Loading summary
Transcript143 lines
- [00:01]
Michael Stelzner
Hey, before we start today's show, if you want to accelerate your AI learning, I have a solution for you. Become a member of our AI Business Society. You'll join me as we go deep with live AI training each and every month. Imagine crafting more persuasive content, creating stunning images and automating those time consuming tasks. It's all possible when you join the AI Business Society. Go to socialmediaexaminer.com AI and join today.
- [00:41]
Audra Carpenter
Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner.
- [00:53]
Michael Stelzner
Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators and business owners who want to know how to put AI to work. Today I'm going to be joined by Audra Carpenter and we're going to talk about how to train an AI model to create exceptionally high quality content over and over and over again. As we get deeper into this show, we're also going to talk about some really cool tools that we've never talked about before that I think you're going to find really, really interesting. If you're new to this podcast, be sure to follow this show. We've got some great content coming your way. Let's transition over to this week's interview.
- [01:38]
Audra Carpenter
With Audra Carpenter, helping you simplify your AI journey. Here is this week's expert guide.
- [01:48]
Michael Stelzner
Today, I am very excited to be joined by Audra Carpenter. If you don't know who Audra is, she is a marketing strategist AI trainer who's the founder of Atech Dynamics, a company that helps businesses scale with AI marketing workflows. She's also the founder of Content Hub os, an app that helps you scale your content. And she's host of the AI with Audra podcast. Audra, welcome to the show. How you doing today?
- [02:13]
Audra Carpenter
Oh, thanks so much for having me, Mike. I am so excited about this conversation.
- [02:18]
Michael Stelzner
I'm super stoked that you're here. Today, Audra and I are going to explore how to scale your content creation with AI. But before we go there, tell me your story. Like, how'd you get into AI?
- [02:26]
Audra Carpenter
I guess like all of us that were doing some kind of service based business when AI came about, that it was a very natural progression into, okay, what is the next technology that we're going to bring in? I started a marketing agency in 2009, so I'm not new to the industry. Anyway, I've been waiting for something like this for a long time. Finally able to get better systems built for clients, better outputs, better consistency, and really start building ecosystems with the marketing instead of everything being so siloed. And AI is just for me, it's a golden ticket of what I can actually do now and bring some of these things that we've struggled with as marketers and really just take those systems that we've got that are doing well and make them do really great.
- [03:21]
Michael Stelzner
Share a little bit about your journey into it. Like when did you realize that AI was going to be something important for you?
- [03:26]
Audra Carpenter
I would say when Jasper came along. So I got in in the early days. It used to be started as conversion AI and then it went into Jarvis.
- [03:34]
Michael Stelzner
Oh yeah.
- [03:35]
Audra Carpenter
And then it went into Jasper. So writing back then, those guys were super early to the market, but they were really able to start carving out what this could look like for us. And by being able to do that, me as a strategist, people that providing these kind of services, we saw the co things that we're going to be able to do, so really dug into Jasper and what it could do, started building out frameworks and like I said, it was early, so we were still very limited as to the results we could get. Early on it was really more work than benefit, but those of us that were patient knew that it would continue. You know, once it was out of the box, there was no going back. And, you know, as OpenAI came, things have gotten better and better. And today we are so fortunate. I'm excited, excited about all the small businesses that can actually finally get some systems down and get work done and grow products and services that maybe they couldn't do before because of a bandwidth issue. So it's been cool.
- [04:39]
Michael Stelzner
Love it. So tell us a little bit about the actual services that you provide today.
- [04:43]
Audra Carpenter
So the marketing services haven't changed that much. Marketing is still marketing, right? We're just using different kinds of tools to do it. Building funnels, building websites. We're still leaning into SEO a little bit, even though we know that's going to change in the next year or so. Social media is still a big thing and I think if I had any takeaways of this stage of it where I'm at, what I would suggest to business owners is to really lean into building that audience because now we all have access to the same knowledge and the same resources. Now not everybody knows how to use them yet. So that does give us experts that have been around a long time a Little bit of an advantage, but there's still a lot that they can do that also helps them build in some systems that they didn't have before. So a lot of system building, a lot of workflows, taking tests that they were already doing and then finding ways. Can we automate this? Does it make sense? Does it save money, generate more leads or sales? And that's really where the sweet spot is right now.
- [05:46]
Michael Stelzner
Perfect. Okay, so for marketers listening right now and creators, why should they lean on artificial intelligence to help them create content? Because that's kind of what we're here to talk about today. What's your message to those that maybe haven't fully embraced AI from a content creation perspective?
