This AI Agent automates your social media content (no code)

In this video, we're going to be making an AI agent that can watch YouTube channels and then create unique social media content from new videos that have been posted. For this example, we're going to be making an agent that can create a post for X, a post for LinkedIn, and then a blog post. We can have it so it can make content for more social medias, you know, Instagram, Facebook, TikTok, but for this video, we're just going to stick to those three. If you haven't seen me before, my name is Rory Ridgers and I'm an AI enthusiast and I'm passionate about helping business owners integrate AI into their business. So if that's something you might be interested in, then I've left a link to my website in the description. You can book in a call and we can talk about some ways to integrate AI into your business. Also, I just want to mention at the start of the video that all the templates that I use in this video will be available for completely free in the description. So if you want to copy what I've made in this video, go ahead in the description and grab those templates. For this agent, we're going to be using make.com for the trigger, and then the actual AI agent is going to be built on Relevance AI. Let's not waste any more time, let's get into it and I can show you how it all works. Okay, so as I said, we're going to be using Relevance AI make.com. This is a Relevance AI setup. We've got the blog writer tool, LinkedIn post tool, the X post tool, and then the send content to email tool. And now these are pretty self-explanatory. The X post tool will write the post for X. Blog writer tool will write the blog. LinkedIn post tool will write the LinkedIn post. And then the send content to email is just going to send all of this content to your email. And then we have two make.com scenarios. We have the trigger here for the AI agent, and then we also have Another one, which I'm gonna show you later on, is this one, but don't worry about that for now. For now, we're just gonna start with the trigger for the AI agent. So this is how it's set up. It's quite simple. We have a YouTube's watch videos in a channel node and then a set variable node, get YouTube transcript, and then a Relevance AI node. And I actually didn't know this until recently. There is this native Relevance AI node in Make.com. It might be new, it might not be, I'm not too sure. When you go to add a module, you can type up Relevance AI and then I think you might have to download it or install it or something like that, but it's free and it makes it really, really easy to connect with Relevance AI. AI agents rather than using an HTTP Make Request node. Anyway, I'm going to show you how this scenario works. So we have a YouTube Watch Videos in Channel. So quite simple, you just connect your YouTube channel and then you put in your channel ID. You can just go to ID Finder, put in your channel handle or the channel handle that you want the AI agent to watch. So this could be a YouTuber in your niche or something like that. Just put the handle in there. and it'll find their channel ID. And then just publish before, publish after date, so you know the time range that you want here, and then the limit to 1. And now if you want it so that it is not in a time range, it's just— it's going to trigger when a new video is posted, all you do is left-click here and click choose when to start, and then, and then click on from now on. And then it's going to do all the videos that are posted from now on as long as the scenario is running. And then what we do is we use this set variable node and The reason we do this is because we need the video link, and for some reason the YouTube node before, it turns the video— it returns the video ID but not the video link. But it doesn't matter too much because a video's link is always just going to be this link here. So https://www.youtube.com/watch= and then the video ID. So we just type that in here. Add the video ID in, and then this will save as the variable video link. And then we use this Dumpling AI get YouTube transcript node, and then it kind of does what it says on the tin. It's just going to get the transcript from this YouTube video. So we put in the output from the step 4 video link, we put that into the YouTube video URL. Include timestamps, I mean, it's up to you. This is just the default settings, I've just left it on. And then Relevance AI node, we send this all to Relevance AI. This is how this node's set up, by the way. We have the role you just type in user. The content, it doesn't have to be in JSON or anything, you can just send it through in plain text, which is pretty cool. So we've got the thumbnail image link, we get this by taking the outputs from this YouTube step and then going to the high resolution one and then just tapping in the URL. And then the video transcript is just going to be the output of this step before. So quite simple, not too complicated. I'm just going to get a run through just to show you how it works. So it has got my video, Build an AI Agent That Can Scrape Anything. That was one of my last videos. And then it's set this variable, so it's given us this link here. We've sent that over to Dumpling AI, and then it has outputted the transcript here. And then we've sent that all over to our Relevance AI