142 | AI Meets Automation: Build a Content machine that converts and scales with Keith Moehring

27:54with some additional details and instructions as needed. If you're using Claude, for example, you really contain everything within the make call or the make prompt. And there's not really much you do on the Claude side to pre-build it, if you will. So the answer to the question is both depending on the tools you're using. Yeah. So, but yeah, this is it. This is the long and short of a very simple example of how you can take a client's success, create a case study within two seconds using these automation tools. But I thought it would be good if we took a more kind of a deep dive example into the real, like really unlock the power of what these tools can actually do.

28:38So what I've done is I've gone through and when you see it on the make, because we'll actually open up make and walk through each of these steps. But this is how, this is the number of different operations that are going on within that one make scenario that we'll take from a thought leadership content perspective. So anything else you want to cover before I jump into the kind of the main show? No, I think we can jump into the actual live example. I think it's going to be the most clear for people and we're going to try to make it as simplified as possible. So it will be easy for you to follow up even though you can't watch it on the screen.

29:13But since you're not watching it on the screen and if you've never used make before, the way it looks like is every one of those steps looks like a little circle and you start, in theory, you can make it move in any direction because you can move the stuff around. But the common way it shows up is it starts on the left and moves to the right. And you can have those intersections where now you can do like ifs and have a fork in the road and go in one direction versus the other direction depending on what's happening. So think about a flow of little circle milestones. Each and every one of them is a step with a different tool and then the progress moves to the left.

29:49And now Keith will describe to you what each and every one of those steps and what the process overall does. Yeah. Yeah. And this one, the purpose of this automation is to, if you say someone publishes an article and you want to write a article or a blog post about that article, but in the style of your subject matter experts, you need to one, kind of if we were taking it as a manual process, What you would do is you would take the article, you would have the subject matter expert read through that article, summarize it, pull out some quotes, and then they would go and they would draft a version of their own article in response to that or commenting on that information.

30:33It would match their tone because they're writing it and their style. And then you would hand that off to your SEO editor. So who would go through and optimize the content to make sure that it has all the right wording in it. It's got all the right features built into it and all that it's everything's optimized for search performance. And then you would take that article, pass it along to an editor who would review and approve it, make sure it's all polished and good for the company website or wherever it's going to get published. And then ideally, you also have social shares that can be created off of that to promote the article.

31:08So maybe we generate two or three LinkedIn shares to talk about and feature the article and drive people back to it. And then ultimately, that stuff needs to be reviewed and approved by your editor. So we'll create a task for them. This workflow does all of that in one and two to three minutes. So I'll walk you through and we'll get into as detailed as we need to on how each step is set up and how it works. So the first part of this, so in this case, this scenario, the trigger for this is called a webhook. And a webhook is essentially a web page or a page on and make site where as soon as information is sent, it runs and triggers the automation.

31:54So what I've done and what we have set up here is, let me get rid of this here real quick. And again, I'll explain in more general is what a webhook is. I told you before that a lot of these tools are already pre-connected. some of them are not connected either at all or not connected in the way you want them to be connected and so a webhook is a non-developer so it's a quasi-technical option that allows you to add more capabilities to these connectors so you can you may connect something in a tool that already exists but you want to trigger it in a different way that doesn't exist in the tool today or you want to connect a different tool most tools today support these webhooks and they're very easy to build.

32:38So that's like a quick explanation of what web hooks are. Yes. Thank you for that. Yeah, that's a good reminder. I have this website, this Google sheet set up. And the idea with this is anybody who finds an article that's worth writing about can add that article directly to the sheet. And so I can add in the date, the name of the article, a link to the article, define the topic the article covers. In this case, I've just who wrote it. And then I've got the final step here. This final column is the trigger column. So in this case, you could set up the trigger so that as soon as a new row is added, it would trigger the automation.

33:17The trick there is that the person who's adding in the article may not act fast enough. And so as soon as they enter in the article, maybe only half the information is available and then it triggers the automation. and then all of a sudden it doesn't work. So there's some issues with setting that up to run like that. So in this case, though, I built a tool like with Google Apps or Google Sheets. There's a Google Apps Script tool that you can build. And again, you can use ChatGPT to help you program and write these little tools. But the idea with this one is on the back end, as soon as I update this to go, So the automation, it's actually going to send this row's worth of data off to this automation through this webhook.

