Loading summary
Transcript97 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]
Andy Crestedina
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 use AI. Today I'm going to be joined by Andy Crestedina and we're going to explore how to use AI as a content analyst. If you roll your eyes backwards a little bit when it comes to looking at data and analytics and all that kind of stuff, and you wish you had kind of a data scientist in your pocket, you're going to absolutely love today's episode. By the way, if you're new to this podcast, be sure to follow this show on whatever app you're listening to so you don't miss any of our future content. Let's transition over to this week's interview with Andy Crestadena, helping you simplify your AI journey.
- [01:45]
Andy Crestedina
Here is this week's expert guide.
- [01:49]
Michael Stelzner
Today, I'm very excited to be joined by Andy Crestadena. If you don't know who Andy is, he's the co founder and chief marketing officer at Orbit Media, a digital agency that helps B2B businesses develop and optimize their websites. He's also author of Content Chemistry, the illustrated handbook for content marketing, and he's been a regular on the social media marketing podcast and social media marketing world. So, Andy, welcome back to the show. How you doing today?
- [02:16]
Andy Crestedina
I'm doing great. I've been thinking it's probably 10 years we've known each other, right?
- [02:20]
Michael Stelzner
It's been a long time, man. I'm super excited to have you here on the show. Today, Andy and I are going to explore you using AI to help you better analyze your content and your marketing. Now, before we dig into all the exciting stuff that I know you have to share with us, I'd love to hear a little bit of your story. It's the first time you've been on this show How'd you get into AI? Tell us the backstory. I can't wait to hear it.
- [02:41]
Andy Crestedina
Sure. So let's say 23 years in SEO and analytics, 17 years in content strategy, social, email, influencer marketing, blogging. So for me, right away I could tell, like, this was a trend I couldn't ignore. I. I need to study this. I need to cram, like, it's finals week to figure out all the best methods to get the best results from this, keep my content up to date, but also not miss opportunities to push our results ever higher. So for me, mostly, AI has been an exercise in becoming more thorough and detailed and comprehensive in many of the methods that I've used every day. So lots of the things that I do with AI are sort of like, I've always done that by hand for SEO. Now I've got a partner to help me do it. I've always analyzed these reports by hand from GA4. Now I've got a partner to help me do it. So, yeah, it's been about two years of research, experimentation, writing, recording, teaching AI methods, and now, yeah, at Social Media Marketing World, at the Marketing AI Institutes conference, at Content Marketing World, at Marketing Profs B2B forum. I'm one of those people who are teaching AI side by side with all the methods that we've been using forever.
- [03:45]
Michael Stelzner
What was your wakeup call? I mean, was it quite literally November 30, 2023 to the day it dropped, or how soon after that? Like, when did you realize, okay, this is something I need to pay attention to?
- [03:58]
Andy Crestedina
You know, I'm not that on top of trends. Like, I take it. It passed me by for a couple of months, but then I had people who were sending me stuff and watching videos and the alarms went off and I realized, like, this was one that I can't ignore. This is one that I should be on top of and in front of. So it was probably February when I decided to just double down and just spend a lot of time. It was actually right at the time when GA4 was kind of past that tipping point. I knew I needed to update a bunch of content for GA4. And surprisingly to me, at the time, like, AI became an even much bigger trend than GA4. So that year it was really just about updating tons of content and finding all kinds of new methods and adapting the whole content strategy for both those two tools. But AI came out in front by a mile, and it's been super fun to. To figure out what works well and what does not.
- [04:48]
Michael Stelzner
Yeah, for those of you that don't follow the marketing community. Google Analytics 4 obviously was a really big deal when they rolled it out. And Andy has always been someone who's been on the explaining the analytics side of things so that marketers can understand what it is they're working. Andy also comes out with a annual study specifically related to the blogging community. Isn't that true, Andy?
- [05:11]
Andy Crestedina
It is. This year was the 11th year for our annual content marketing and blogging sur, which is a fun piece. It creates a little splash which is good for our brand, but it answers questions like how long does it take to write a blog post? It's just under four hours. How long is a typical blog post? It's about 1400 words. Lots of research. And now with the AI question you can start to see like what percentage of bloggers are using these tools and for what reasons. And no surprise, from last year to this year, the adoption of AI tools has doubled. There's like only 20% of bloggers report not using AI is down from 40% the year before.
