in this video I'm going to show you an advanced AI blogin automation that uses perplexity new deep research feature as well as Claude 3.7 Sonet which is a new Advanced hybrid reasoning AI model we've got two different versions of our bloging system on the go one uses make.com the other uses the n8n automation platform so you can choose whichever one suits you the make.com version is a bit easier to understand whereas the N ATM version has a few extra advantages you can bulk generate articles using an air table base for both of these systems however the N ATN version also comes with the additional capability that you could use an AI agent such as one through telegram to trigger the generation of the article directly check out the link in the description if you want to see a full run through of that I'm going to run through this in our me.com scenario I've gone through both these systems in separate videos but I've made some key updates in order to integrate perplexity deep research feature and to use CLA 3.7 solet if you want to get the blueprints to these automations then check out the link in the description to Community a key benefit of this automation is that we're using CLA 3.7 Sonet which results in just a simpler automation previously if you wanted to create a blog post of that length you need to create an iterator to generate the article section by section like we did in our more advanced blogin blueprint a few months back but now this new hybrid model really simplifies that process and makes it a lot more accessible to create these long articles in one shot so let's start with an example I have an air table base here I have this pulse table set up and each row here corresponds to an individual article I've added in the first row here and I've added in an article title of what are the best AI models in 2025 it's going to need to use a deep research feature to fetch timely information for this now in status I've moved this to ready for article generation I could ask it for an outline but in this case I'm going to streamline the process and move straight to generating the article and pushing that to Wordpress I've just run this Automation and now we can see the data that has flow through so it's taken the air table record we're set a bunch of variables at the start here this is just an easy place for us to set the model name and the website root URL and some of our API keys and later on if you want to use a different model for example you can just update the model name here which is great from here because we decided to move straight to generate the article it's going to do all the research within this section where we branched off from the main flow but you can do that in a separate step whereby you create the article outline like here and then review that in air table and then choose to generate the article based on that so we're using this to get the XML sitemap for the website most websites have an XML sitemap such as your website address with pulse sitemap.xml at the very end the address can vary a little bit but your website probably has an XML sitemap that you'll be able to scrape for this automation so earlier on we added the website root URL within these variables this is a simple httv request where we're getting that XML sitemap and it's return the response we're using join and map functions just to clean up that data a bit and just get the URLs from the sitemap then we're passing that into into Claude and we're just simply asking Claude to pick up to three URLs from the url's list we're passing in the title from a table and then the list of URLs and then it's going to choose relevant links that we will then use in the article the response from that was these three URLs the rides of artificial intelligence what is AI the future of AI so these are relevant articles that I will try to link internally because we decided to jump straight to generate in the article then it branched off to this separate section and it used perplexity deep research feature and I've just asked her to create a summary of facts figures and information to include for a blog post do not respond with introductory text do not number the headings and then after that the topic is the following title and we're asking it for a specific structure of key points data information keywords any elaboration on each key point and a bunch of other items in the prompt here the response from that you see choices message content it came back with this really long chain of thought process at the start of its response which is pretty cool but we don't want that because it's just going to be extra noise later on in the automation so we're going to remove that in the next step it's come back with a ton of insights and statistics and information across a lot of different AI models from proprietary to open source we see specialized AI systems Trends performance benchmarks and recommendations Market statistics there's a lot of good stuff in here next up I passed it into this text parser and this horrible regular expression is just filtering out that pink part from the start of the response and the result of that as you see here is just the perplexity response but without that starting section that we don't want then we're passing this in to Claude and we're referencing the model name that was a variable at start of the scenario which is 3.7 Sonet now you can pause the video if you want but this is going to create an article outline for the topic based on the following topic and Analysis and so we're passing in the article title and we're adding in the analysis as well as instructions so if you provided Specific Instructions on air table then it will also include those as well and we've specified a very specific structure that we want of key points keywords facts figures retain the data and then we've added in the analysis which is the result from perplexity and the result of that is a