34:05So again, for those of you who don't see the screen, the final column on the Google Sheets is basically the trigger. So there's a drop-down menu in that Google Sheets that says whatever, but one of the options is go. When you switch it to go, that's what triggers the automation. That's basically what that webhook listens to. It's looking for a new one of those to change to go, and then it will run the thing. if you don't know how to do the script, you could have said, once there's the number one on this column, do the same thing. And then it doesn't look as fancy as Keith's spreadsheet, but it will do the same thing.

34:38Because once the number one, or once you type yes, or once you type go, or once it has any content in that final column, only then it will trigger the row to actually get the information. And that verifies that all the information is there before you send it over. Yeah. So this may be a little bit of a kind of a advanced way of doing it, or you could just do very simply, as soon as a new row is added, that triggers the whole thing. But the way this works then is, so the information comes into the trigger and that row's worth of data is then passed along to this first operation. So the first operation, and this is one that's natively built into make.

35:17And what it does is it will take a, I'm going to open this up here, it'll take a URL. So the information comes into this. And as part of that, what's created is a packet of information, one of which is the article URL. And so I can add that into that first field in this first operation. And what it will do is it will go out to the Internet and scrape everything that appears on that web page. In other words, it's going to grab all the HTML. It's going to grab all the content. It's going to grab all the links. Everything that's on that page will come back in this one object. And it's not overly usable when you've got all that HTML code baked into it.

36:00So that output here is the HTML code of the full web page. So I'm going to pass all that packet of information into that next operation, which essentially strips out all the HTML from that text block and leaves only the content behind. and HTML goes in, out comes just the text block of the article itself. Real quick, if you're setting something like this up, the thing to note is these tools can't go through and grab content from like a paywalled article. So it's got to be a freely accessible article online. But if you can do that and you can get it in here, then this will strip it out. And now I've got the content of the article.

36:47Just a quick question, because there's a question, how does it strip it out? Is there like a tool built into Make that just does exactly that? Yeah, it's a native operation that's built into Make. And so if there's code on the back end that actually just goes through any sort of HTML elements that exist, it will just completely strip them out. Awesome. Yep. So now I've got the copy of the article. and what I'm going to do is then I'm going to send it into my first of several OpenAI assistants here. And so what I'm going to do is it's going to go in to this first one, which is called the Article Analyzer.

37:26And all I'm doing in this little widget here is I'm calling out and defining what assistant I want to use within my OpenAI account, defining myself as the user, and then in the message, All I'm doing is adding in that block of text from the article. Now, if I go into OpenAI and open up that assistant, this is what the assistant looks like on the OpenAI side. So I have the full prompt here. You're a content marketing specialist. Read the provided article and provide a summary in two to three sentences. Grab two to three notable or quotable statements from the article. Make sure they're pulled verbatim.

38:05And then also grab the author name. And in this situation, what I want to do is I want to send it back in a structured format. So I'm using what's called a JSON structure where I give the data a label and then save that information. So the first label in this JSON structure is summary. And then it'll add in the summary of the article next to it. And then the next one is quotable statements. And then it'll add in the quotable statements next to that. And so what I'm going to do is then I'm going to take this assistant is going to send that information back that is extracted out of the article and it's going to pull it right back into the make scenario in that format.

38:51So that's the. I just want to pause you for a second for those who are not watching this and just listening to this. First of all, there's a YouTube version of this. So if you want to watch what we're doing on the screen and you're listening to a podcast and you're not driving right now, because if you don't do this, I don't want that on my liability. But if you can't watch the video, there's a link in the show notes that will take you to see the video on YouTube if you want to do that. But to recap, we've done four steps so far. Step number one, start of the process based on somebody filling up a Google Sheets.

39:20Step number two, grab the HTML from the article. Step number three, turn the HTML into just the text. And step number four summarizes the article and puts it in a format called JSON, which you don't need to know what it is, but it's a standard structured format that multiple types of software know how to use. So think about it. We like to put stuff like that in the table. That's like the digital version of a table where things are very clear what is what as far as the data package. And then you can send it to multiple different tools so they know what to do with it. Exactly. Exactly. And this and this after this step, again, I'm sorry for the podcast.

39:57This is a very visual type of demonstration. Hopefully you do get a chance to watch the YouTube version of this. But so once we've got the JSON summary of the article, and although we got summary that go out the quotes, we got the author. There's one of two things that happen. We'll get into the second thing. So the secondary version of this is every so often with these GPT or the generative AI tools, they will run into errors and they will return an error. Something won't fire for whatever reason. I'll get into how we handle that at the end. But assuming everything comes back from this tool properly, it's going to go into the next operation.