- [05:45]
Michael Stelzner
Okay, so we have a lot of marketers listening right now and they are probably using AI for all sorts of other things and not necessarily helping them with analysis. So help them understand the upside, the unlock if they pay attention to what we're going to be talking about today.
- [06:02]
Andy Crestedina
Yeah, you use tools, we all use tools. And some of these tools have reports. Many of those reports can be exported. Exporting a tool was not useful to me before, but today it is. Because if you export a report from an analytics tool, social, email, GA4, any tool, and then if the data is cleaned up enough and easily ingested by the AI, you can just really upload that and start talking to your reports, asking for insights, looking for correlations, finding new ideas from the same data very quickly. I never had the budget to hire a data scientist. Now I don't need to because AI almost works in that way. So the key is to look out for that export button and then sort of know how to prompt and ask it for various types of insights. It's super fun.
- [06:49]
Michael Stelzner
Well, and what I love about what you said, I've never had budget for a data scientist, but now I kind of do. I think that's absolutely true. And what's really fascinating is so many people are not analytical. Maybe they're more creative. There's a lot of marketers that find themselves to be exceptionally creative and their eyes roll backwards a little bit when they get inside these analyt tools and they wish they could have just talked to the Tool, for lack of better words, and gotten the data that they want and it started slowly getting there with some of these tools. But now what I'm hearing you say is you can take any source of data, you can upload this source of data into the right AI tool and you can start asking it questions. And that is going to unlock for a lot of people all sorts of possibilities for, for example, all of a sudden you can get a little bit more evidence to back up what your intuition is telling you that's happening, or you can get evidence to disprove that your hypothesis is wrong. Right now all of a sudden you've got access to a very smart tool, these AI platforms that allow you to make a case, right, or to debunk a case. Do you find that's one of the big unlocks we're talking about here?
- [08:02]
Andy Crestedina
100% for the data driven marketer. Every idea that you've ever had in marketing is in fact just a hypothesis. All best practices are hypotheses. So the scoreboard and the validation for checking to see the efficacy of any action we take in marketing is the data. So there's more to life than data, of course. And there's lots of things I do for which there's really no measurement. You know, trying to be top of mind or get referrals or make make something memorable. Lots of things don't show up in your data and that's fine. But many of the actions that we take do have a measurable impact. And when that's the case that shows up in a report, it's very powerful to have a new way to look at this data. Of course we have to make sure that we have permission to upload that data. I'm only uploading the data from my own accounts unless I talk to the client first. So that's a concern for some people. Really there isn't anything in any of these reports that I know of that's personally identifiable information, PII or anything that's confidential or financial or healthcare or anything like that. So we can probably set our minds at ease about that. But no, it's amazing, Mike. Not only do these reports get analyzed by the AI, but the subsequent action that you might take based on that analysis also possible within the AI. What worked well in which channel, it gives you some insights. Now help me, you know, make recommendations for how to improve traction in that channel or suggest new things that might work in that channel. AI, unlike normal reporting tools, actually can help you take the next step and execute on the ideas that you have based on the data, which is not something any of my tools did before.
- [09:29]
Michael Stelzner
Perfect. So we're going to dig in for the rest of the interview today to talk about, first of all, how to analyze content. And then we're going to get into how to analyze data. Now, data and content, you know, it might sound like they're the same, but they're not necessarily the same because you can have data on almost anything. We're going to start with the content perspective. So let's explore together, Andy, how AI can help me analyze my content. And let's just set the stage for what kind of content, you know, you could be analyzing with AI.
- [10:03]
Andy Crestedina
Yeah, anything that you have that's public facing. An ad, a social post, email, subject line, an article, your homepage, whatever it is.
- [10:13]
Michael Stelzner
Your YouTube channel. Right. Your blog.
- [10:16]
Andy Crestedina
Yep.
- [10:16]
Michael Stelzner
Right.