- [06:04]
Audra Carpenter
So if we look at marketing, kind of like a bicycle tire, we have our product or service in the center and each one of the spokes is a different channel. Right. Marketing isn't just social media, it isn't just paid ads. There's a lot of different moving parts when it comes to doing that effectively. Where I would say is you start with where you're already winning. You know what that system looks like, you know, how it works, why it works, how it's supposed to work. That would be the best one I would say is, okay, this thing that's currently generating me money, leads, or opportunity, can any part of that be automated? And then I would, you know, do some research on that, find out if it can, if it can't, keep doing what you're doing, and then look at some of the other systems in your company or tasks within your company and say, can they be offloaded into something that's more manageable? That maybe doesn't free me up because I'm the lead gen person, but it frees up my assistant so she can work on higher skill tasks that I really need to get done and she doesn't get done, or something like that.
- [07:10]
Michael Stelzner
Okay, so we're here to talk about how to really scale our content creation process.
- [07:17]
Audra Carpenter
Yes.
- [07:17]
Michael Stelzner
So where do we begin?
- [07:19]
Audra Carpenter
Do you want me to go into the little experiment I'm running, or do you want me to talk a little bit more into the content?
- [07:24]
Michael Stelzner
Well, tell me more about the experiment that you're running.
- [07:26]
Audra Carpenter
Okay, so owning an agency, as I mentioned earlier, from 2009, I've always played the role of second in command. So I'm the person creating the plans, developing the roadmaps, building out the teams, strategizing and getting all the pieces moving. And then the owner typically will show up and say, look at all these millions of Dollars we made. I have actually not spent a lot of time building a personal brand. I have never ran an ad for any marketing in my company or for me as an individ. I've never boosted content for myself. Now mind you, I've done this for hundreds of clients but for me personally, I've never done it. And so I've gotten to a place where AI where I'm recognizing that as we evolve in the next couple years, it's not going to be so much about our services, it's not going to be so much about our products as it is going to be about the audience that we serve and the people that we've built in our little tribes or community. I don't have one of those. So I thought, okay, I've done this for hundreds of clients, I need to probably take a few minutes and start building this out for myself as a what, 16 year, 17 year veteran. So Chatgpt and I spent some time together doing some brainstorming as to the best way to approach this and I decided to kick off a challenge. Now mind you, all my social media together, I'm probably 20, 25,000 again, don't really promote, don't really do a lot of kind of stuff. And in all transparency, I don't think I really found my lane there. I felt like I couldn't go out there and just be another Me too. And I know a lot of business owners feel the same way. I'm not really on video, I don't really care for that. That's not who I am. I just want to sell my widget right? So it's a little bit tougher to get a plan. Regardless of what we know, what we need to do and what we actually do are two different things. So I decided I was going to document where I'm at and try to build everything in public and be completely transparent how AI can lean in and help me do this, maybe remove the emotion or the head garbage and see if I could just let AI drive a little bit tougher when you're the one that typically drives. But I decided to do that. So about four weeks ago now, I kicked off a plan to see if I could move my followers from the 25,000 to 50,000 in 90 days. And I decided to start with LinkedIn. I felt that business wise and the AI topic and marketing and that kind of expertise would make the most sense starting with that platform. So put together a little plan. Now if anybody knows anything about me, they know I don't take any Action without actually having a strategy. I mean, that's just kind of a staple piece of life for me. I need to have a plan. So I start creating a social media plan and I really don't know how to write for LinkedIn, not in my voice, in for my content, for my audience. So we re engineer it the other way around. I said, okay, so what does an effective post look like? What is getting engagement? What is moving the needle? What are the algorithms doing? How do we create the best kind of content and spend some time just really mapping out what a good LinkedIn strategy looks like? How does AI fit in? So I go to Claude this time and I say, what would I need to create a Training database for LinkedIn? We map out all the steps in a spreadsheet.
- [11:19]
Michael Stelzner
By the way, what do you mean by training databases? So people understand that.
- [11:22]
Audra Carpenter
Oh, that's, that's a very good point. So when you want AI, any of the large language models to write content in a specific way, it's better if you provide them examples. So if you think about the chat models, they're just a big box of information and they're going to do their best to provide you the outcome that you're asking. But if you give them a little bit of help of what an example looks like, they're gonna not have to look at that whole box and they're gonna say, oh, okay, she's trying to get me over here. Let's create some really sexy stuff around this. So that works well.
- [11:56]
Michael Stelzner
Okay, so let me just summarize where we're at so far. So here's what I've heard you say, and just to kind of like bring a little structure to where we're at, folks. We're talking about where do we start? And we've been using Audra's story as an example, but what she said is like, first you gotta come up with a plan, right? Because not everybody, she does it by her very nature, where she's strategic and she wants to have a plan. Just out of curiosity, on the plan, what kind of things did you have inside your plan just so people maybe when they're planning, because obviously this is all going to tie together for AI, but can you kind of give advice on what kinds of things someone should have in a plan if they're going to use AI to create content?
