[00:00] Hey, my name is Sethi and in today's video I want to show you how I have built an agent that can write a newsletter or a blog article in your own words or the words of the creators or copywriters you are most inspired by.
[00:16] And then it also knows how to structure the actual structure of the blog with the right white space, bullet points, the right titles, and everything from A to Z. The way I build it is I took the experience of managing the marketing agency with over
[00:33] 82 employees, and I literally reverse engineered what might be needed to create a successful article. and from the examples they need, the SOPs they need, the instructions they need, the examples they need
[00:51] and have built agents that can replicate the work that the team used to do. So this is what you're going to see and let's begin. So what you see here is the complete automation.
[01:04] The way I will structure the video is first I'll start with the demo, Then I'll show you the complete process and which agents are included in these different steps. And then we'll go step by step on how I build this, including all of the different elements that you can now see on the screen.
[01:24] So to start, I'll just test the workflow. As you can see, this one is the new one, February 12. so in essence this this is the complete article it wrote so that that's first of all that's like
[01:42] amazing right like when you see it we have the h2 title a complete step by step with numbers billing in your sleep so as you can see it ended like bullet points numbers bringing it all together
[01:56] and that's the complete blog post article and in essence it's with a really good copy a lot of people think they're drawing in task and too much
[02:08] doing in fact they're drawing in lack of clarity it's not as blah blah so how to fix this let's talk about the breakdown and it breaks down everything building in your sleep let's talk about what's selling while you sleep or the news and
[02:20] then break it down different steps of transformation the best businesses keep blah blah blah. Free value, so it literally broke this down into different funnels that you can use online to return lost, secure, awakened, doubt, training, belief, belief the one, and then people
[02:37] with a mission. And the step-by-step value ladder, so vision building, idea distribution, low risk magnet, so amazing, a complete process based on the context I send them. so what will happen is that first it will go to notion so in notion i have a complete second
[02:57] brain the second brain is a system i use to put everything i have in the digital world in one place from managing my business different businesses all the things that are private like
[03:11] my real estate investment business inside personal inside the different systems i have for mindset the vision of the future I have, everything is in this one space in Notion. So I just wanted to pop here real quick to say that this second brain,
[03:27] you will be able to find it in my free school community. It's in the description. You will find there a link to go into the school community where you will have not just this second brain that's not all the second brain,
[03:40] but some of the second brain that you can actually use in this automation. you will also find the complete automation that you can just import into your NNP and start using it
[03:52] and use the complete blog post creation tool with all the agents I've created. So if you want, it's in the description and enjoy. So the way this automation works is that it first starts by going to a specific place in the value creation system,
[04:09] which is my content creation system, in the content flow. in the content flow whenever I create a new page and in this new page as you can see I will tag the platform as a newsletter so instead of having a YouTube video or an Instagram post it will
[04:25] actually be a newsletter and I will put some input so that could be any idea you have in mind right so let's say you have watched an amazing video or you saw a great reel and some idea pops into
[04:38] your mind, I'll just go and I will write this idea. It can be on my phone, it can be in my idea capturing the machine here on Notion. And then I'll just go and write some stuff about this idea.
[04:50] What it does the moment I create this new page with my input and my newsletter, it actually goes to the automation and it grabs this page. Then it checks if the new page I've just created is
[05:05] having the tag platform newsletter, if it does, and if it also has input, then it will continue to the next action. Now, the next action is very important because one of the things you find out
[05:18] when you run a team of marketing, and in essence, I think that it's related to all marketing aspects, right, is that every single piece of content you ever create should aim toward a specific type of
[05:30] audience. So if we write a newsletter without providing input of who is the audience we are targeting with this newsletter, the editors, the copywriter, even the newsletter title agent
[05:43] won't have enough input about who is he creating the title, the article for. So this is why it actually goes into, again, into my Notion database. And here, it will actually go now to the brand
[05:57] blueprint into the buyer avatar. So I've created a table with all the information about my buyer avatar from here, like the segment itself, the values, the main archetypes, desires, frustrations,
[06:11] emotions, sub-archetypes, main fears, what triggers them emotionally to take action, what they expect after they take action, and just a description of this targeted audience. And what it does is that
[06:23] It actually goes here and it goes to this specific Notion page and it grabs, as you can see, all of this information. Then once we have it, we are going to trigger the first agent.
