Can AI Make Marketing More Human?

Mick (00:02) anytime I found myself prompting, I have more than one prompt. It's kind of the same topic, if you will. And I mean like three times. If I've done it three times, then I have a need to make a GPT, especially for internal stuff. Benjamin Ard (00:39) Welcome back to another episode of Content Amplified. Today I'm joined by Mick. Mick, welcome to the show. Mick (00:44) Hey Benjamin, glad to be here. Benjamin Ard (00:45) Yeah, make I'm excited. This is going to be a subject that is going to be a ton of fun and you have some crazy cool examples that you can share. So I think everyone's going to get a lot out of this episode, a lot of tactical and strategic content. But Mick, before we dive into that, tell us about yourself. Let us get to know you here about your work history and background. Mick (01:05) Yeah, for sure. As he said, Mick, I work with Power. I'm head of growth marketing at Power. I've been with them since May of 22. It was my second company in the software space. And it's definitely its own animal. We're on a freemium model. We're fully product-led with no sales team, which presents its own set of challenges, especially as it relates to content, as you know. Been in the space for 20-plus years. Most of my time was in outbound sales. and then made the switch over to marketing full-time in 2017. Married with one kid and one nice little pity named Jolene, and we live in the mountains of Georgia now, and I spend all my time behind this monitor and not in the mountains yet. yeah, that's quick little notes about me there. Benjamin Ard (01:49) Love it. That's amazing. Well, Mick, you have taken your AI usage to what I would consider the next level. Like a lot of us are really good. We have some good memorized prompts and things like that. But what you've done is really delving into the world of custom GPTs and not only for like your own business, but things that you've shared with other people to help them even on a very basic level. So to take a step back just to make sure everyone's caught up and on the same page before we go deep. What is a custom GPT and what is kind of your experience been with AI thus far and how is it helping you? Mick (02:23) Yeah, I was what I consider kind of an early adopter with ChatGPT. I got in soon. I realized its potential, first of all, as a work partner more than anything else. I never looked at it as something that could replace marketers on the team or anything like that. I saw it as an opportunity to level us all up to another degree. So early on, obviously, I learned that how you prompt and what you say in prompts has a lot to do with the output. and then came along custom GPTs where you could basically make your own version of ChatGPT for a unique and specific purpose. So ⁓ that's where I started. And like with anything, I looked at my overall weekly structure and the different sprints that we run and what are we spending the majority of our time on. And the majority of time at the beginning was actually editing articles that were incoming. I don't have a huge content team, so I rely on ghost writers all over the world that write content for us. But I have very strict guidelines that obviously follow SEO for actual copy ready production. So in the early stages, it was myself and one other person that was manually editing all of these articles. So we had a backlog of three or four months worth of content that was ready to be edited, but it was taking so long to get them actually done. So the very first GPT that I made, I called an Article Draft Inspector. And basically what I did is I took our guest post guidelines that are very detailed from start to finish, I put that into the knowledge base of the GPT. And basically I set up that GPT to inspect every article draft that we had come in. And it would give a set list of recommendations of things to fix, whether it be your meta title is too long, there's no FAQ. your images don't have alt text on them, all these things that are important for SEO and ranking, it would give me basically a checklist. And then we would take that checklist, give that to the author and say, hey, these things need fixed. So before where it would take us a day to do three or four article drafts, we were getting 20 and 30 done in a day. So we were able to really, first of all, replace that time with this GPT to do that for us. I knew nothing was going to be missed and we got our content scheduled. far enough in advance, we're scheduling articles right now for October. So we're that far ahead. And these are fully SEO optimized articles that are scheduled in our CMS queue to be published. that was the first one that I made. And then it's just been like a run to see how many of these things can I make at this point. Benjamin Ard (04:39) Wow, that's incredible. love that. So that's amazing. So you saw the need. This is taking forever. Can I use AI to fix that problem? And it is absolutely amazing to do that. when you're making a custom GPT, how are you training it to get the results? Like, do you throw an additional training material? trial and error, like what's kind of your standard guidelines about how you make these to be productive? Mick (05:15) The one big thing was adding knowledge base to the GPT. That was the one thing that I noticed that seemed to really help. And I guess in a roundabout way, that should be included in the time to make one because I do do a lot of research beforehand. And you can save it in any type of context. You can save YouTube clips in the knowledge base of a GPT. So that was the one big thing I noticed was really helpful, especially once I got into ones that were more complicated, was adding... adding files and things to the knowledge base. And once I did that, then I would just iterate and test. I