- [12:34]
Audra Carpenter
Sure. So I start with the goal. What am I trying to achieve? How am I going to measure it? I mean, it really goes back to our simple SWAT stuff that we were taught, right? But what is my goal, how am I going to measure and then what vehicle or tool am I going to use to try to achieve that? All those variables can change. If I'm tracking revenue, if I'm tracking likes, if I'm tracking product sales or content creation, all of that would evolve based on what am I trying to achieve with this action.
- [13:03]
Michael Stelzner
Got it. And then you talked about your target audience. Right. And this is really important for the non marketers that are listening. Maybe creators don't understand that, but how descriptive are you developing your target audience so that when you feed that into AI, you've got that level of detail.
- [13:18]
Audra Carpenter
I actually went the other way around. So I go a little bit different when it comes to the whole prompting thing. And the reason why I say that is I'm getting very good content being conversational versus having to go the prompting route. And the reason I'm finding that I'm more effective doing that is because I'm an expert in what I do. So it's a very give and take conversation that I'm able to have. I think when you're out of your realm of expertise, you probably need to lean a little bit bit more into an effective prompt. But if you're a small business owner and you don't know marketing but you know your industry, talk to it. I'm trying to get to this. This is normally who buys my product. This is normally what they do. These are the questions they normally ask. It will help you get there. We got to get over that. We think we have to be the smartest person with the answer when we go to these things. They've indexed the Internet. You're not catching up. Let it go.
- [14:18]
Michael Stelzner
I love it. Okay, co Basically, for those of you that are thought leaders, you kind of know like Audra knows your stuff and you can kind of just converse with the AI. Totally got that. Okay, so this is going to be like the foundational stuff that we're talking about. The next phase really is. It sounds like the, the research phase that you and I talked about when we were prepping, when you were talking about this training database a few minutes ago. Why don't you talk about why that's so important and maybe even like where we get context or examples to train the AI model well.
- [14:51]
Audra Carpenter
So if we think about it this way, our mode of operanda, mode of operation was before this. We would create social media blog posts, content for any of our marketing channels based on our experience, not based on data. For the Most of us, 9 out of 10 of my clients over the last 15 years. Do not want to read the data. Right. There's, there's very few. So the content that you're creating is not based on that. So as I'm thinking through the research part of this, I thought, okay, wait a second, clearly, let's come from the other angle. Because AI has access to what's working. So I actually went to the AI and I said, look, I know you're an expert marketing strategist on LinkedIn. You know what a successful post looks like, you know what moves the algorithm, and I actually built it the other way around.
- [15:47]
Michael Stelzner
I love this.
- [15:48]
Audra Carpenter
Which normally we don't know because I didn't have the data.
- [15:51]
Michael Stelzner
Got it. So explain why it's so important to those listening to make such declarative statements to the AI, like, you are a blank. Who understands blank. Like, why is that so important? Because not everyone in my audience understands why that's so important.
- [16:05]
Audra Carpenter
I would say the biggest thing is because I'm trying to put them in the right lane. Remember, they indexed the Internet. I need them to lean into that little tiny expertise of being a marketing strategist on LinkedIn. I've niched them down into their expertise so they can say, okay, wait a second, we only want to talk about this.
- [16:26]
Michael Stelzner
Got it.
- [16:26]
Audra Carpenter
I know a lot of things, but she only wants me to focus on this.
- [16:29]
Michael Stelzner
Okay, cool. So keep going with where you were going with that train of thought. You are an expert in creating LinkedIn content, maybe written posts, and you understand the LinkedIn algorithm, I believe is what I heard you say. So keep going.
- [16:44]
Audra Carpenter
I would even back up just a second. So how do I measure success on LinkedIn? Well, if I go back to my data, if you are successful on LinkedIn, you're probably measuring three things. Follower growth. Again, we're just at the beginning. I'm not trying to sell anything. My challenge was to grow my following. So my follower growth, my engagement and my comments, that's it. That's all I want to measure right now. Because from engagement and comments will also come the growth.
- [17:12]
Michael Stelzner
Got it. And when you say engagement, you mean likes and shares. Is that what you mean? Because obviously comments are sometimes perceived to be engagement. And what do you do with that? Are you telling the AI that these are the metrics you care about?
- [17:23]
Audra Carpenter
Yeah.
- [17:24]
Michael Stelzner
Okay, good.
- [17:25]
Audra Carpenter
100%.
- [17:26]
Michael Stelzner
So keep going.
- [17:27]
Audra Carpenter
Okay, so then once we've decided I have my goals, I have a timeline. Right. Which is another one that's important. Is it going on forever? Let's do a 30 day sprint. Let's do a 60 day sprint. 30 days is long enough to know if it's working or not. And then if you need to reassess and create a, to, you know, evolve your strategy. So I would say set 30, 60 day goals. That's it. Then we said, okay, how are we measuring it with the followers, likes and comments, engagement type stuff. And then originally I was going to start on all my social media. What I found was it was too much. Again, I'm trying to prove this out, me doing it by myself with none of my staff or none of my other resources, as if I was a normal person and didn't have access to the things I have access to.