[06:35] And that's the title creation agent. Yes, we have an agent that all it does, it creates the title for the article. And this is very important because I don't know if you ever heard this sentence, but 80% of the work done on, for example, YouTube or if it's articles or anything else is finding the right title.
[06:58] That's what will hook people to actually click and go into the article itself. So here I've created an agent that all he does is taking this input. So it takes the newsletter context, what you write before, and it takes specific information about the targeted audience.
[07:15] As you can see, this is all the input. It will take the name, so that will be the online business owner. It will take the audience description, so audience description, their ambitious coaches, consultants, online service providers, agency owners, blah, blah, blah.
[07:29] And then it takes the emotional trigger and the emotional action-based outcome, which means what will trigger them emotionally and what they expect from an emotional perspective to feel once they took the action, once they clicked.
[07:45] That's the information that the title agent really needs. And then it creates the title, and the title goes, as you can see, the system that sells in your sleep, create a movement, that's the title.
[07:58] And what it does is it goes to the newsletter planning agent, and this one just breaks down everything we've provided so far. So like the title, the context, and the entire information about the target audience from the fears, frustrations, desires, and it creates four different titles.
[08:15] Now, the way it does it is based on a specific formula. So as you can see, it generates four high-level engaging H2 headlines based on a specific formula called APAG.
[08:27] That's a formula that I learned from Dan Coe, and it's Attention, Perspective, Advantage, and Gamify. So in essence, what it does is that the first title is there to grab their attention.
[08:39] Then the second one is to provide the perspective, so your perspective on the idea, on the concept, and then the advantage. Why your perspective is right? What's the benefit that audience, your buyers can actually see when they use this perspective?
[08:56] And then it gamifies. So it creates a list of step-by-step process that people can actually take to create the reality that you present them. So once it does this, and we have those, as you saw, we have those titles, the introduction, the unseen peach, why your system shines, steps to movement mastery, it goes into the project planner.
[09:18] And what the project planner literally do is that it just take those four titles and break them down into a newsletter section. So we give them the information, the table of contents, that's what we had before, and we just break those down into four different user sections.
[09:33] And then we split those from one item, so from one table of contents, one section with those four, we break those down into four different items. And the idea behind it is that now we are using a research team.
[09:49] So instead of sending the information to all the information to one person, we are actually sending it into a research team that takes one by one So it takes every single section as you can see we have four items in the output and it takes every single action
[10:10] and then it goes and do all the research needed to bring the right information about the topic we want to talk about in the newsletter. So what we give it as an input is the title
[10:24] of the complete article, then the context. So the input I provided in the beginning, so we know where to start, and then the section title. So every time you will get those four different items, right?
[10:38] So every time we'll get one title with the title, with the context, section one title, section two title, section three title, and then we'll go and do the complete research. So this agent is responsible for delivering the final well-researched content for a newsletter.
[10:56] Now, it's not really the final piece of content, but when we tell it that it's the final, you will actually provide a more holistic output. Now, in essence, what we want to give it is the newsletter title, the context, and the targeted audience, of course,
[11:11] and the section title. So the way the research works is that it takes the title, the context, and the title, and then it goes and grabs the right information related to all of those. The idea here is that we give them examples of what we want to get.
[11:28] So we give them the instructions and we give them a couple of examples. So what's the title, what's the context, what's the targeted audience, and the section title.
[11:40] And then we want to give them a complete output. Now, if you don't know how to write instructions, that's probably going to be one of the next videos that I will create. But for now, what I did here is that literally provided them examples of good copywriting and told them that's what I want.
[11:57] Now, it's a lot of text, I agree. But in essence, the thinking process behind it is that I really wanted to understand how to write content really well, even though it's not yet the copywriter.
[12:10] It's like we want to level up the level of copy with every single agent that we use throughout this process to get a very good copy in the end. So I gave him an SOP, standard operating procedure, to what are the different steps he needs and the final notes and gave him this information.
[12:30] So what he does is he goes and uses a pre-tool that I've created. What you can see here is a sub-process, and the sub-process, what it does is that it takes copywriting examples, if it's from you or people you're inspired by, and it saves this into a database in Pinecone.
