Conversations agent, so you can talk to it. We leave this as tools agent because we are using tools with the agent. Uh the source prompt which is what is the input that we're getting. In this case, if I execute the previous nodes, I can

see that the input is this. So we give it the prompt user message because every AI agent has two different types of prompts. Okay, it has a user prompt and it has a system prompt also assistant but the main ones are user and system.

And the user prompt in this case would be what is the variables that we're feeding into the actual AI agents for it to do its thing. And a prompt user message here is what are the variables like newsletter topic, tone of voice and

target audience which we're pulling in from here that it uses to actually do its thing, right? So that's for the user message. And then we have the system message. Now the system message is a prompt that gives it the instructions,

right? The instructions for the thing. It doesn't actually give it the variables because the variables go in the user message. But we give it the overview, the context, the instructions, the tools that it uses, um, examples as well, input, output, input, output. So

these are assistant prompts because we're giving it an example here, example here, example here, example here to make it, I guess, to give it more context as to what we want to do. Uh, cuz sometimes AI it's so creative sometimes that it

just spits out what it wants to spit out. So if you give it a few example, it sort of like guides it. It narrows down the the focus that it needs to have when making the blog for you. Um, and so this

is the full prompt. And again, at the end of the video, I'll show you exactly how you can get the whole system for free. So don't worry. Uh, but you can also take a screenshot of this and drop

it into chatbt and ask it to extract text. Uh, that's what I used to do all the time. And now you have the system message right here. And you have the user message. And we can see we ran this

before. And this is the output that we get. Of course, this is cut, but if I go to JSON, I can see here that this is a full thing. So, table of contents. So, we are generating table of contents for

the newsletter, right? And we're using perplexity as a tool to do so. Why is because you want to do more research on the actual topic to then give it proper table of contents. And table of contents is like, okay, what's going to be in the

actual newsletter, what's going to be in the blog. Once we have this, let me pin this. Then, we go ahead in the next step, which is generate an outline. And here, we're not using an AI agent. We're just using a simple step. Now to connect

your OpenAI to NN, all you have to do is press create a new credential. You can go to platform the OpenAI platform.opai.com.

You can go to login, choose your account, and then you want to go to dashboard.

Get your API key right here. Press this button. Get a key. You can copy and paste back right here. Bear in mind, this is not free as in you still have to pay for this. Um and he pays for this

using something called an API credit. Now the API credit is something that you can see where is it billing the left hand side and you will add money here.

It is very very cheap. So it will only cost a tenth of a cent when you run this. Um so but make sure you have about $5 cuz that is the recommended amount that we use. All right. Now that we have

this we have connected our openi to and then we can use text because that is the thing that we're manipulating. The action that we're taking is messaging a model which means that it's sort of like we're going to chat GBT and we're typing

something to get an answer back. Then we're using the model 40 mini because it is the um we're going for like speed and quality, right? And a for many is great.

And the prompt is this. Your job is to split out the table of contents into an individual item for each section. I'll put each section separately in a field called newsletter sections. When doing so, keep in mind that the newsletter

target audience is this and the tone of the newsletter is this. So now we're pulling in the variables from here. So variables. Here's a table of contents.

So what we're asking it to do is this. We're asking it to basically get the table of contents, which is a block of text, and then output an array, right?

An array. I'll show you exactly what it is. An array is a something, right?

Where it contains different items inside that something. So in this case, it contains the four different sections. Uh each section has a title and has a description. Title and description.

Title and description. Why do we do this? Well, you can simply and by the way, this whole system you can simply replace by just using one prompt from TGBT to make a blog. But because we want to make it higher quality, right, and go

through basically generating an outline and taking each bucket of that outline. What I mean by this is taking each newsletter section and then giving it to AI to do research and then making it more optimized and then making it

better, right? And so what we're doing here is we're asking it to take this text which is unstructured because it's a block of text which doesn't make much sense to us and we can't really do anything with it and then giving it to

AI and saying hey take this block of text that we have here. So this is the input and basically split it out. So I want you to split out an array which is something that has different items in

it. And you know it's an array because typically has a square bracket. Square bracket. Yeah, square bracket here and here. And inside the square bracket, there are a bunch of items which is this and here and then this and here and this

here as well. And same thing with this, right? And so these items are all pretty much the same. And those are the four items, which means that there are four table of contents, four different parts of the blog that we want to iterate. We

want to split it out, right? Each one and then use each one for research, use each one to write the blog, and use each one to make the blog better. So once we have this then we can spit out right. So

I mentioned that we let me pin this. If I press this you can see this is now four items. Why is because the array itself is best for us to use when we want to get the array which can be found here. So if I go here

new set section this now will split up four different items first second third and fourth. And that's exactly what we then feed into our AI agent which is a research agent right here to do more research. So the input of course is a

tools agent because you're using tools. Uh the source prompt is user message which we'll define below. And then the user message itself which is what are the variables that are going into the AI agent for it to use to do its thing are

the section title which is this section description which will be this right here. Where is it? Generate outline.