agent. So you can see this was the message, the thumbnail link, and then the video transcript. So perfect. Now I'm going to show you inside some of these tools on Relevance AI. So we'll start with the X-Post tool. We hit edit, and now all the user inputs for these are exactly the same. They're going to be the thumbnail image and then the YouTube transcript. So this is obviously just what's sent over in the Make scenario here, as I showed you just a second ago— the thumbnail link and then the video transcript. I'm just going to go into the thumbnail image here and then the video transcript. And then we only have one step, and that is an LLM step. And we've basically just prompted this to create an X post based off this transcript. So you can see, you can read through this prompt if you like. You are a content creator. Your task is to take a YouTube transcript and turn it into a concise, engaging post for X. Here's some instructions. I'll be giving it the tone and style, the length, the formatting, and then also the image attachments. I've said at the end of the post, embed the image via the link I give you. Don't output the link, I want the actual image to be outputted. And then extract the main idea, insight, or value from the transcript. Simplify into a single message or small cluster of points that resonate with the audience. And then I've given it the transcript here, given it the image link that I want it to embed, and then told it to output it in HTML. Just because when you send stuff on Gmail, when you send it through mate.com, you can send it as HTML and it will come back to you in the email as plain text. So just makes it look a little bit nicer. That's why we do it in HTML. And I've just put at the bottom, don't use emojis, because sometimes when you send it via email with the emojis, they come out all stretched and warped for some reason. Not too sure. So made sure that he doesn't use emojis. So we're quickly just going to test this tool out. I've got a transcript of my YouTube video here, just using a free online transcriptor. So I'm going to copy that, paste that into this input, and then I'm going to put in my thumbnail image. Okay, so that's uploaded. Perfect. We're going to hit run tool. And there we go. That was pretty quick. You can see this, what it outputted in HTML. So this might not look like much now, but when we actually see it as plain text, when it's sent through the email, it'll look nice. You can see We just got everything on separate lines here. Tired of spending hours on market research? In my latest video, da da da, want to simplify your research process? Grab free templates. Let's chat about how to integrate AI into your business. And then this is obviously the image link. So perfect, that's cool. We can save that and then come out of it. And then we have the LinkedIn post tool. And now the X post and the LinkedIn post tool are fairly similar. You can see we have the same inputs here, the thumbnail image, YouTube transcript, and then we have an LLM and it's prompted. And it does look very similar to the X Post tool, but there are a couple different things. Like on LinkedIn, it's a little bit more professional, whereas X is a little bit more laid back. And also on LinkedIn, there's a higher word count than X, so we've included that in. And then we also have a little bit about hashtags. So you can see the prompt is fairly similar. But we're just going to test this out quickly and I'm going to show you that they are in fact different. So if we put in our inputs, okay, so we've got the thumbnail image and then the YouTube transcript and then it's just generated it. You can see it looks very different to the X-Post. So we've got a big bold title and then some text here and then I think these might even be bullet points or something like that. It'll make more sense when we see it in plain text and then we have the image at the bottom. So Perfect. Not much to touch on there. The blog writer tool is where it kind of gets a little bit different. So let's have a look inside here. Same inputs again, thumbnail image, YouTube transcript, and then we have an LLM step here that is basically going to do what the X post and LinkedIn post did. It's just going to create the content, but obviously with a blog post it's a little bit different. There's a lot more text, so we've added that in. There's going to be, you know, H1s, H2s, H3s, bullet points. A lot more words. So we've added all of that in, given it the transcript and then the image link. But then what we do after that is we have another LLM step here and it's going to take the output of this step. So this is going to look like basically the full blog post. It's going to take that full blog post and it's going to find a little bit of it where it thinks that an image would be relevant and would be helpful for the user. And then once it has that bit, it's going to output a prompt for an AI image generator to then generate this image that can go into the blog. So yeah, it will generate a prompt. I said try to avoid infographics as the image generator has a hard time creating text and graphics, so more like actual images rather than infographics. And then giving it the blog post. And then we have a text-to-image generator here, and we literally just put the output of this step into the prompt because this should be a prompt to