- [10:16]
Andy Crestedina
Yep. All of it. If you train the AI first on your target audience, teach it about your reader or your buyer, upload your ideal client profiles, probably take out your brand name first. If you're concerned about AI's memory, or have AI actually help you by creating a quick AI Persona, once it's trained on your target audience and it sort of knows their hopes and dreams, their fears and concerns, their emotional triggers, their information needs, then upload the asset, whatever it might be, to AI and say, to what extent does this asset meet or not meet the information needs of my audience? What are the unanswered questions I wrote from this article? What key pieces of information and unaddressed objections are on my homepage? Whatever it is, you can talk to it. So it's all based on that Persona and garbage in, garbage out. So you need to work very hard on that. It's a bit lazy to just use the. And I could share it. There's a Persona prompt that will make a quick Persona for you, but validate that, look at it, check it carefully, and then it will basically tell you what gaps are in that thing you made. One thing that I kind of try to highlight here is that I'm not really using AI to find efficiencies. I'm using AI to find deficiencies and to make more impactful, stronger assets for anything that I'm doing. Because, you know, once you have that Persona loaded in, you're basically have like a synthetic member of your target audience that you can talk to day or night about anything at all. It's really fun. So that's a qualitative approach to auditing your. Your stuff and doing it without. Without any data at all. But AI is Extremely useful at another point of view on any. Any content or any asset anywhere.
- [11:51]
Michael Stelzner
So let's explore Personas a little bit, because even though I've had other people on the show talking about this, it'd be really useful. How detail of a Persona and just to define a Persona from everybody's perspective, I mean, well, why don't you define what a Persona is and what the elements of a good Persona is? I won't even make any presuppositions as to what you might define it as.
- [12:10]
Andy Crestedina
This isn't my main focus in my marketing, but I'll use the definition from Ardath Albee, Persona expert. You and I both know a Persona is a format that makes your audience intelligence actionable. So it's really just a summary of the person that you're selling to. It sometimes includes a bunch of demographic information, which Ardif downplays. She says mostly in B2B. That's not that useful. Who cares if this person is, you know, 42 years old and drives a Volvo? We're trying to sell them some enterprise software or something, so the demographics is less. She tries to deportize that, but in my prompt and I sort of listed it, what I want back from my Persona is their hopes and dreams. Interesting. Often includes kind of a personal thing. Their fears and concerns. What would stop them from moving forward in this sales process or through the marketing funnel? Their emotional triggers. Frequently your buyer sort of wants to look good. Esteem is one of their goals. They have career goals as well, which come up in this research. And then finally their decision criteria for selecting a company like yours. Everybody needs to know the decision criteria for selecting a company like yours. Because if you. You don't know that, how can you even write a web page? What does your homepage say? Like, you have high risk of being way off. So, yeah, in that prompt, it's basically build me a Persona for a job title in an industry at a company size, in a geography with responsibilities and goals and challenges. List their hope streams, fears, concerns, emotional triggers, and decision criteria for selecting a company in my vertical. It'll make one for you and it will be wrong and you'll have to tell it to fix it. Remove that concern. Add this concern. Let's deprioritize that, you know, ad. So you have to work with it back and forth. It's called Chat GPT because it's a conversation. And then when you have the Persona that looks pretty good, you feel like it's like 90% correct. I would copy that out of the AI and put it in a PDF file and share it with your team if you're not a solo marketer. And then now you have a foundation for all of your subsequent conversations with AI. I don't think AI is very good. Before you do this, I joke. AI might as well stand for average intelligence or average information until you give it your Persona. So at that point, yeah, you can show it anything and say, to what extent does this meet or not meet my Persona's information needs? And it will list for you all of the most important questions that your article, homepage, social post, email failed to answer. What are the most important questions I failed to answer? And it will tell you, and some of it you won't agree with and you can just ignore it. But one or two of those nuggets might be extremely important and make a real impact on leads and deal flow.
- [14:39]
Michael Stelzner
I love this. I want to ask a few questions here. I wish I could type and talk at the same time, but I can't. So. Okay, back to the Persona real quick. How long typically is this? We talk in paragraphs, pages, one paragraph. Like, just give me a sense of how detailed you go normally.
- [14:54]
Andy Crestedina
I did one this morning with a client that didn't have a handy Persona. We should all have one. We should all have documented ICPs, ideal client profiles nearby. But if you want to take the shortcut or get a head start, that Persona prompt, it's probably four or five sentences, the prompt.
- [15:09]
Michael Stelzner
What about the output though?
- [15:10]
Andy Crestedina
The output is usually four separate bullet lists. So it kind of lists, you know, gives you some high level view of this person. It'll name them. ChatGPT usually names it Alex. Okay, your Persona is probably going to be named Alex. And then it'll have hopes and dreams. Here's six bullets. Fears and concerns. Here's six bullets. You know, emotional triggers. Here's four bullets. So it's just this very structured thing that you can kind of go through. And it's important though to review it again because everything afterwards is going to be very, almost like literally like side by side comparison with those bullet points that it put in your Persona. But I made one this morning with a client that didn't have a handy Persona. And you know, it's this kind of long thing. And then that thread becomes where I do all my subsequent analysis or I copy and paste it out of there. And I now have a folder on my desktop with probably Personas from 15 different clients. And I just use them all the time.