- [18:16]
Michael Stelzner
Perfect.
- [18:16]
Audra Carpenter
Because I can't tell people to do what I'm doing if there's too far of a gap between us.
- [18:21]
Michael Stelzner
Okay, cool. So in this research phase when we were prepping, you were talking about how you, I believe you told me that you had identified some really great content creators to model. Right. Talk to us a little bit about that.
- [18:33]
Audra Carpenter
Okay. So what Claude suggested I do is to find some creators that are creating great content that's working on LinkedIn. So I spent probably four, eight hours going through different kinds of profiles on there and I didn't pick marketers. I don't need to learn how to talk like a marketer. I want to know what's working on the platform. Remember, I'm second compared to what I'm trying to achieve using AI. If you build it based on you, the model's not reproducible. You need to build it from AI and then that model can be taken and applied to anybody. So remember, I got my, I mean I still have my marketing hat on and I'm looking at scale and everything else, but if we build it the other way around, then you'll get much better results and not as bias.
- [19:26]
Michael Stelzner
So where did you find these people? And like what were you looking for really?
- [19:29]
Audra Carpenter
I just started random. So I would just spend a couple hours on there, surfing, clicking, going into different industries. I've got about 10 different industries. My database has about a hundred different posts on it. And I'll go into kind of what I mapped out on the spreadsheet itself. But I have fintech on there. I have coaches, I have other service providers, I have venture capitalists. I do have another AI person on there. But again, I tried to pick people outside of my industry. I don't want to be another. Me too. I mean, we have marketers, we're some of the most competitive people out there. You don't want to produce content just like everybody else. So you always want to look outside of your industry for just a little different positioning. Marketing's marketing, right? A funnel's a funnel. But how can you position it so you can stand out a little bit with all the noise that's out there?
- [20:27]
Michael Stelzner
When you were looking in all these different niches for content creators, were you looking for samples of content that seemed to outperform based on like, let's say they had had 10,000 followers, but this post got way more comments and shares. So obviously there was something going on there, like help people understand once you found someone that you thought was creating great content, how you decided which pieces of their content were going to be the ones that you were going to put in your spreadsheet.
- [20:53]
Audra Carpenter
So a couple things that I ran into. One, I built this example for a buddy of mine. His training data. He went out and he got people like Gary Vaynerchuk or Richard Branson. And those kind of thought leaders don't work because they could post crap.
- [21:13]
Michael Stelzner
Yeah, because they have a huge following.
- [21:15]
Audra Carpenter
Right, because they have a huge following already. Correct. So you need like the B players or the high C plus players that are really creating some solid content, have a decent following and are consistent in the kind of content that they're posting. So I looked at things like there was nobody under, I think 40,000 followers and they were up to about 300,000 followers. And what I would do is say I found a really good one in the venture capital space. I would go look at who's commenting or who he is following and then I would check out a couple of their people. That person would lead me to the next person. That one would send me over here. And I really was just connecting the dots. Like I said, it was more of a research exploratory and then also the content that jived with me. Right. You do spend some time, but the quality of the data that you're going to collect, it's fire. Nobody can touch it. It's so good.
- [22:19]
Michael Stelzner
Are your competitors already using AI to create better content and faster? Are you struggling to keep up with the pace of the ever changing AI landscape? If so, Social Media Marketing World 2025 is your solution. Our dedicated AI track brings together the brightest minds in AI marketing to help you master the latest and greatest AI tools. You'll learn from experts like Matt Wolf who will share 25 AI tools to transform your content creation forever. Chris Penn will show you how to use AI as your second brain for analyzing data and building winning Strategies Molly Mahoney will reveal how to quickly create months of high converting content with AI. As one attendee, Melanie Miller said, quote, the AI teaching was, was mind blowing. I realized had I not gone there would be a good chance my business would have been obsolete. I'm so far ahead of so many people because Michael Stelzner brings only the best to teach. Thank you, Melanie. The most affordable way to attend Social media Marketing World 2025 is with our dedicated AI ticket, which gets you the AI track and access to all of the keynotes in San Diego this March. Visit social mediaexaminer.com Aicon today and future. Proof your marketing career with our AI ticket. Don't delay. The event starts March 30th. I hope to see you there. Is AI able to assist you at all in any of this part? Or is this kind of, as of today, kind of a little bit of a manual process?
- [23:49]
Audra Carpenter
This portion here to start, it was manual. So I created a Google spreadsheet which I'm happy to share with anybody if they want the kickstart of it. But what I had was the category, the audience, the intentional message. Kind of like the summary. What was the gist of it, the actual post, the link to the creator bio, how many likes, how many comments, and then a link to the image.