There we go. description, newsletter, target audience, which should be this right here, the target audience. There we go. And newsletter tone of voice. And so we feed all of this in so that it has context as to what the actual section is, right?

And so this being the user message and this is a prompt that we use for the research agent. And so we have an overview which is you are an AI agent responsible for delivering only the final content for a newsletter. Then we

have context uh which is all necessary details including the section title, the description, the target audience and the tone will be provided. Then the goal and the content must be supported by research and then we give it instructions. And this is usually how

the the actual prompt goes, right? We have overview, context, instructions. Now in this case we have tools. It's only one tool because we're connecting this to perplexity to do the research.

And with that said we have citations, examples. So we give it examples and SOPs as well. And this is a very very good prompt as in it has markdown formatting which is the hashtags which is basically saying hey this is heading

one this is heading two and so on. And then it has a bunch of stuff as well. So rules guidelines all that stuff that he needs and from here we start using claude comparison to openai right here because claude because now we're

actually writing content right and claude is the best content. And to connect your cloud account, press create a new credential. Anthropic console and then you can go to get an API key. You can go to create an API key. Just name

it whatever you want. You get a key that you can then paste back here. Now 3.5 sonnet is one of the best for content.

Let me actually search it up. Um, which cloud model is best for content? I believe it's 3.5 set. I think so. Not cloud. It's cloud. Let's see. Cloud showing you guys the full raw content uh because that's what it is. Claus Opus 4.

Okay. Cloud Opus 4. Do we have that available to us? I don't think we do.

Yeah, we don't have an API version. So, let's do API model cuz not all models are given to us for an API. Um Cloud 4 generation, Opus 4 and Sun 4 are hybrid models. Opus 4 is the most scalable. Yeah, son of four.

Son of 4 um is designed for speed plus quality as well. And since we don't have four here, we're just going to use 3.5.

And that's the thing that it uses as its brain because again an AI agent usually has a brain, which in this case is LLM.

And then it has tools that is connected to which are in this case perplexity, the thing that it actually uses to take action. And if you want to build your first AI agent from scratch, check out this video up here. Once you do here is

we run this. What this will now do is it will go through each section one by one.

So four different items at the same time. Not at the same time, obviously separately. And it does research on each one. And the reason why we're adding two different inputs is because the first input is the actual research that it

does. And the second input is the description. All right. So now we finished the actual research. And bear in mind that each section of the blog will be researched, right? And that's what this does. So, I'm going to pin

this so we don't have to rerun it again because it does take a while. Obviously, it got cut in the video, but I've been here for about 4 minutes waiting. Um, and now with the merge node, what we do

here is let me press execute step. This will now get the output of the research agent, but it will also get the title and the description. So, the merge node is usually combining, right? two inputs because you have one and two and we're

putting them all together. So you see how it went from here, here, and here to all three right here. And these are four different items, which means that it does it for every single section of the block. All right. Now that we have this,

we can then aggregate it because we don't want to send four items individually to the editor AI agent. We want to send all together, right?

Because we have four items. Let's put them all together. And as you can see by the diagram here, this node right here is to take all the different separate items and put them all together into one paragraph. So let me go here execute

step and here we get all the titles and we get all the outputs because again four items for the four outputs in this case for the research agent. And that's where we put individual fields. The input field name will be title because

it's title. Same thing with output which is this one right here. And that's what it takes to merge the list. Again, merge list just means merge everything together like this. So, we're able to then feed it into the editor AI agent,

which will be the thing to actually write the blog, right? To actually put everything together. Um, it will be a tools agent. The prompt will be the list of titles and the list of article content, which we get from here and

here. So, we're giving it, let me show you this. We give it this a full user input. This is going to be a big prompt plus we give it a system message as well. So this is a message will have an

overview. So you are an expert editor specializing in creating and finding content to output a high quality formatted article. You're given a list of titles and outputs and you will use these to create a newsletter tailored towards the defined target audience.

Create a section in the article for each title with a hyperl source in each section based on the content. So when I was making this workflow or this AI agent, I thought of the fact that obviously you want to have inside what's

it called? Hyperlink um yeah hyperlink um citations. So for example, when you go into the actual word and then when you press that word, it goes directly to the source of that news. And that's exactly what this is. This is a fancy

way, a HTML way of saying, hey, this is the URL that we're giving. And this is the name of the actual thing that when we press this, it takes us here. And so that's what it is. And we have the

objective as well to create content. Uh we have the citation management. We have the source section. We have the output format. We have a ton of stuff in this prompt right here. And then we want to output a thousandword maximum or else

the automation breaks right because there's a limit. Uh and today's date is this. Why do we give it to this date is because sometimes within the actual blog it needs to sort of site um it says tomorrow, right? Like how does it know

what today is? AI is not the best at this. So that's why we give it the actual date. So now with this prompt right here, I can press execute step.