40:35And so the next operation is designed specifically to create a usable variables off of the JSON that the GPT created. In other words, I'm going to have a independent variable called summary with the description or the summary of the article. I'm going to have a variable called quotable lines with those built into it. And I'm going to have a variable for author name. And now because I've got each of those individual pieces of information available to me, I can use them wherever needed throughout the rest of this workflow. So now once I've got that information saved, the next step is to, and this is probably one of the cooler things that you can do.

41:20It's definitely, I recommend it. It's going to give your automations a lot of power. So in this case, what we've got here is I've got the summary of an article. I've got quotations. I've got the topic of the article. But I, in my company, I have a subject matter expert for content marketing. And I have a different subject matter expert for HubSpot. And I have a different subject matter expert for generative AI. So I have all these different topics that we may write about and a subject matter expert for each of those. And each of those subject matter experts has their own unique style and tone.

41:55They have their own examples of work and social shares and all that stuff. So what I've done is I've created a Google sheet. And within that Google sheet is essentially, it's a directory of resources specifically related to that topic and to that subject matter expert. In this case, I'm showing now a Google sheet that in column A is a list of topics. So content marketing, paid search, generative AI, and HubSpot. Next to that is a column called semantic keywords. And so we'll get into the SEO side of this here in a little bit, but it's essentially a list of IDs for documents on Google Drive. So this is a Google Doc ID for content marketing semantic keywords.

42:43So I have one doc on my drive with a list of keywords, and that's the ID to that specific document. The next column over column C is the subject matter expert tone and style. So we have descriptions of their tone and a description of the style they like to write in. And that's saved in a specific doc. I have a example blog posts of past posts they've written in a doc. And that's in column D. So the ID for that document on my drive is saved there. And then the final doc is, or the final column E is their example social shares. Again, I've copied and pasted them all into a Google Doc. And I've put the ID of that doc in this row.

43:23So within the make automation, what happens is this will go in and open up that file. And what it's going to do, it's going to filter it so that it only returns the row that matches the topic of the article. And that was something that the user inputted into the spreadsheet when they essentially put the article to be created. So in column D of that first spreadsheet that we set everything off with, the trigger spreadsheet, I've defined this first article as content marketing. And so now I'm going to use that phrase to look for the row in my directory that matches up with content marketing. And as part of that, what I'm going to pull in is the ID for the keywords, the tone and style, the blog post and social shares all related to that content marketing.

44:12So I want to pause you just for one second, A, to help people understand what we're doing and B, to say how that connects to what I said in the beginning. And I'm actually going to start with that. Like I said in the beginning, that part of the problem with just creating content is that it does not take into account your tone of voice. It does not take into account keyword research. It does not take into account your expertise in that topic and stuff that you've written before. And what we're doing here is basically solving for literally all these things in this one step. And the way this was done is in advance, Keith has prepared these reference files that can be used.

44:52And this could be different reference files for different topics. So for each of those topics that he mentioned, there are different reference files, different keywords, different tone of voice, different referenced past experience on stuff that you want your company to talk about. And what this amazing step does is it really prevents you from running four different automations, each one for a different topic, which is what I would have done before I learned this trick from Keith. So before I saw Keith do this on stage, I literally did this once and then created the automation again to be triggered with the second topic and then created the automation again, created with a third topic.

45:31And if you have 20 topics, now you have 20 of these automations, which makes absolutely no sense. So what this step does is it literally says, oh, if this is your topic, use these three files as references for your tone of voice, for your past expertise, the stuff that you've already written and examples, and for your keywords. And then you continue with those in mind and you can show us now how this is done. Yeah. And I'll show you here too. So these are the docs that it's pulling that information from. So the ID for this doc, you can usually see it in the URL of the Google doc. You can do the same thing with Word docs and SharePoint.

46:09But this would be my list of semantically related keywords to content marketing. So everything I write about content marketing, I want to feature and include these specific terms. As the author, here's my tone. Here's my style. Again, it's just a block of text written out, but it accurately describes how I write and the style in which I write. And then I've got a list of, okay, here's some example blog posts I've written on this topic in the past. And again, we've got my style, we've got it, but also it's got bits of my own expertise baked into it because I wrote these articles. And the final step is example blog posts.

46:45So example LinkedIn shares that I've put on about these topics. Again, copied and pasted verbatim. But now the automation has access to all of this information that I can then apply to the next thing that we're going to do, which is going to be to write the article. After it goes through that directory, now I've got essentially a list of IDs that I can utilize for the next few steps. So the next step here is, and the next operation is going to go into Google Drive, and it's going to grab my tone and style guide. And it's going to grab the content off of that document. So now I've got those blocks of text available to me within this workflow.