create an image that will go into the blog. And then we have another LLM step here that is basically going to take the original blog post, then take the image, and then take the prompt of what the image is, and it's going to rewrite the whole blog post, and it's going to put the image into the right bit of the blog post, and then output all of that in HTML with the thumbnail image, the full blog post, and then also this AI-generated image. So we're just going to test this out again, exact same thing. Let's put in the YouTube transcript. And then we have the thumbnail there. So let's run it. And this one takes a little bit longer because there's more steps and it's also having to generate more content, so you might just have to give it a minute. Okay, so all the steps have just finished. We're going to quickly look through what happens. So the first LLM step, it created the blog post, output it in HTML. You can see it's quite a lot longer than the LinkedIn and X posts. Okay, so nice. And then this LLM step here, in the blog post, suitable area for an image would be the section titled Meet Your Agent. Here's the prompt, giving us the prompt. And then we put that into the image generator, created us this nice little image here. So if you've watched that video, you know that we talked about Instagram, LinkedIn, Trustpilot. It's managed to get them in, which is pretty cool. And then this LLM step here should have just put them all together. It's kind of hard to tell while it's in HTML, but we should be able to see— you can see this link here has the try relevance in it, so that means it's the— this image here. So it has managed to put the image into the blog post, which is really, really good. That's exactly what we want. So perfect. Now I'm going to show you the last tool, and then I'll show you inside of the actual agent on Relevance AI that puts all of these tools together. So the last tool is Send Content to Email. And this is kind of more Make than Relevance AI. I'll show you what I mean. The only inputs are the X-Post, the blog post, and the LinkedIn post. This is literally just going to be the outputs of all of those tools before are going to go into these respective inputs. And then we have an API step. It says invalid parameters. I'm not sure why, because I've tested this and it does work. And all we're doing here is we're connecting this to a Make.com scenario via a webhook. And if you haven't done that before, I'm going to show you how to do it. It's really, really simple. We're quickly going to come over to Make and show you inside this scenario that connecting to. So this is what it looks like. We've got a webhook, set multiple variables, and then a Gmail. If you haven't used a webhook before, what you do is you just put it down, you'll get this URL here. And when we're connecting to it on Relevance AI, we do the method as POST because we want to send data to the webhook. And then the URL is just going to be this URL here. And you put that in, and when you first set it up, you'll have to redetermine data structure and then run it, and then they should connect to each other. And then we don't really need to do any headers for this. We can set the body to raw. And then this here is the content that's actually going to be sent over to the Make.com scenario. And we've got the X-Posts, the blog post, the LinkedIn post, and these are separated by dollar signs. And I'll show you why in a second. It does make sense. So we send all of that off to the Make.com scenario. And then what happens is we have to set multiple variables. So this is why there's a dollar sign. So what we do, basically, this data here will get sent through as one variable, and we want to split that into 3 separate variables. So what we do is we separate them with a dollar sign and a space, and then we go to set multiple variables. And then what we can do is we can use these equations here on Make, and we can split up this value based on this, if that makes sense. So you put in your variable here that you want to split up, and then you put in here what the sort of separators is. And usually it'd be a comma, but in this HTML there's loads of commas, so I found that it was getting split up really weird. So I did a dollar sign because I thought that probably won't be in it. Splot. So split up by a dollar sign, and you can see there is also a space here. So a dollar sign and a space. And then what we do is we put in number 1 here because this is going to be the sort of first variable in here. You can see X Post is the first one in here. And then we do the same thing but for number 2 to get the blog posts because that's the second one after the dollar sign. And then third one is going to be the LinkedIn post, as you can see. So I hope that makes sense. When we run it through, it'll make a little bit more sense. And then all we do is we just send this to email. So I put in my email here, subject line is going to be new content, and then I've put in the data here. And if this looks a little bit weird to you, it's because I've got these HTML tags here, which is just br, which is just break or line break. Basically just starts a new line, because if you just hit enter like that, when it comes through on Gmail it will be on the same line. So you just do these little brackets, br, and then a closing bracket, and it will line