would test 30 times and I would notice little things. I would keep notes in a separate notes tab of things that I noticed and I would give all the notes at one time. And that seemed to go faster too. Instead of like nitpicking one, I would just get a list of all the things that I didn't like or... It gave things the wrong way using very clear descriptors like always and never are really important. So little stuff like this, just I learned trial and error like a lot of stuff. Benjamin Ard (06:14) Well, and it's just like teaching any human, right? Like as you ask them to do something, you get the output and then there's training and it's kind of the same process for AI to learn. When you talk about adding knowledge base, just to get really tactical, do you do like MCP integrations? Are you talking about like some connectors or is it typically you're uploading specific example documents into the custom GPT? Mick (06:21) Yeah. Just uploading it to this point, we just started to kind of toy around with MCPs and we're starting to build those out a little bit more. I am not an engineer and I always make that clear that I don't know how to, I'm not a developer and I don't speak that language. So a lot of that's getting more into the engineering realm. So we actually made what we call Ninja teams, which are just small teams within the company that ⁓ work. So we've got an engineer, we've got a PM and then we have myself and somebody in support. building those to have literal connectors. Now, but to this point for this conversation, I'm not using anything that technical. Benjamin Ard (07:08) Very cool. I love that. for you in your experience and for anyone listening, I think they're going to ask the question of when should it just be a prompt and when should it be a GPT? How do you know that you should actually make a GPT instead of just prompting? Mick (07:19) Mm-hmm. When I start doing the same chats over and over, that's a definite one. That's a clear answer. And anytime I found myself prompting, I have more than one prompt. It's kind of the same topic, if you will. And I mean like three times. If I've done it three times, then I have a need to make a GPT, especially for internal stuff. I've made a lot of tools. For example, A-B testing. Whenever we A-B test, don't think anyone really knows how to do a correct sample size. And that will completely pollute your data if you don't have the right sample size, right? So I made an A-B sample sizer. And it asks for certain criteria in the beginning, and it will tell you that you need to get it in front of this many people for this length of time. So that's just another example of an internal tool that I knew that we needed. Same with the watch this for me. Like I knew all of us had all these videos we need to watch. just don't have the time to do it. Benjamin Ard (08:13) I love that. Yeah. No, I love that. Like I remember taking a statistics class in college and in most of the stops in my career, I was teaching people here's what the law of large numbers looks like. Here's how you actually know it's statistically significant. And like, you know, like a lot of people are like, oh, if we do like X amount, we should be good. It's like, there's an actual way to know this. Let's, let's do the math real quick. So for everyone listening, you have some great Mick (08:13) Ha Yeah. Yeah. Benjamin Ard (08:41) custom GPTs. Do you mind sharing a few of what those are, some examples to kind get the wheels turning for anyone listening? Mick (08:47) Yeah, I'll give you a couple and I'll even give you a couple that I'm actually got in the lab right now that I haven't actually released if that's cool with you. Okay, cool. yeah, again, it's all about saving time. We're a very small team. So anything we do repetitively, we do a lot of, I'll make a new one. So another one that we use daily is our email spam checker. And basically what that does is ensure us that any type of email copy that we send out does not have spam words. Benjamin Ard (08:54) Yeah, absolutely. Mick (09:13) that spam words that might trigger that filter to shoot you to spam or even to the promotions folder. There is some logic that's changing around that, I understand, but still I think it's important that emails are clean. So we have over 200 workflow automations for emailing that we have going over 700 emails. So keeping our email health is important. So basically what you do is you take, write whatever you think your email should say. and then you upload your subject line in your email copy and it will tell you this word is a potential spam trigger. Try these other words and then it will actually offer to rewrite the entire email for you. So that's one. Another one that we've been using a lot is ⁓ it's called the AEO optimizer and it is for optimizing our product pages, our landing pages to be surfaced in AI citations and different AI models, LLMs for. Hey, you should look at power. There's some specific criteria to rank for an AI citation. So that AO optimizer, you basically extract the page source code, upload that, and it will tell you you're missing FAQ schema. You could possibly benefit from a shorter meta description or whatever the case might be. It'll give you optimization tips on how to get that page to rank for an AO citation. that's another couple that I'm working on currently. I'm working on one to build LinkedIn carousels. that will give you the LinkedIn copy as well as the copy for the screenshots, the slides that you put into your slideshow. And then another one that I'm kind of toying around with a little bit now is a Google Analytics 4 Assistant, essentially. I don't know how much you use GA4. It's a nightmare. It's a black hole in my opinion. Benjamin Ard (10:47) ⁓ I'm already like, you know, I'm begging you for the URL for