This will now give all of this and these uh this prompt right here to be able to give us a final optimized sort of blog that we can use to send ourselves an email. All right, so we have the output

right here. You see here with all these lines and and P's and A's and H2. This is called HTML. HTML is basically the way that we make our text look pretty.

You know, when you see emails, when you get emails with headings, you get with colors and beautiful things, that's all HTML, right? And so we want to do that because of the fact that we want headings, we want a text when you press

on it, hyperl, right? You press on it, it goes to the source, all that stuff to make it look pretty and to make it look more formatted. We use HTML. And then we use another AI right here, which is an A

step. Again, same connection. Message a model for mini, which is fine. And the prompt is create a title for the incoming newsletter. The tone of the newsletter is authoritative. The target audience is 50-year-old men. And here's a newsletter. So, we give it the full

newsletter. And then we say only output the title in plain uh in plain text, no quotation marks, and capitalize the first letter of each word. Example, is artificial intelligence a friend or foe.

So, basically, this is making the title for the newsletter, which we can then feed into the subject line of the email that we're making. I can press execute step. This now gives us the actual content, which is the title of the blog.

quantum computing and AI transforming the future of technology and that's the thing that we would use to then be able let me pin this to send ourselves an email so in the email section just connect your email and go here and sign

in with Google connect your account bring it back and you're all good and then the thing that we are manipulating is message the action that we're taking is we're sending the email that we're sending it to of course is my personal

email the subject line in this case will be the title here right will be this right here content we bring it across here and this is Now the title that we use and then the email type because again Gmail gives us two options whether

you do text whether you use HTML. Well, in this case, because we wrote the email in HTML, then we get this, right? We get a full thing, which is the thing that you saw in my email. And then we want to

turn the append and end attribution off, which just means that usually when you send yourself an email, at the end, it will say NN send this or sent by NN, right? We want to take that off. And as

well, we want the name of the actual person to be daily newsletter, which is why we add here sender name daily newsletter. All right, we can just test this execute step. What this will now do is it will send us another email. I go

here and refresh quantum computing and AI. And this is a newly written email with hyperlink text that I can go here and I can go to the actual source of the research which is amazing.

And it has headings. It has more research and the citations are coming from credible citations, right? Which are great to have. And at the end we have sources as well here. And now this you can use as a way to keep yourself

updated but also to post it on your you can say on WordPress or any of the sites that you make blogs in, right? Because you have a ton of research, a ton of value that you can get because this

workflow right here is a series of AI agents. And the reason why we do a series of AI agents is because when you break things down and you actually go in depth into each topic or each section, the quality is is just much better. And

so as I mentioned before, you can have a form which is the input which is the thing that what do we feed into the actual system, but you can also have a Google sheet that you can run through

every single row and you can then add a you can say on the schedule right here at the start which runs maybe every day right so every day and then it runs and then it takes the topic from a Google sheet and

then it runs through the whole system step by step. Now, within our agency, we've implemented these sort of systems, especially with blogs and content into tons of businesses just because of the fact that businesses know they need to

make content, yet they don't have the time. They use AI anyways. So, might as well automate the whole process and save them a ton of time and make it actually valuable and and with actual research and with good prompting so they can

actually get results. And so these systems right here are perfect for them because it allows them to have such an efficient process to generate this content that can either be sent to our email that can either be sent to a

WordPress. WordPress is just the way that you host your website. You can have blogs on your website, blogs on LinkedIn, blogs on whatever platform you use. Uh that is highly effective, highly researched, well researched, right?

because if it is well researched and so the system is great because it saves the company tons of time into making content and we know it's good because we have good prompting which companies usually don't. All right. And if you want a full

system for free, check out the first link down below which will take you to my free school community. You can go to the custom section. You can go to the templates vault and then you can go to the latest video which will have a

button right here to download the end automation blueprint and you can import it into your own account. And if you have no clue how to do that, no worries at all. You can go to the welcome start here and you'll have a tutorial

importing blueprint to an event as well. And if you apply and you get in, you also get access to the AI automations 101 course which is a very comprehensive course that hundreds of people have already taken which takes a real

beginner in AI automation to someone who's actually able to build automations for themselves or for other businesses.

And as a disclaimer, we're not letting everybody in. As you can see, we have tons of people looking to join our community. So, please put themselves into your answers before you apply. and I'll see you on the inside. All right,

and if you want to watch more videos like this, check out this video on the screen where I show you step by step how I built an email classifier AI agent inside of any land. With that being said, I hope you found value from this

video and I'll see you in the next