nicely cleaned up article outline that we can use to create the article and it has a bunch of different sections which is great now we're using the set variable module as a way to get the data back into the main flow you cannot merge flows in the same way that you can do in N ATN for example so we're using this set variable option where we're setting the article outline and citations and deep research and then we're setting them there and then when we go back to the flow so in a router in make.com you right click and go to order routes you can see the order that those routes are executed in this is the first route that's going to be run and if that has been set there then by the time we use this get multiple variables here it's going to be able to get get the value from that into the main flow that's a pretty cool trick if you've never used it before and it just means that we do not need to duplicate this entire flow here just because an outline was not created to start with so it just can help us simplify our scenarios quite a bit so now that we have the article outline and citations and research directly in the main flow we now need to generate the media for the article before writing the article now we have these modules here which have called flux 1.1 Ultra which is one of the best image models available currently and once it generates those it uploads them straight to Wordpress and then we use this data for SEO API which is a really powerful API that we can get lots of great insights for SEO but you can also use it to scrape Google search and what we're doing here is we're just passing in the article title along with youtube.com at the end and this is effectively just doing a programmatic Google Search and the result from data for SEO was a pretty comprehensive result set for each of the individual search results on Google but we do not want to pass all of that into Cloud so what we're doing here is we go down and we're using the join and map functions which are some more advanced functions of make.com and we're just extracting just the URLs from each of those search results so when that was actually run last time the prompt that we gave it was you were an assistant that will return the first YouTube video URL you find from the following so we have all these URLs this is the first on the list and we see that in the text response it did respond with that so now that is a usable YouTube video that we can pass into the main article writing prompt now technically we could do this without using AI such as we could just use some pattern matching but there are potentially a bunch of different YouTube URLs that could be linked to and we really specifically want a video URL for this so that's why we're using AI to just try and get it to make sure that it will respond with that video URL if you want to see these in more detail which is to get the images and get the YouTube results then again check out the link in the description where I go through these in more detail but to get the content image let's have a little look at that first so this called file. and the result of that was this URL so let's have a look at that URL and that was a URL that was generated directly from filei and that's a flux 1.1 Ultra image that was generated directly from this automation what been uploaded directly to Wordpress we now get that WordPress URL and we're getting the media item after generating it because we want to get the correct Source URL that we can then pass into this right article prompt so now let's have a look at the right article prompt this is a pretty long prompt so pause the video if you need to we're using claw 3.7 Sonet which is quite an advanced model so it's able to really understand nuanced and detailed instructions but within the text prompt here we're saying you will be provided an article brief below write an article based on it you will be provided with a considerable amount of data and insights in the resource section It's Your Role to create an easy to understand blog post based on this in relatively concise and simple language within the air table base you can add in a tone of voice you can add in keywords you can also add in instructions we're adding those into the prompt if they've been populated in the air table base but if not we ignore it so we're using this if statement so if tone of voice optional is present it's going to write in a tone of voice otherwise it's going to add in this section here which is blank it'll do the same with keywords and it will do the same with instructions later on in the promp we're instructing it to create a highlevel NLP friendly introduction at the start here then we're asking it to create a key takeaway section there'll be a maximum of five bullet points in the key takeaway section So within the article itself here's the key takeaway section and here is the introduction we're asking it to bold one or two important words per bullet point which it did there and then after the key takeaway section we're asking it to inject in the YouTube video right below that it's added in this YouTube video which is relevant for the topic then write the remaining section of the article based on the provided brief each section should start with a h2 tag which is a head in two tag and that's in HTML we're asking it to add in this image after the first section so that's the image that was generated earlier on which is great we're asking it to include list items where necessary to demonstrate your points so if we go through that article that was created it has oh yeah it did it there which is great so there's a little bit of variation in the sections here it does not do it in every case which is kind of what you want you don't want every section to be exactly the same so again I'll just run through this outline you can pause the video and we've just added in some general directions of of writing asked to use contractions in your