- [16:01]
Michael Stelzner
Love it. Okay, so step one to analyzing Your content is really to make sure that you've got a really well baked Persona. Right? And obviously a lot of people have probably learned that there's things that they're missing from their Persona based on what we just talked about here. For example, emotional triggers and decision criteria for selecting a company like yours. And like you said, AI can help you come up with stuff, but it's going to make it up and then you're going to have to go through it and you're going to have to validate it by either talking to your other people in your marketing department or your sales team or whatever to kind of make sure this is accurate. And then once you have this, you can reuse this in future prompts, which I absolutely love. So now I would love to talk about this concept of gap analysis, which I know you've kind of been alluding to a little bit. But let's take one example. Let's take the example of a sales page, especially B2B. People generally have a page that they've designed that explains what their offer is. So how could, how could we do a gap analysis on something like that?
- [17:00]
Andy Crestedina
Well, I'm doing this so frequently. I could give you an example from one just like hours ago.
- [17:05]
Michael Stelzner
Let's hear it.
- [17:06]
Andy Crestedina
It's a client that has online coding boot camps and they have these pages from which people are coming from search and landing on the page. And these pages have click through rates on the call to action. So the first thing we do is measure the click through rates and the call to action and set a benchmark. It's something like 2% of people who land on this page click on the call to action. Okay, how could we improve that? Well, people click on calls to action when the page addressed their concerns and answered their questions. And it had some supportive evidence. Great. I make a Persona, we look at it, we think it's close enough, and then I take a full page screenshot of the page. I'm using Chrome. There are Chrome extensions for this. The one I'm using is called Go Full Page. You can also use another one that's called awesome Screenshot. Techsmith's tool Snagit, makes full page screenshots. Use whatever tool you like. But the visual full page screenshot rather than the link or the copied and pasted copy is important because for conversion optimization, I want this thing to analyze the page with the visual hierarchy.
- [18:03]
Michael Stelzner
You're talking about AI now you're specifically when you say the thing.
- [18:06]
Andy Crestedina
Yeah, I'm going to give AI a page and this is a page about coding boot camps. But I want to give it a page in the, in the full visual of the page because some of the conversion elements on this page are like logos or awards, you know, or things that AI won't really see if I just give it the code of the link. So best practices for doing conversion optimization using AI, Give it the Persona and give it a full page screenshot of your URL, the money page, the page where the person's persuaded or they're not.
- [18:33]
Michael Stelzner
Well, and I will add a little bit of my thoughts in this. For those of you that have really long sales pages, AI is going to freak out. It's not going to accept it. So you're going to have to break it into multiple screenshots because we have really long sales pages on some of our products and it's just too big of an image. And in that case, break it into multiple images and that resolves it. So, okay, you're specifically talking about a problem where you're trying to determine why the click through rate at the bottom wasn't strong enough. So keep going with that story. How is AI helping you with this?
- [19:03]
Andy Crestedina
Yeah. So for these coding bootcamps, a person decides to sign up for a coding bootcamp if they are confident that this investment in time and money will help them reach their goals. So they have to sort of get their questions answered. Is this an online coding bootcamp or is it in person? How much does it cost? What are the placement rates? What are my career opportunities if I take this coding bootcamp, Do I need coding experience prior to signing up for this bootcamp? These are just the basic questions. If you were the salesperson and I was the prospect and I called you, what would I ask you? What would your answers be? That's how to make a page. That's how the Internet works. The highest converting pages emulate a sales conversation with the top rep, the order of messages, the answers, the addressing of objections, the supportive evidence at each step. So you want to see like case and success stories and people's lives who were changed by getting these new skills and programming. So those things must appear on the page. The page as it was really wasn't going deep into those things. It was kind of like, what is coding? What is programming all about? Like, it wasn't talking specifically about the boot camp or the impact on this prospect's life. And the AI laid it out, it just said that it's like this page does not include any urgency. Oh, that's another one. These coding bootcamps have deadlines to sign up. Just like World. The page didn't mention, it didn't there's actual scarcity and it didn't indicate scarcity. Thereby it failed to trigger the loss aversion, which is a cognitive bias that has a powerful impact on conversion rates. So all kinds of things that we just saw at a glance. Literally it was we were only four minutes into this exercise by the time we had this list.