- [24:17]
Michael Stelzner
Real quick, on the post, was it a link or was it the copy of the text in like a cell? Do you understand what I mean?
- [24:23]
Audra Carpenter
The copy of the text.
- [24:24]
Michael Stelzner
Okay, that's important to know. Okay.
- [24:25]
Audra Carpenter
Exactly the way it was with emojis, with spacing, the whole thing. All I did was copy, copy and pasted it.
- [24:30]
Michael Stelzner
Okay, cool, keep going. I love it.
- [24:32]
Audra Carpenter
Okay, so. But the only thing I filled out was the actual post, the date, how many likes, how many comments, and then the links to the creators and if there was an image, because then I gave it to Claude because again, she's been. Claude's a girl for me. For some reason, she's been instructed exactly what we're building and what my goals are. Right. So she's got the layup and then I would copy just the post to her. She did the rest of the work. So she told me the category, the target audience, the messaging, blah, blah, blah, blah, blah, and gave me the rest of the data.
- [25:11]
Michael Stelzner
Okay, I love this. So let's just. We're going to get technical here for a second. Yeah, we're using a cloud project.
- [25:16]
Audra Carpenter
I do.
- [25:17]
Michael Stelzner
Okay, so inside of the cloud project, presumably you've trained up this cloud project to kind of understand the objective, right? And you've told it all, all the stuff that we've talked about and you pasted in the actual post, not the spreadsheet. Is that what I'm hearing you say?
- [25:32]
Audra Carpenter
Correct. Because we needed to fill out the spreadsheet first.
- [25:35]
Michael Stelzner
Yeah. So you've essentially in this cloud project have said, hey, you probably, you're really good at analyzing LinkedIn posts and discerning these things. Right. Like what's the topic that they're talking about? What's the blah, blah, blah, blah.
- [25:46]
Audra Carpenter
Audience and Right, category.
- [25:48]
Michael Stelzner
Audience and category. So then you're putting that back into your spreadsheet and then how many rows of data are you collecting here before you're kind of.
- [25:56]
Audra Carpenter
I have a hundred.
- [25:56]
Michael Stelzner
A hundred. Okay. Do you recommend everybody have that much?
- [25:59]
Audra Carpenter
Claude recommended 150 and I told her sister, I'm done. See what you could do with this.
- [26:06]
Michael Stelzner
Okay, so once you've got all this data in a spreadsheet and Claude is helping you populate the spreadsheet.
- [26:13]
Audra Carpenter
Yes.
- [26:13]
Michael Stelzner
You've got this really big Google sheet. Now what are you doing with this Google sheet? Like what's the next part of the process?
- [26:20]
Audra Carpenter
Okay, so the first expectation here was I was actually going to build an OpenAI training trained fine tuned model, a custom GPT basically. Right, right. But on the API side.
- [26:34]
Michael Stelzner
Okay, got it. Okay.
- [26:35]
Audra Carpenter
Yeah. So I get all this data hours, you know, by this point I probably got two days of work in here. Hours of work. I go to upload it to Chat GPT and it won't take it and it won't take it and it won't take it. And I'm like, lord help me. So had to do some more back and forth with ChatGPT to figure out why. Well, note to self, note to you guys, when you build your own fine tuned model, again, I would say it's a step or two above a GPT. It's really on the developer side and you get a lot more control there. They only want three fields. The user field, the messaging and the system.
- [27:18]
Michael Stelzner
System prompt. Yeah.
- [27:19]
Audra Carpenter
So it would not take all my data.
- [27:22]
Michael Stelzner
Oh, fascinating.
- [27:24]
Audra Carpenter
So I had to. I created another clone, the spreadsheet, removed all the rest of the stuff and just showed those three categories and then it uploaded and it was able to take it. Here's a challenge though. I don't have as much control over it. The output, not sexy. It just was not. It was. Even with all that data, the output was still as if I just went to the retail window and typed it in. I was not impressed.
- [27:54]
Michael Stelzner
So what did you do? Did you end up going, creating another cloud project and uploading the data there or what did you do?
- [27:59]
Audra Carpenter
I kept it in the project. Okay, so now I just go in there and I say, okay, today we're going to write a LinkedIn post about AI automation and customer service. Please refer to the training data.
- [28:13]
Michael Stelzner
Okay, so just so we're clear, because I think we might have skipped this step. Okay, you did at one point take your Google Sheet.
- [28:20]
Audra Carpenter
Yes.
- [28:21]
Michael Stelzner
Did you connect it to the project or did you upload a CSV file?
- [28:25]
Audra Carpenter
I actually just downloaded a PDF and uploaded it to Claude. Now it did take up probably 60% of the memory.
- [28:36]
Michael Stelzner
Got it.
- [28:36]
Audra Carpenter
But yeah, so it's all attached there now.