[12:54] So the way this one is constructed is I put it, the trigger itself is when executed by another workflow, just because that if in the future I would like to integrate this into a more robust agent, I can do it.
[13:09] But for now, I just use this workflow to get the information into the Pintone storage. Then I use an HTTP request to epify, which we will go through in a moment.
[13:21] So you know exactly how it works. Then it converts this information that I got from the scraping, API what it does, it scrapes a specific website, takes the information and then creates a text file from the information it gathered.
[13:37] And then it goes and saves it into the Pinecone Vector Store. So first, let's understand the Vector Store. So the way you do it is you create a new account. so you go to pinecone you log in or sign up to a new account then you'll generate a new api key
[13:54] in the api key can be let's say an article and you will create keys then you can just copy the key right and you'll go in here create a new credential paste your key and then save the moment
[14:09] you do it you will have this connected account and then what you want to do once you have done it you'll go to database and in database you want to create a new index as you can see i already created
[14:23] the copywriting examples and that's the samples i have used what you'll do is you will create a new index you can give it any name this is something internal so i just gave it like copywriting
[14:35] for a newsletter right and then what you want to do in the configuration you want to make sure that you check this one so you choose text embedding three small it's
[14:48] very important because here in this automation we will use the same embedding so we want to be consistent throughout the process so you choose this you keep it serverless AWS whatever and create the index so now that you
[15:04] have created the index what you want to do is you want to go back and here you can go into again find convector you will do insert document because what we want
[15:17] to do is insert a document and we want to choose copywriting from user the one that we have just created now let me explain for a moment about epify so what
[15:29] epify does is a website that holds a lot of crawlers, scrapers, extractors so it helps you gain information from a lot of different sources online so like Instagram, website, Facebook, you can even go to YouTube which is it's mind-blowing
[15:46] and I'm going to use it in a lot of different tools in the future that I will probably upload to the channel so if you want to see this you can just subscribe to the channel and for now what I've used is the website content
[15:58] crawler so what you want to do is you want to come here and the first one you want to do is you want to create an integration so if I'll go back into this one so what I did is created an HTTP request here you want to put the URL
[16:16] right and the way you do it is you go to integrations in integrations you will use API endpoints because that's what you want like the API endpoint and we'll go to run actor so actor is like agent on epify and we go to run after
[16:34] synchronously and get data set items because this is what we're trying to get we're trying to get items to put into a data set right so i'll just press here to copy this it's already with the token so the api key is already integrated i'll come here and i'll just
[16:51] paste it here after i did this i don't need any authentication and i'll just make sure that i'm sending a body and the body is json and again i will specify a specific json and what i've
[17:05] pasted here is simply going again into epify in the input so here you can put the actual start url so the way this website content crawler works it's very smart because what it does it actually
[17:23] goes to the first page so that will be like the high the bird's eye view page the first the initiator. So from that, it will actually start finding different links in the page,
[17:35] and it will go and grab information from those pages. So if, for example, you'll go to this page, right, so like my letters, what it will actually do is it will fall into this page, and it will start grabbing all of those articles, and it will get the information from those articles.
[17:53] In essence, you just put the URL, it can be your own content that you have written in the past, if you want to train it with your copywriting examples, or you can go to the top creators or someone you want, you're inspired by.
[18:09] Of course, from a copywriting perspective, it's not the best thing you should do, but it's still possible. Then once you put the information here, I'll go to run options and because I'm using the free account, I only have 8GB.
[18:22] So what I'll do is in the memory, it will start from 8. So I'll just drop it down to 128 because time-wise it's fine, I'm not in a rush. And I'll press here instead of manual, JSON.
[18:36] So now it will give me the complete JSON file. Before you copy and paste it, what I will really recommend is to change the max results. You don't need to scroll. So if you send it to someone who created 100 articles,
[18:51] you will literally go and get every single link on this huge index. So you don't want to do it. You want to take the right amount of results. So five, from my perspective, is enough. You can do seven, eight, nine, ten.
[19:04] But I just put five. I grabbed all of this, and I went back here, and I just pasted this information here, right? So once I have this information here, what you have to do when you are doing the pinecone vector, so you will be asked, so if I'll add another, just for the sake of the example, so if I'll add a pinecone vector, I will choose add documents to the vector store as we did, and I will put the space for this one, right?
