hello Hans here um and in this video I'm gonna show you inside the machine room or the engine of how I constructed this logging system automated blocking system so let's dive in and I'm gonna apologize on start with that this is going to be a bit Technical and a bit nerdy but um it's still more visually pleasing to look at this than a python script so um let's get into it this is the main script which generate outlines it generates titles intros conclusions all that for the article it writes sections for the article um it uploads images to Wordpress which we'll use in mid article format and then it um uploads articles to Wordpress and it also gets articles back from WordPress and if we ever want to get an oracle back so how this looks like um is I can show here that this world's most expensive coffee probably lewark we can then go and generate the outline I'll show you how this looks this is webhook then gets activated and brought in the information from Air tables and I have some sleep timers in then it

routes down to this other router, and then this is the outline router because I have different outline generators, depending on which outline I would like to generate. So, this is an outline where we have scraped a URL and we have a lot of information. This one up here is if we don't have a scraped URL, if we just want to make up an outline based on the keyword research without having too much information. I also have a review outliner and how-to outliner. So now we generate the outline, and it's sent that back, and we can see this is what it generated. We can now go into this box here and see what we're actually doing. We are calling panchivity or because that's the best large language model at the moment for writing. We add a system prompt: You're a professional blog writer. Task: Writing a blog post outline and following the instructions. Always write the most important mission first. And that is there's a lot of prompts here, but these first ones are the most important, and I always like to tell it to write.

the important stuff first because I like to write. Yeah, the important things first. So why not have it write that as well? This is a system prompt, and then we'll go down to user prompt. So it will create a blog post outline based on SEO outline review summary, and then we draw in directions from air tables. And then we go down to how it should be formatted, the outline create. This is how many paragraphs or sections, and then including between one to three subtopics in each section. Write the sections on the outlines. Then we go down and show it how an outline will look. So it will look like this in H2. Enter the keyword cluster here. If we have any external links, put them there. And then H3s, then divide them by these and so forth. And at the bottom, we enter all the information. This is the main keyword, and then we have the keyword cluster. Um, and all that SEO data, the SEO outline, and the review summary based on all that information. We get this back. So search intent. Ah, sorry, that wasn't it. We're down in this one.

here this one now we get it back H2. How is the world's most expensive coffee made? Good question. Um, might be too many h3s, but again this is what it does. We then review the outline. Okay, sounds fair. Um, once we are happy with the outline, we then send it to generate full blog post. Canine intercept that. Yeah, I could. So now it generating full blog post. It then goes in, writes an SEO title. You can see from the message, its content. No, yeah. Ah, that was the input on the output. There is world's most expensive because I kept this. And you can see some of the prompts in here that I won't. I want the AI to write really short, uh, titles for my articles because in my opinion, they convert better. And you might have another opinion, and that's up to you. I just like short and concise articles or titles that has the search keyword in them and has a good search intent or answer intent. So we actually send this information to Airtable while the article is being produced so we can go here and see.

is I would then add the world's most expensive coffee worth it. It's because if it added the it would come and go over the limit of what I set it to because AI has a tendency to just write super long fluffy stuff if you don't constrain it. Um, but now it's writing each section, and that's why we had those um, those little arrows in there in the outline because it breaks up each section like this. Then it breaks up in another section and then feeds each of these section outlines into the block Rider. We then save each of the outputs and then aggregate it into one long article at the end. So that was the four paragraphs sections. We then aggregate that, put that into a too long did not read or summarizer. Um, I just like this format. Some people might not, but this is how I like it. Then send that back to air tables. That was the whole article. Now we can see that the summary is put here. Then we have the intro. We then have the block content and so forth. The conclusion may be a bit

long and then a short and tight um meta description with world's most expensive coffee in it. So let's go back to make whoops get back there and all these nodes are constructed with a system prompt. You are an expert headline writer, be self-critical, follow instructions. You never write headlines longer than 50 characters, so that's the constraint. Why I didn't add the to the title before and then we feed it um this main keyword and we also feeded the block outline. And I have some headline examples here um and some other stuff. Don't use ultimate best secret amazing like don't do that and we have the block intro. You're an expert writer, your task is to write catchy block intros to get people to read the full article again we constrain it to dunk wire so it doesn't write too much then we feed it more information on how it should be writing this is taken directly from Google on right to adhere to eat then we add some Avatar information this is information about our Persona how like they are coffee.

enthusiasts who like to x y and z. Then we give them the prompt, um, pulling in the information from Air tables and feeding the outline. And that's more or less what it does all the way through. Image prompt is a bit different, slog name also a bit different, but still. Then we get up to router and then an iterator. The iterator splits the outline at these four here and then feeds it in to the block. Block Rider, and which again has a system prompt, your fun and engaging article Rider and expert within expert description tasked with writing one article section out of X. So we are telling it you're writing one section out of a multiple of sections. Again, you're self-critical about writing, never at fluff, always write the most important information first. So that's just how I like the AI to write. You can add your own spin to it and have it do whatever you want in the system, systematized manner. So we have a role assistant here taken directly from Stanford question answering data sheet. This is how you

answer questions with the highest NLP score. So we're feeding that information into the article to the AI, which means that again we prompt it here with a lot of stuff on how to write. For example, use the Stanford Question Answering Dataset method to answer questions with as high a persistence as possible. And right after the question is asked, this will say to the AI, Please answer the questions by mirroring back the question if possible and then answering it. So like how Google likes questions answered, we're telling the AI, Please answer questions this way. Yes, this is what works for Google, so we might as well put it in. And then after the initial answer is given, you can then elaborate on the topic. Um, then I ask it to write in markdown. Start the section by writing these two and so forth. Then we feed it in the outline and add all the information here. We're at the outline plus the review summary, and it loops through all those, generating the article, then summarizes it and sends it back to make. So that was more or less how it

generates articles. Um, I can then say, for when we are uploading to Wordpress, we do a post. WordPress here. Then we do markdown to HTML because WordPress wants it in HTML and not markdown when you are feeding it through an API. We then get, get the hero image, and we add a um, what's called an ALT text or this image. So all images we upload through air tables have automatically assigned an ALT text within convert. If it is PNG, we convert it to jpeg, we resize it, then we upload the media to Wordpress. We can then see here, we have the title of the image, which is the title, and then we have an ALT text. We can add more if needed here, and then we add the blog post itself with all the information needed. And that's more or less the block generating script. So I hope you enjoyed this video a bit more technical. Um, but let me know what you think in the comments, and I'll see what I can do. Have a great evening or day wherever you are. Bye.

How I use Make.com To Automate AI For Blog Creation - Nerd Version

This channel is about automating business using No-Code and low-code tools combined with AI.
Nothing cute, just stuff that actually works.
You can get all our free stuff here: https://scalebytech.com/free-stuff