- [28:39]
Michael Stelzner
Just because I've done this before. There are other ways you could do this potentially, right? Like you can make each article potentially into a PDF, but the problem is that it's not going to necessarily have the categorization and all the stuff. I've taken huge CSV files that had like customer testimonials in one column and then another column it had the name of the customer. So I know that you probably could, with Claude, export this as a massive CSV file, but did you ask Claude to do a little reverse engineering on what makes a good article given all this data, or did it just know? Do you understand? Like you have hundreds of articles now on the spreadsheet, right?
- [29:20]
Audra Carpenter
A post. Yes, yes, yes.
- [29:21]
Michael Stelzner
Yeah, hundreds of posts. I'm sorry I call them articles, but hundreds of posts inside this spreadsheet. Did you ask Claude to like analyze it and look for common threads or did it just intuitively know what to do with all this data?
- [29:33]
Audra Carpenter
No, it knows what to do. And here's the even better thing about it, because we categorize them. So I may have, out of those hundred, I could have 10 of those or how to post. Ten of them are mindset posts. I mean, I don't have that many. So I can go in and say I want a how to post look at the training data. Here's what I also noticed that I, I don't think I've ever paid attention to. Me personally going through and reviewing all that data. I saw the frameworks and the structures of all these posts are, and I have to tell you, they're nothing. Mind blowing. The content is not. Wow, I read that and my life has changed. They're leaning into what the algorithm is rewarding. They're leaning into what is a successful post. It works right now, three months from now, LinkedIn may start rewarding something else. What I found was I tried to use that same workflow that I'm currently doing with LinkedIn on Facebook. It's not the same and it doesn't get the same results from the growth that I'm seeing on LinkedIn, because my growth has been stupid.
- [30:45]
Michael Stelzner
You could apply this very same methodology to any platform though, right?
- [30:48]
Audra Carpenter
Correct. I'm already building Twitter. I'm already. Correct.
- [30:52]
Michael Stelzner
Yeah. So you could do this on Facebook, you could do this on X. I really like that. Okay, so so far, let's just summarize where we're at. Okay. We started with kind of a goal or objective of where we want to go. And in your case, it was doubling your. Your number of your audience on LinkedIn. And you had already in mind what your objectives were and what you wanted to measure. You went out and you did a lot of research on your own to kind of find high quality data to build this database, this, this training database, as you've referred to it. Right. And you've used AI to help essentially analyze each of these pieces of content, to provide labels for lack of better words on each of these types of content. And then you pulled it all together, uploaded it, in this case into Claude. And what did you do once. I think there's a step I want to ask about, which is how did you now ask Claude to create content? And how did you make it sound like you.
- [31:48]
Audra Carpenter
Okay. I do have brand voice guidelines.
- [31:52]
Michael Stelzner
Okay.
- [31:53]
Audra Carpenter
So I just uploaded that with the training data. I also went a little bit further. So there's one thing about getting your brand voice, which I know a lot of business owners run into, because they don't publish a lot of content. Somebody else is typically writing the blog post or posting social media from the brand voice, meaning the company voice, not the individual CEO or CFO or CMO even, for that matter. But brand voice, I think, only gets you so far. What I also did was I went in and I wrote a document about my experience, my professional experience, things I like, things I don't like, things about my personality. And then I've also got this document that kind of leans into the human side of me. And you know what is important to me? It's one thing to make it sound like me, it's another thing to actually know me and share my stories and share my journey and my expertise. And that has been the winning sauce to kind of bring all of this together. If you go to my blog and read my content and I just relaunched it, that was part of this rolling out the Brand thing. It has found my voice. It sounds like me. Anybody that personally knows me would have a very hard time saying, I didn't write that.
- [33:11]
Michael Stelzner
Okay, so any tips on how to collect this data for people that don't really know how to do this? Have you learned anything in the process of making this. These documents that are kind of of all about you and your stories and your life?
- [33:22]
Audra Carpenter
I would say the easiest thing would be, is pop out your phone and talk to somebody about it. Have somebody sit with you, a mother, a friend, a coworker, just somebody that can interview you, essentially.
- [33:34]
Michael Stelzner
You could ask AI to do it, couldn't you? You could use ChatGPT, advanced voice mode. Right. And you could give it a prompt that says, I want you to blank interview me to what are some of the things we should ask it to.
- [33:47]
Audra Carpenter
Do to help me tell my story?
- [33:49]
Michael Stelzner
Yes. And ask me as many questions as you can about my work and about my personal life and the things that I'm working on. And then when you're all done, you could copy and paste that, couldn't you, right into the models. Right.
- [34:01]
Audra Carpenter
And I would say to make it a little bit more organic, use the voice mode.
- [34:05]
Michael Stelzner
Yes.
- [34:05]
Audra Carpenter
Do it on your phone if you don't have a microphone, but use a voice mode because it feels a little bit more conversational and you're not going to feel like. Like trying to type it out like a letter versus how you actually talk.