[19:34] and as you can see we don't need to add anything else and then I'll need to choose two different things first of all the embedding and the embedding is just like what we choose before so embedding open AI and I'll choose the text embedding three small just like we did in the
[19:49] pine cone and then in the documents I will do the default backup loader and here as you can see what I did is it not a JSON file right Because we are not uploading a JSON file we are uploading a text file So I choose binary and then load all into data And the only thing I added in the metadata was file name
[20:15] and copywriting samples. I did this because then when I prompt the system for the different agents, I can use this value as the value that triggered them to go and search for those samples. So once
[20:30] have this completed process ready, in the end added a text splitter, so what this does is that it actually split the text that it sends to pinecone, so instead of sending like a huge bulk
[20:45] of information that it's very hard to go through and then grab all the information you want, you can just put a chunk size, so like 1000 characters and the chunk overlap is how many characters you
[20:58] want to have overlapping between different records in the index. So if I'll go here just to show you in my copyright example when I added everything I had those records that went into the index
[21:13] and in those what happened is that some of this information is duplicated here and duplicated here and duplicated here right so in essence what it does is that if you put let's say 20-50
[21:26] characters, it will use the same 250 characters in the next record and so on and so forth. Once I had this ready, I just tested the workflow so that should work for you and grab all the
[21:40] information from the website you have put and integrate this into your pinecone storage. And then if I'll go back here, what I did is that I gave their research team a way to communicate
[21:56] and not just grab the information and it's important because on the other ones I actually gave it the way to grab the information. On the research team what I wanted to do is not giving them a way
[22:08] to grab the information is to actually use another OpenAI chat model to actually communicate between them so the research team will communicate with this chat model to grab answers based on the data
[22:25] I stored. So the way I did this is I integrated in this tool and answer question with vector store, right? And when you do it, I just gave it a data name, so copywriting, and I gave it a description,
[22:40] copywriting sample. So as you can see, for example, in this process, what the agent did is what is the copywriting style of the system that cells in your sleep create movement. So as
[22:53] can see we have here a bad response but in the other ones how to write the unseen page it actually gives him information from the data set on how to write right so when you ask him what is this he
[23:09] didn't understood it but when he asked it how to write it actually what he did is use this chess model to go and grab information from the pages so what I did is I added this
[23:23] answer questions in it I added a limit of 10 so only 10 different pages can go from the database back and then here in the pinecone vector store I did retrieve
[23:38] documents so instead of the insert documents now we want to retrieve those documents and we tell them which from which list so for example it's copywriting samples and then it goes and retrieves this and again use the same embedding as
[23:52] we used before embedding to a small so now what is what happened here is that the research team go ask a question this AI goes grab the information and answer this question now for me I don't need this AI this agent to actually know how
[24:09] to write really good because he just need to write an okay article but focus on the length and it's important because if every single team member from this agent team knows what is the
[24:26] most important thing that he needs to do for example he needs to really focus on writing in the specific length we need so when it goes to this copywriting we don't need to really emphasize
[24:38] on the lens, but more on the copy, versus the editor who we don't emphasize on the copy, we emphasize on editing the structure. So once the research team is finished,
[24:51] there is an aggregator. So it literally takes all of those four different items, right? So like section, content, section, content, section, content, and it aggregates this into one specific
[25:05] output and I did this because when we send a request we have a specific amount of input and output tokens that we can send to OpenAI so if you do all of
[25:21] them in one go maybe you won't reach the length that you want to reach with each section of the article but if we do four different requests what we will get is a
[25:33] longer article and then when we aggregate this we literally played the system so now we have a complete blog post with 1,800 to 3,000 words from those
[25:46] four items and that goes into the next part of the process and that's the copywriter the copywriter focus is to write really good copy so in essence
[26:00] what he does is we give him the information the title the context and then we give him the raw content so that's the section title and the section content so he gets all the information
[26:13] about the complete blog post and now we need to go and write the content for the blog and his entire emphasize is to be master in copywriting. So the way I did this is pretty different from
[26:29] other ways I saw, where it's just giving an input and output. So what I did here is I played on a couple of different systems on copywriting. The first one is I created a persona. Now this persona
[26:47] have a specific information so there is description of the persona the tone the voice the language the sentence structure the thought patterns how this persona thinks and then i told him in the
[27:01] role itself that he needs to channel the persona right so like the name of the persona here and he just need to channel this persona and from this from channeling this persona he will write
[27:16] the content right so the prompt itself emphasizing a lot of times to use embody 100 of this persona right i still give them some of the apg frameworks so we keep on working with the same framework we
[27:31] used so far and once i give them the persona one of the main key ideas here is that now we like embed this persona this agent embedding this persona and then I instruct him to go to the
[27:47] copywriting memory tool but now as you see I didn't give them just the answer questions like before because I don't want him to ask questions and get information the problem with asking questions
[28:03] is that he can ask him how to create one, two, three, four, how to write. And then the other AI will actually tell him, you need to write with this metaphor and this. I wanted to give him this specific persona.