47:26And then the next step is it's going to go in and grab the past article content. Again, pulled directly from Google Drive. So I've got the text verbatim sitting there. And then the next step here is I've got everything I need now. I've got the summary. I've got quotable links. I've got my tone and style. I've got past articles I've written. I've got everything I need to go and write this article to have chat GP. or in this case, we're going to have Claude, because I like using Claude to actually generate the content. I'm going to have Claude write the article. And so as part of that, I'm going to pass in to this operation.

48:02So AnthropocClaude, I'm going to set it up so that messages is from a user. It's a text prompt. And I've got my prompt written in here. So you are a content marketing specialist and ghostwriter. Please write a 300 to 500 word blog post based on a recent article about, and then I'm going to put in the topic of that article. And then it says, use the following information to create the blog post. Here's a summary. And then I put my summary variable in there. Here's my quotable takeaways. And I'll put a list of those in there. And then I also then provide it. Here's the subject matters, tone and style.

48:38And I include that variable. Here's past blog posts they've written. I put that in there. And then I essentially lay out the task, review the summary and keywords, analyze the SME's past writings, and generate a point of view on the topic using supporting evidence and insights, and then draft the copy. I also make sure that it calls out and said, please include quotes from the author throughout to reference their piece of content, and then ultimately have it spit out something of, generate the blog post, and then provide it in a format that I can use. So Claude will go through, do all of that work.

49:20And on the other end of it, I have the draft of a blog post ready to go. And now I could end it here. I could simply just save this doc on my drive and be done with it. But the other thing that I really like doing, and I forget who mentioned this to me, but I really thought this was brilliant, is you can have one of these generative AI tools write the article. But what's cool and handy is to have a second, a different generative AI tool, actually review it as if it's an editor. So you get essentially two different training data sets to draft and then to review the article. The other thing that I've done with this second one is I've set it up so this editor is also an SEO expert.

50:02So at this point, after the article is written, I have the automation go in and grab the list of semantically related keywords for this topic. And so now I have that keyword data available to me within here. And then I'm going to pass all of that information over to a second OpenAI assistant. And that second assistant is, oops, wrong button. Okay, so I call this one my SEO content editor. And in this case, I've already preloaded it with a prompt of you're an SEO expert, content optimization. And what I've also done is because you can do that with these assistants is I've preloaded some resources into this assistant.

50:45So I've got a Google SEO starter guide for optimizing for Google, avoid search engine first, whatever it is. I forget the name of the article. In other words, I have just given it an education in SEO best practices. And I've baked that right into the assistant. So when this thing fires and this thing runs, what it's going to do, it's going to run through this prompt, but it's also going to draw from all of those resources and essentially give it a solid expert level background on how to optimize content to perform well on Google. So when I call this assistant here through the automation, I'm going to provide it the keywords and I'm going to provide it a draft of the article copy that was generated by the Claude operation.

51:32and what it's going to come back with is a version, another version of that article, but optimized with new keyword, with the semantically related keywords. It's also going to come back with new features to add to the article. Say for example, a frequently asked questions section, or here's the recommended page title and meta description for each of those for that article. And so by the end of it, I'm going to have a fully optimized piece of content with additional features to bake into it. And now I can take that information and I can save that full drafted out, optimized blog post on my Google Drive so that when it's, and now that's the thing that can be reviewed by my content editor.

52:15So first of all, this is absolutely brilliant. I want a quick summary of this step and we're short of time. So maybe we'll stop here. Maybe we just want to show the last few steps quickly. But what this really does is you created another step that is an SEO expert. And the way you create the SEO expert, you give it a lot of SEO best practices. So you give it whatever articles or books or whatever they are that you want to reference for this thing and say, okay, here's a blog post that I've written. Here are the keywords I'm trying to optimize for. use all that knowledge of best practices in order to evaluate and update my initial draft in order to make it a better version.

52:59And you can do this for literally anything you can imagine, right? This could have been any other step, but this just allows you to, when you hand it off to a human evaluator, already after it's been optimized once or twice through these different filters. The other thing that I want to touch here is, in theory, this could have all been done in one step, but these tools, A, this allows you to use different tools for different things. So in this particular example, you're writing with Claude, which is probably the best writing tool out there right now. And then you're evaluating with Chach EPT.