- [34:18]
Michael Stelzner
Yeah. The advanced voice mode, what's great is it will transcribe everything it asks you and everything it tells you. Okay, so we're now at the point where we're inside of our LLM and let's say in this case it's Claude. And let's say you want to write a post. Like, what are you asking it to do? What kind of insights are you providing in order to write a post about whatever is in your mind?
- [34:35]
Audra Carpenter
It's like five minutes, if that. It's like two minutes. So I typically log in, hey, sister, today we're going to write a post on xyz.
- [34:43]
Michael Stelzner
So you could write a post about you appearing on Michael Stelzner's podcast.
- [34:47]
Audra Carpenter
That's it? That's all I have to do?
- [34:49]
Michael Stelzner
That's all you say? That's it?
- [34:50]
Audra Carpenter
Yeah.
- [34:51]
Michael Stelzner
Okay, cool.
- [34:53]
Audra Carpenter
Now, there's a couple other variables I haven't shared with you yet.
- [34:56]
Michael Stelzner
Please go there. Yeah.
- [34:57]
Audra Carpenter
Okay. So besides doing this stuff, which all of us can do, there's no excuse. If you need help, reach out. I'm happy to fill in any gaps that you feel like you're missing. I'm already in the automation, building these baby agents and leaning into that quite a bit. So I've taken this workflow and built it into besides Claude and besides using it in OpenAI, I've built in an automation in N8N, which is an automation platform, little technical, so I wouldn't say suggest a start there, just n the.
- [35:29]
Michael Stelzner
Number 8 and then n. Is that what you said?
- [35:31]
Audra Carpenter
Correct. A developer tool, kind of like zapier or make.com but even a little bit geekier. And then also I'm using a tool called Relevance AI which is very good for building agents. And inside that agent I've built a fine tuned model. So I also took my spreadsheet and uploaded it there into Relevance AI. And then I built in a couple of workflows. So I mentioned earlier I got my 10 creators. Well, they're still posting content, I don't want to go scrape it. So I built a LinkedIn scraper that follows those 10 people, takes what they've recently posted, brings it back and loads it into my database. So after I built it the first time, I don't have to build it again.
- [36:21]
Michael Stelzner
Okay, so a couple clarifying questions. So okay, you had 10 creators and, and presumably you had 10 pieces of content for them. Is that how you got to the hundred?
- [36:29]
Audra Carpenter
No, I mixed it up. But yeah, some could have five, some could have eight, whatever.
- [36:34]
Michael Stelzner
Talk to us a little bit about Relevance AI a little bit more because I think that's kind of the cool. Because I don't think I've ever addressed that on the show.
- [36:41]
Audra Carpenter
Okay, so Relevance AI is a retail model, so they do the heavy lifting in the background and what we do is think of a workflow that you want to get done. So if we go back to this LinkedIn thing, I need to do a couple different things I needed to update because those creators are still creating new content. So I need some kind of tool that's going to go out, scrape LinkedIn and bring it back, put in the database. So now I have access to it. What else do I need it to do? Well, I want it to write the LinkedIn post. So besides Claude, I can actually have relevance AI write the post, schedule the post and then actually send it. So think of just a workflow. As you're building workflows on anything, regardless what tool you're using, think through the physical steps of what needs to happen and then it'll make sure that whatever you're building is connected where these workflows break. Is where you've not thought through all the pieces or the actions that need to happen to make that workflow work.
- [37:50]
Michael Stelzner
Work. I'm looking at their site and it looks like they've already got agents or functions for sales, marketing, customer support, research and operations. Does that mean they kind of have pre built, for lack of better words, agents that plug and play? Is that or do you have to build these from scratch?
- [38:03]
Audra Carpenter
They do have some. Okay, the one I'm talking about I got pieced together from a bunch of other people that were are doing. I'm in a ton of AI automation groups.
- [38:14]
Michael Stelzner
Okay.
- [38:15]
Audra Carpenter
So I typically will learn from there. But yes, they have quite a few off the shelf things that at least get you started and start thinking through what agents are. So just to clarify those that are kind of listening to this, we have workflows which you are the driver. You say do task 1, 2, 3, 4. AI assist by following directions. Agents will be the other way around where we say here's the workflow, now you go do it without my instructions. See, now we're getting closer to that, but we're not quite there yet. The output unfortunately is not consistent as operator and these other new tools come out with the large language models. We'll get closer, but we're not quite there yet.
- [39:04]
Michael Stelzner
Okay, so just to help us understand, first of all, the journey we've been on is we built this inside of an LLM, which is mostly CLAUDE at this point. And then what we've done is it sounds like we've got a bunch of different AI agents that are doing different parts of this. So what I heard you say is you've got an AI agent that can go out there and like do the research and keep updating the best practices for the data set. Right. What other agents are you using in this context for the content side of things?