[28:16] And now that he has this persona, I wanted to give him the actual examples of how this persona writes. So I send them straight forward into this copywriting, write samples of copywriting into this list of copywriting examples.
[28:30] and I'm telling him that he can go review the copywriting memory tool text provided as a reference, analyze how the persona writes their choice of examples, phrasing and tone,
[28:45] let this memory tool be your blueprint for mastering the persona writing style. So I give him the persona and then I'm telling him go there and grab more data on how it writes when you write, right?
[28:58] So there is all the instructions, the HP AG framework, another SOP, and then some outputs, and then some final notes. Now, as you can see, there was in one of the runs probably a timeout, right?
[29:15] And that's because here I'm using 01. Now, in the other ones, I use GPT-4. so in most of them it's GPT-4 as the main model but only the copywriter because and again you can
[29:30] use GPT-4 but I really wanted to have an amazing copy for the blog because that's probably one of the most important things in this entire process is that we have just one agent that is better than
[29:43] everybody else and this agent we want to leverage the most important task in the complete automation flow, which is the copywriting. Because, okay, of course, creating a title is much easier than
[29:58] writing a complete article or a newsletter copy The highest leverage the one that from a business perspective will give us the most conversions on the tool that we creating right if we are creating a newsletter and we want to convert people to the specific action or if
[30:14] it's an article to our website and we want to convert people from a specific banner or something then that that will be like the copywriting will give us the highest leverage because it will raise the highest value from it from the audience perspective the value will be higher if we have
[30:32] really good copy. So this is why the copywriting is using 01. So I'll retest this step to get the information. While it gets the information, I just want to talk about another piece of the process
[30:46] that you can use. As you can see, I didn't integrate it right here and that's a polisher. So the idea here is that you can have another engine between the copywriter and the editor.
[30:59] and his specific goal will be to take the information and just polish it. So the way I did it, it's like a system iteration system where what I wanted to have is an agent that can actually go and see
[31:15] and it doesn't need, if you can see here, if you see here, your job is to review the provided newsletter content and adjust any generic AI-like or off-tone phrasing so that the final output sounds exactly as it were written by a down-to-earth,
[31:32] care-digit, frame-delivering, actionable, blah, blah, blah. So in essence, what I wanted to do is to have an agent that can go back and watch every single word and phrase in the system
[31:44] to make sure that what we created is actually channeling the persona we created for the copywriter. So now that it's finished, it will go to the editor, right? and let's continue this step.
[31:58] And what it does is that now we have the complete blog post written by the copywriter. So listen how different it is from a copy perspective versus a generic blog, right?
[32:15] So close your eyes for a second and picture Neo from the metrics. In the beginning, he's pretty much entrenched. No real sense of what's real. Just a vague itch that there is more to life than a gray reality around him. Fast forward a bit and here's the one.
[32:28] Fui awakened to his abilities and driven by a purpose bigger than himself. That, right there, captures the core of how transformation works. People don't start as heroes. They grow into the role step by step, stage by stage.
[32:42] So, from my perspective, that's a great copy, right? It moves the audience. It creates a story to start the blood force. So, that's a great result. Now, as you can see, every time you see here the slash in, that's a new paragraph that we have in the article.
[32:58] So in essence, because the copywriter wasn't instructed to create a really good structure, what he did was he just created a very basic structure for the blog itself.
[33:10] So you can see that it's like there are not a lot of paragraphs, not a lot of new lines that we are sending here. The editor itself gets this information. So the title and the newsletter.