writing do not start consecutive sentences I have a black list of words to not include some of these are kind of dead giveaways of robotic AI writings so I've just added these in I've asked it to include a source section on the following citations below and I've provided Specific Instructions for how to format that so if we go on to the very bottom of the article and we see at the very bottom here we have the list of sources then here we're using these XML tags Claude is particularly good at been able to understand these XML tags because the these were used when training the model so what you can do is just write your prompts in this kind of format we have topic and then closing tag topic article brief closing tag article brief and then we're adding the data in from the previous parts of the automation that's a title the article outline the citations and the Deep research so we're loading all of this data into the prompt to create the article and then we're asking it to include the following internal links naturally within the text and then we're added in the internal links again earlier from the automation so let's have a look for those internal links automating business operations so that's a link to the future of AI in automating Daily Business operations that is a relevant link that was contextually added in by the AI which is fantastic technology as an anchor tag was added here and then we have another internal link of the open source AI Revolution so it's pretty cool that the AI is able to do this and then finally you must bold important words throughout the article with strong tags and you must respond in HTML format so it's bold of these words and it responded in the correct format because we send this to Wordpress and this is formatted very nicely up front after that we're generating a featured image and then getting the featured image uploading the media item in the same way as we did previously and then we're creating the post so we have the title I'm just using some replay statements here to clean up the output a bit because I was getting some display errors at one point when loading the data to Wordpress the last call to CLA here is to generate the social media text and that's going to be drafting Instagram Twitter Facebook and Linkedin text based on the article that was just generated and this is going to respond in Json format which is fantastic because it means that we can generate this all in one call to CLA instead of having to create four different modules to call claw individually so once it's done that we'll press save and we see the output of that text response is this big long output but we've passed it in to par that Json afterwards we have them as separate data items which is really good but it's not as easy to get Json output from clae as it is from open AI so what we're doing is we're pre-filling the response we're adding in a separate message here which is where we're effectively starting its response what happens a lot of the time with Claude is it's going to respond with something like here is your Jason or it'll respond with some extra character that you really do not want and we'll then mess up this automation when we try to parse that Jason afterwards so what we're doing here is we're using an assistant role which is different from when you're sending the model a message of using the user role we're sending an assistant role here and we're starting it with this curly bracket so the Json should start with a curly bracket the AI model is going to see this and say okay we have a curly bracket we need to now complete our response based on that so we're effectively starting its response on its behalf then when we're passing that into this par Json we're mapping the response from Claude but because we already finished the first character in its response we need to add in that Curly bracket to the very start of the response and then we press save the result of that is all of the drafted social media texts then at the very end we update the air table record we move it to status of article generated we load in the featured image URL the file name and then we map each of the values from that social media call so we have Instagram Twitter Facebook and Linkedin and they are all mapped there and then with an air table you'll see each of those show up separately then you go to your air table base you click on the article link you see that you have a fully generated article within WordPress that's currently saved in draft mode and then you can go in make whatever changes you want and press publish if you're very confident in this process you can even go in create a post and change the status to published then that will Auto publish the post instead of creating it in draft mode after that the status will have moved to article generated with an a table you can then go in to the social media text and adjust that however where you want when you're happy with that you can move this to push to socials again click click this automate my post button in that case you need a web hook at the start of the scenario otherwise click run once have this run on schedule it will go down through this route which is the push the socials route it will get the image file so it'll get the featured image that we got there and then upload each of those to the social networks check out the links in the description because we have individual tutorials on how to connect to each of these social networks within make.com scenarios because the instructions vary a little bit based on the social network make sure to check out the link in the descript deson to our community where we'll get access to all of our automation blueprints including the ones in this video we've got an active Community fantastic support via live workshops and tech support calls and when you sign up you get instant access to all of these courses to give you a head start thanks for watching