this one. Like, yeah. Now I think every marketer wants that one. Mick (10:54) Let me tell you, it has been a good example is that I could not find a way to know, am I being cited in ⁓ AI tools? Is ChatGPT recommending us? So I do what I always do and I start with ChatGPT and it wrote an entire funnel exploration that can show me every single LLM, I'm talking DeepSeq and Mistral and some that most people probably haven't heard of, I know if they're citing us. And then I even went a layer further and I can tie conversions to it now. So that entire funnel exploration I built with ChatGPT. So that's one that I still haven't fully fleshed because there's a lot that goes into making one, especially one that robust. But the goal is to say, and Gemini is trash inside Google Analytics for it does not work. My goal is to say I want to see X from X and with X result. Benjamin Ard (11:40) Ha ha ha ha. Mick (11:45) and it will give you exactly how to build the report, how to build the exploration, what filters, dimensions, metrics. ⁓ So that one's going to take a minute, but I think that one might actually be productized if I get that right. I think I could sell something like that actually. Benjamin Ard (12:00) Very true, yeah. I love that, that's amazing. So are any of these public GPTs that people can actually download and use, or do you typically keep them just for the team at Power? Mick (12:12) That actually came up recently because Power is a small business, right? We are still a small business. We're not three people, but we're still a small business. our CEO was like, hey, Mick, if you're using these tools for marketing, I'm sure other small businesses could use them for marketing too. So right now I think there are seven of them that are published and we recently just got a lovable subscription. So I actually made a landing page that has them all in it. ⁓ Benjamin Ard (12:38) Very cool. Mick (12:38) so you can get to all of the custom GPTs from that individual landing page. there's, I think there's seven or eight that are public right now. Benjamin Ard (12:45) love it. So for anyone listening, what we will do, I will coordinate with Mick and we will get the URL and put it in the show notes below so that if you're listening to this and you want to check out any of these or use any of this kind of stuff, just scroll down, click on the link and you'll be able to find all those really easily. I love that. Mick (13:01) Yeah, and I think it's worth mentioning a lot of people think GPTs cost. All of the GPTs in the GPT store are free. So just putting that out there. Benjamin Ard (13:09) Yeah, I love that. So Mick, we're almost out of time. So one final question for anyone listening to this and they're like, okay, I need to build some custom GPTs. I need to do the work. How do they actually get started? Like, do you have an idea of like, you obviously shared some examples of where you got started, but is there like a common problem that you've seen where you're like, this is a perfect use case for a chat GPT, go make this and maybe a couple steps and how they can go ahead and build it. Mick (13:36) Yeah, so first you do have to have a pro subscription to ChatGPT to make your own GPTs. So I think that's important. But the way I've looked at it in the beginning and even now is what do you do repetitively and that takes a lot of time to do. So if you spend a lot of time on writing landing pages for your website, then let's make a landing page builder GPT and start from there. So I would say anything that you have to do over and over again and anything that takes a lot of time out of your day to do that. Make a custom GPT, simply explain what it is that your problem is and how much time it's taking, and then convert it. Say, want to convert this to a custom GPT, and it will help you step by step make your own. Benjamin Ard (14:17) I that. So I'm going to throw in two cents here. This is going to be kind of meta. If you are making a custom GPT and you're getting outputs and they aren't quite right, copy and paste the output, open up another GPT window and say, here's the output I got from a GPT. Here's what I'm going for instead. How do I fix my prompts or my training in the GPT? Tell me and help me fix it. It's one of the weirdest things where you can ask GBT everything. Mick (14:40) He ⁓ Benjamin Ard (14:42) And it will guide you and help you get closer to your answer. So there's one little tidbit that I've seen to be really helpful. Mick (14:48) That's really cool actually, I've never thought about that. That would work. Benjamin Ard (14:52) Yeah, you pit it against itself. works pretty well. It's a lot of fun and it's interesting. With that, you do often have to ask JET GPT to be a little critical because it likes to stroke your ego sometimes. And so it is helpful for it to be overly honest. So I love that. Mick, this has been amazing. I am excited for everyone listening to have the wheels turning and actually be thinking about some of these custom GPTs they can build. Mick (15:01) Mm-hmm. It's Yeah, that's really cool. Benjamin Ard (15:18) For anyone online who would love to connect with you, reach out and find you online, how and where can they find you? Mick (15:23) I'm probably most active on LinkedIn. It's at Mick Essex. All the other socials, it's at Mick underscore NLA. I'm on all of them, of course. So everywhere other than than LinkedIn, that's it's at Mick underscore NLA. Benjamin Ard (15:37) Perfect, I love it. So anyone listening again will have the landing page in the show notes and we will link directly to Mick's LinkedIn profile so you can click there and connect with him there. Mick, this has been amazing. I really, really do appreciate all the insights. Thanks for coming on the show today. Mick (15:51) Yes sir, it was a pleasure.