- [39:32]
Audra Carpenter
Oh, just for this one. So in N8N, I've built a. It's kind of an agent, they're calling it one, but to me it's not like a true agent where I've connected Claude. So I use a tool called Open Router. Open Router is a dashboard where you can instead of me getting an API key from OpenAI and from Claude and Perplexity and Gemini using Open Router, they're a dashboard where I can access all of them.
- [40:03]
Michael Stelzner
Okay, cool.
- [40:05]
Audra Carpenter
All I do is use the Open Router API key and then I could pick whatever model I want doing whatever I want inside there. So my research, I have Perplexity do my product descriptions I can have OpenAI do my blog content. I have cloud write. I mean, and I could mix and match whatever I want without having to go in and change those modules individually.
- [40:31]
Michael Stelzner
Talk to us about the other agents, because I could tell that you're using agents in other ways. Like, let's. Let's keep going. Like, what are some other ways you're using agents right now?
- [40:39]
Audra Carpenter
So Mike had mentioned the Content Hub. Now, here's what I want you guys to think about. Everything that gets done, and marketing's my lane, so I got to lean into that. But everything that you're trying to get done, what are the steps? I'm lucky that this is a superpower for me because I am a step person. I can typically map out every single thing we want to write a blog post. What goes into a blog post? Well, I need a title, I need an excerpt. I need internal links, external links, keywords. I mean, the list, right? Start with a list of what goes in to make that one piece of content happen. Then you could say, okay, well, what is our workflow look like now? Map that out. Okay. How does AI fit into this? I think I'm getting such winning success right now is because I'm leaning into the stuff I was already doing instead of trying to figure out how to make AI run that stuff and completely go against what I was already doing that was working. And once you kind of do it that way, then you'll realize it's just a tool. Does it help here? No. Does it win here? Absolutely. Integrate it in. So I've built automations for Slack for. You can record on your phone a voice message. Slack transcribes it, dumps it into Claude. Claude writes a Transcript for my AI with Audra podcast that goes into 11 labs, that creates a podcast episode with my cloned voice that's downloaded into my podcast player. I'm done. I mean, so there's so many different pieces that we can connect together that really will make our lives better and consistency and better results.
- [42:24]
Michael Stelzner
Audra, this has been a really fun exploration, and I know we've just scratched the surface of your wisdom. If people want to connect with you on the socials, what's your preferred platform? And if they want to check out your products or services, where do you want to send them?
- [42:37]
Audra Carpenter
I am just Audra Carpenter on any of the platforms. So whatever your drug of choice is, you can find me there. And then my website's audra carpenter.com.
- [42:47]
Michael Stelzner
And what about this Content Hub thing that you were talking about? If they want to learn more about that.
- [42:51]
Audra Carpenter
Oh, so it's built and it's beautiful. I need AI to get a little bit further where I have some control over the output. So just to kind of wrap up, you know, close the loop on the LinkedIn thing. I've got it to a place where I've had huge growth. My comments are more than I've ever had in the 12 years I've been on LinkedIn. I got invited to be on this podcast, I got invited to speak at two events. And this is just this little tiny thing from the work that I've been doing on LinkedIn. So it's working. So it's worth the effort that it's taking to set it up. The Content Hub hub. I need to be able to consistently get these kind of results no matter what industry they're in. And when you set up an automation, you don't get to change the prompt as you're running through it. So I got to nail that part of it. Then the Content Hub will be able to take one piece of content and do all of the assets. One blog post it'll write. It'll create the email, the social media, the YouTube script, the Reels, post it or schedule it, post it, and then go into a system to redistribute it. It's all built. It's beautiful.
- [44:03]
Michael Stelzner
Love it. Audra Carpenter, thank you so much for coming on the show today.
- [44:06]
Audra Carpenter
Oh, Mike, thanks for having me.
- [44:09]
Michael Stelzner
Hey, if you missed anything, we took all the notes for you over@social mediaexaminer.com a46. Be sure to follow this show on your favorite podcasting app. If you've been a longtime listener, would you let your friends know about this show? We we would love it if you would do that. And also if you'd be willing to give us a review on your favorite app, that would be amazing. And do check out our other shows, the Social Media Marketing Podcast and the Social Media Marketing Talk show. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful.
- [44:46]
Audra Carpenter
The AI Explored Podcast is a production of Social Media Examiner.
- [44:54]
Michael Stelzner
Are you ready to transform Your Marketing with AI at Social Media Marketing World 2025? Our exclusive AI ticket gives you access to cutting edge tools, proven workflows and strategies that drive real business results. Join industry experts like Matt Wolf and Chris Penn this March in San Diego. Visit social mediaexaminer.com Aicon to secure your AI ticket today. The future of marketing is waiting for.