[33:22] So if you want to see the complete process, that will be how it looks after the copywriter. That's like the article itself after the copywriter. And what the editor does is his role is to actually start breaking down the structure itself of the newsletter or the article.
[33:44] So he wants to add proper line breaks, double line breaks. bullet points where it's important. And there are a lot of different information here about how to do it. And I use the same system, I use the same concept of creating a persona, but this one
[34:01] is the precision formatting persona. So it's not the same persona as the copywriter, it's a different persona. But this persona entire idea is to actually create a very good structure
[34:15] for the post. I give it some instruction and an SOP and again he can go into the pinecone vector but his instructions versus the copywriting instructions is not to go
[34:31] to the vector like the database and learn about the copywriting but learn about the structure we use and then he will create a complete structure. Now because a notion and that's the
[34:44] end of the process because Notion have only 2000 characters that we can upload into the page then what I did is I integrated an AI transform which in simple words it's an amazing tool that
[34:59] you can just instruct with like plain text and it will create a javascript code that will help you do the whatever you want so for here I just wanted to him to split each section of text into three
[35:11] different sections and I also told them to make sure it doesn't cut a word or midpoint or sentence so in the beginning it just gave me like this jumping between things and if you want to completely automate even the upload of
[35:25] from notion to your blog tool so it can be WordPress or framer or webflow whatever you're using in your website or if you want to send it as a newsletter if it's convertkate or MailChimp and other platforms then you don't want to
[35:39] suddenly have jumping between the roles and sentences so this is why I added here what it does is literally taking all the text and putting and cutting it into different sections and then I again created a notion one I added a database
[35:58] page so now it's create a page in the beginning we wanted to get the page now we create a page and this one is in my second brain if we go back here this will go into the newsletter system and it will go here into the newsletter
[36:14] management okay so last thing we have to do is just upload everything to the notion as you can see this one is the new one February 12 so in essence this
[36:29] this is the complete article it wrote now I will add that if we have used all one so that that's first of all that's like amazing right like when you see it
[36:43] we have the H2 title a complete step-by-step with numbers billing in your sleep so as you can see we added like bullet points numbers bringing it
[36:56] all together and that's the complete blog post article and in essence it's with a really good copy a lot of people think they're drawing in task and too
[37:08] much to do and in fact they're drawing in lack of clarity it's not as blah blah blah so how to fix this let's talk about the breakdown and it breaks down everything building in your sleep let's talk about what selling while you sleep really means and then break it down different steps of transformation the
[37:23] best businesses keep blah blah blah free value so it literally broke this down into different funnels that you can use online to return loss to curious to awaken doubt training belief belief the one and then people with a mission and the step-by-step value ladder so vision building
[37:41] idea distribution low risk magnet so amazing the complete process based on the context i send them now i have to say that from a structural perspective but that's not perfect it's good but it's not really perfect. And if we use here, based on my tests, if I would use 01,
[37:58] I would get much better results, like really line breaks one line at a time and much more. So it's amazing to see how we can use different AI models to get different results. And the last things I
[38:12] want to touch here is that you will see that in the complete process, there are two times where I used the parser. So in the research team that was the first one. In essence the way I used this was
[38:25] I wanted to take the information and again tell him that it should write 400 to 750 words and it will give me two different strings. So like the actual section title and the description and here
[38:38] in the after the editor I just wanted to bring everything back home. So in essence what it needed to give me is the newsletter title that's like the first thing and then for each section that
[38:51] help me take the complete information and then if we see the scale we will just see that we have the newsletter title and then the different sections so each section includes the section title and the
[39:03] text by doing this instead of getting one big chunk of text we actually break this down and And then we can organize this in no shame. So yeah, that's the complete automation.
[39:18] I hope you have enjoyed. Again, if you want this automation, everything with the instructions, with the system messages and everything, check in the description. There is a link to my free school community
[39:30] where I help you both learn about the marketing foundation. So everything you need to start building your business from landing pages, lead magnet, content creation, and all the systems you saw in my second base,
[39:42] as well as everything related to automation, specifically to scale your business. So you will be able to go there, download this entire automation, and just start using this from where it is.
[39:56] You will also have access to parts of my second brain, and not everything, but you will be able to use it. And in the community, you will see how you can use it. All the information is there. So make sure you go and check it.
[40:08] and until the next time, I hope that was great and I see you. Ciao!