break. So it'll just go down a line. And then we just put in our posts. I've labeled it X Post, and then the X Post will go there. Blog Post, Blog Post will go there. LinkedIn Post, LinkedIn Post will go there. And you can do these as attachments as well, as PDFs or Word docs, whatever you prefer. But I found it kind of got a little bit skewed with the HTML. It kind of didn't look right when I sent them as attachments. So I thought I'd just send them in the email body and then they're formatted correctly and they look nicer. So just to give you an example of how splitting up these variables work, I'm just going to put in some placeholder information. So I'm going to say 1, 2, 3, because I'm not writing out the full HTML for all of these different inputs. So we'll just do 1, 2, 3, and then we'll send this off to the Make.com scenario. You can see how it's split up. You can come over to Make and see it's been triggered. 1, 2, 3. Obviously it's 1 dollar sign space 2 dollar sign space 3, and then it is split up like this. You can see this is how it looks. So this was what was inputted, and then it was split up based on that dollar sign, and then it outputted X Post, Blog Post, LinkedIn Post. And then when we want to use these variables in other steps, you can see we have the separate variables here, 1, 2, and 3. So that's how that works. Now we're going to put all of these together in an AI agent and you can see how it would actually work. So obviously I've already gone ahead and made the AI agent. If you're going to make a new one, you just hit New Agent, but I've already made one here, the Social Media Agent. Okay, so we've got an agent name, agent description. The agent name, Social Media Agent. Agent description, creates content for social media from a YouTube video. The name and description don't really matter for now. They will matter when we come on to use multi-agent systems, but For now, don't worry too much about that. The main thing that we need to do is add in the tools. So to do that, we just hit Add Tool, and then you can select the tools you want to add. And something you need to do— this, this by default will be set to Approval Required. You need to set it to Auto Run just so that the agent can automatically use these tools. Then we're going to look at the core instructions here. The role— I've just said you're a social media assistant that can generate X posts, blog posts, and LinkedIn posts based on a based on a transcript from a YouTube video. Then I've said you have 4 tools. If you want to be able to directly reference the tools, by the way, just hit a forward slash and the tools will pop up here. That way the agent knows exactly what tool you're talking about. So when you're triggered, you will receive a thumbnail image link and a YouTube video transcript. I want you to use the X-Post tool, Blog Writer tool, and LinkedIn Post tool first to create 3 pieces of content, and then send all of that content to the Send Content to Email tool. I'll give you a breakdown of each of the Tools now. Then here I've just given it a breakdown of all of these tools. This tool will create a post for X based on the YouTube video transcript. The user input required are thumbnail image and YouTube transcript. I want you to put the thumbnail image link into the thumbnail image input and the YouTube video transcript into the YouTube transcript input. And for the X Post Tool, Blog Writer Tool, and LinkedIn Post Tool, the descriptions here all pretty much are the same except I've just said create a post for LinkedIn and create a blog because the inputs are the same and they pretty much work the same way. You just input the data and it comes back with a piece of content. The only thing that's different is the send content to email tool. I've said this tool will send all of the outputs of the previous tools to make.com to then be emailed to the user. The user inputs required are X-Post, Blog Post, LinkedIn Post. I want to take the output of the X-Post tool and put it in the X-Post input, the output from the Blog Writer tool in the Blog Post input, and then the output of the LinkedIn Post tool into the LinkedIn Post input. So simple enough, we can now test this out and see how it works. So what we're going to do is we're going to manually trigger this scenario because we're not going to wait around for a new video to be posted. So if I want to manually trigger, I can just go choose where to start, since specific date, and then I can choose a day. I'll just say the 15th. I've obviously put that before, so I think I posted this last video around the 16th or 17th. So we hit OK, hit run once, it should pick up the video. Which it has. Build an AI agent that could scrape anything. And then it's set the variable, which is the link. And then we have got the transcript here. And then we've sent that all over to the relevant AI agents. Now we should see it inside of the agent. Yep. And now this does take a minute just because this is such a massive input to put into all of the tools. So it does take a few minutes sometimes, but This is going to be running autonomously in the background after you post a video, so it doesn't really need to happen instantly. As long as you get these pieces of content within sort of 5 minutes of posting a video, I'd be happy with that. So I'm just going to give it a minute and we're