53:27So that enables you to do that. But also, even if you're using these tools, the same tool, it's always better to break it into individual steps because then you're going to get a better outcome for each of the steps versus if you try to combine it all together. And the last question was a question from somebody in the audience who's watching us do this. So those of you who's not watching us, every time that Keith mentions parameters in the prompts that he's sending, they show up as these little colorful boxes because what they are, they're literally parameters from the previous steps that we've done in Make.

53:59So whenever you want to enter a parameter, you do a forward slash, you get a drop down menu, and it shows you all the different things that happened in the previous steps. And you can pick the parameter you want and basically place it as part of what you're sending over to the chat. So every parameter that happened in any previous step is now available to you to build the next step and send the information over. Yeah. And just to hit the last couple of points on this then is after that draft is done, what I'm going to do is pull the past social shares for the author and send that into Claude. And it's actually going to generate for me two to three LinkedIn shares to promote the article.

54:39I'm going to save that information to a Google doc. And then at the final step of this, so essentially the output, what I'm going to do is assign my editor to go through, review the content, review the social shares, get it all set up and polished up, and then upload it into wherever it needs to go, whether it's LinkedIn or the content management system. That's the one thing I will absolutely, I want to make sure that I leave you with, never, ever let these tools just publish automatically. You have to have a human in the loop. You have to have someone have eyes on it before it goes published because there are things that will come out of this.

55:16You'll be like, where did you get that from? But that kind of stuff happens. So you need to have that extra set of eyes on it. Awesome. Yeah. Keith, this was absolutely brilliant. I think this process is literally mind-blowing. And again, this is very well-structured and it ticks literally all the boxes that I talked about in the beginning that are problematic with just, oh, I'm going to have ChatGPT write a blog post for me because it misses all these really amazing capabilities. And I want to touch on two big other points that are critical. One of them you already touched. You can put as many human-based steps in the middle and the way to move forward in a human step.

55:59So let's say you wanted another evaluation step in the middle is you're in your case posting the task to click up, but this could have been Monday, Asana, Jira, like whatever task management tool you're using. And you can say that once you move it to the next status of tasks, so this is for review, it goes from for review to reviewed, that's going to continue the automation from there or start a new automation, however you want to build this. So you can do that as well. And that's how you can get humans in the loop in that process. And the other thing is this whole process takes minutes. So this whole thing that would have taken a company of people, either because of handshake time.

56:38Okay, I got an email and I didn't check it and I checked it the next day or the time it takes to... Would have taken days to do. Like this whole process that Keith just showed us takes minutes to run. And then you just save people in your company a few days of work and it still goes to a human reviewer that can then edit and make whatever changes. But he's already starting at 80 % of the work he's done. In some cases, 100 % of the work. But in, let's go for the worst case scenario, 80 % of the work is done, you add your two cents, you make some minor changes, and you're ready to publish. And you saved a few days of work, both in means of effort as well, definitely means of timeline.

57:19As I mentioned, absolutely amazing, Keith. If people want to follow you, if they want to connect with you, if you want to use your services, if they want to know more about what you're doing, what are the best ways to do that? Yeah, I'm on LinkedIn. I'm also, our website is It's just l2-digital.com. So we're ramping up and putting up a bunch of resources, kind of talk through and show how to do all of these types of automations. Got a bunch of specialized use cases for those. So those would be the two best spots, LinkedIn or through the website. It would definitely be the easiest and most efficient ways.

57:54Awesome. Everybody, by the way, both on LinkedIn and on the chat on Zoom are going like, oh my God, I'm blown away. This is the best step-by-step breakdown I've ever seen. there's a bunch of those on both chats. So as I said, that's how I felt after seeing your presentation on stage. And I'm very happy that I got a chance to host you here and share this with my audience. Thank you. Thank you. Thank you. No, I appreciate it. And just for the fun of it, when you started talking about how long this thing takes to do, I put go, the process is already done. I've got the task. That's awesome. Great.

58:28Thanks everybody who joined us live. As I mentioned, we do this every Thursday at noon Eastern. We also have the AI Friday Hangouts, which we do every Friday at one. And this is just like an open mic kind of thing. Anybody who wants can join. And we have conversations about what happened this week in AI and problems people are having. And we try to help each other solve them. So more of a community get together. And it's a lot of fun. So if you want to do those, again, there's a link in the show notes where you can come and join us to these events. and again thank everybody for joining us live whether on LinkedIn or on Zoom we had a few dozens of people join us and I appreciate all of you and I know you have other stuff that you can do in the middle of your day on Thursday so thank you for being with us and thank you again Keith for doing this and for sharing your brilliance with all of you.

59:13Thank you and thank you for having me.