going to come back when it's finished. And I could have made this where it automatically posts this content to your social media, but I haven't done it that way just because I feel like AI isn't fully there to make these sort of decisions on its own. I mean, can for some people, but if you're sort of really, really particular about what goes out on your social medias, I think it's best for the content to be sent to you first for you to review, and then it could be posted out either manually or by the AI agent. In this case, I didn't set it up where it could be posted by the AI agent if you approved it, but we could easily do that just with some Make.com scenarios. They can all connect to these softwares and post content, but for this video I didn't do that. Okay, so you can see it's used the LinkedIn post tool And now, okay, sometimes with this LinkedIn post tool, it wouldn't put the thumbnail image for some reason. It was really, really weird. So all I did with this is I tried to prompt the AI agent a little bit better and tell it to actually, you know, put the URL into this input. But sometimes it still wasn't doing it. So what I did was I set this input to optional. That way, if the AI agent didn't do it, the whole tool wouldn't just break, because if the inputs aren't there that are required, the tool just won't run. So the tool just wasn't running if the AI agent didn't work out how to put the thumbnail URL into this input, but looks like it's done it this time, so that doesn't matter. It's also saying it sent the content to us via email, so we can have a look. Okay, so this is the email just received. Hey, I've generated an X post, a blog post, a LinkedIn post based off your recent video. Find them below. I hope you like them. So the X post, we can read this. Tired of spending hours on market research? Check out the free templates in the description. And then it's given this image. The image is a little low quality, but I think this is just where I'm previewing it through Gmail. And the blog post, same again, a little bit low quality image, but I don't think it matters too much. And then we have this, create an AI agent for market research. What is an AI agent? The tools you'll need. That's really cool, it's going through the transcript of the video and worked out what tools a user would need to build this. And then you can see we have this AI image here. It's actually done a pretty good job at creating these different tools. So Lead Researcher tool, Trustpilot web scraper. Instagram scraper, LinkedIn. Yeah, I mean, it's not perfect, but it's pretty much there, which is really, really cool. And obviously you can fine-tune this a little bit so it only creates images that you actually want. This is sort of just a demo to show you guys how this would work. And then it's given us a step-by-step guide here. This is really cool. A conclusion: visit my website. Let's see if that link works. And it does. That's really cool. So we've got the LinkedIn post here, unlock the power of AI for market research. It's giving us the tools again to put some hashtags in, which is really cool. And then it's given us the image again at the bottom. So as I said, I could have set this up where if I just replied to this email saying yeah, they're good, I could have had the AI agent then send these out quite easily. Actually, all I'd have to do is set up a couple Make.com scenarios with webhooks, and it could just take this HTML data that would be passed through the webhook and then we could just set up an X node, a LinkedIn node, and then also a WordPress node or wherever it is you're using to host your website, and then just post this content. So it could be really easy to do. We can also set it up where we can make some adjustments. So I could say, yes, the blog post and the LinkedIn post are good, but for the X post, can you add some hashtags or something like that? And we can have the AI agent then go back into the tool and regenerate it how we want it with these Adjustments. Okay, so you should now have an agent that can create unique social media content completely autonomously after you post a YouTube video or someone else posts a YouTube video. And as I mentioned at the start, you can do way more social medias than this. You aren't limited to just X and LinkedIn and blog posts like I did in this video. They're just examples. Uh, yeah, this is a really, really helpful tool, especially if you're a content creator that's also running a business like me. It can be kind of time-consuming sometimes to get content out on all these different social medias. So this would be a really, really valuable tool to someone like me, or as I said, someone else that's a YouTuber that's also running a business. So if you were interested in working with me so we can integrate AI into your business, as I said at the beginning of the video, I've left a link to my website in the description. So if you want to go ahead, book in a call, we can talk about some ways that we can integrate AI systems like this into your business. And also, if you want to go ahead and grab all the templates I've used in this video, they'll be available for completely free in the description. Also, if any of you have some video suggestions that you'd like to see, please do just drop them in the comments and I'll try my best to get around to doing them.