- [20:37]
Michael Stelzner
So what's the question you're asking AI? You're asking it to identify problems that would stop the ideal target audience from taking the desired action. Is that kind of what we're doing there?
- [20:46]
Andy Crestedina
Thank you for reminding me to take that half step backwards because I didn't really talk about that prompt. This is the important part. The prompt that audits the page after it has the Persona is somewhat detailed and it's basically my best practice after having built a thousand B2B lead generation websites over 20 plus years, it sounds like this. You're a conversion optimization expert. You are skilled at evaluating pages and their ability to both inform and persuade the most compelling highest Converting pages on the Internet share some common traits the following are our best practices for service pages. The header clearly indicates the topic of the page, letting the visitor know they're in the right place. The copy answers the visitors questions and addresses their objections. The order of the messages aligns with their prioritized information needs. The copy uses supportive evidence to support its marketing claims such as testimonial statistics, case studies, awards, logos, years in business. The page connects on a personal level by using human elements like faces and quotes and stories. The copy triggers cognitive biases in subtle ways when relevant, such as loss aversion and urgency. And it has a compelling and specific call to action. Now I'm giving you a web page. Create a list showing the ways in which the page does and does not meet the best practices above.
- [22:00]
Michael Stelzner
Love it.
- [22:00]
Andy Crestedina
Kind of a detailed prompt, but then it actually will look at that URL as exported from your full page screenshot against all those criteria.
- [22:07]
Michael Stelzner
I love it. And you can apply this prompt to just about any marketing persuasive thing that you're creating, right? You could probably do the same thing in a presentation if you're giving a sales pitch or anything else for that matter. Now, cognitive biases. You kind of hinted at that a little bit. I'd love you to like unwrap that a little bit just so my audience can understand what that means. Missing cognitive biases first of all, what are they? And just help us a little bit.
- [22:32]
Andy Crestedina
Understand that the reality is, and there's a ton of research on this, humans are not really good cost benefit calculators. We are inefficient and we use heuristics and shortcuts to make quick decisions. And ultimately we are sort of making decisions based on emotional reasons and then rationalizing them afterwards. That is, read any book about behavioral economics or Daniel Kahneman or Dan Ariella or any of these thought leaders and you'll they have strong evidence for this. Therefore, it's an opportunity for marketers to at least not miss opportunities to trigger a cognitive bias when relevant, you don't have to hit them over the head with it. It can be subtle, but in this example, there is a deadline for signing up for that course. Reminding people that they could miss this opportunity is important, right? It's important for the conversion, it's important for the visitor, and it absolutely triggers loss aversion. Humans will expend more energy to reduce the chance of loss than to seek a gain. It's a very strange human phenomenon. Loss aversion is a real thing. You could look it up, go to Wikipedia. It'll give you a list of all kinds of cognitive biases. Nancy Harhud is a leader on this topic. She's got a great book about it and I've got other prompts that basically look at any URL from that lens and say, here's 15 cognitive biases listed in Nancy's book. Rate the extent to which this page missed opportunities to trigger those biases. It will audit the page from that perspective.
- [23:53]
Michael Stelzner
Love it. And Nancy will be speaking at Social Media Marketing World, which is going to be awesome. Many of the top experts you've heard on this show will be speaking at Social Media Marketing World 2025 and with your AI ticket, you can attend at a very economical price. You'll discover practical AI workflows and advanced AI automations that will increase your productivity and save you time. Imagine getting live and in person training from Matt Wolf, Chris Penn, Jeff J. Hunter, Rachel Woods, Molly Mahoney, Brian Piper, Jeff C and many others Isn't it time to enhance your career by fully embracing AI? Grab your tickets now at social media examiner.com aikon so what I love about what you're saying here, and I think an even more simple version of what we could do here, is once we've developed this Persona on who this Persona is, my guess is we could take any piece of content that we're creating, like an outline for a YouTube video or a script from something or anything else. And we could just simply ask it, what is this content missing? It could have been something we previously created, like a blog post, Right. What is this content missing that my core audience, my primary audience that I'm trying to address would need to understand? And then we could use that to create another piece of content.
- [25:14]
Andy Crestedina
True, exactly. So you've got the Persona and then any asset, any public facing or even internal documents, you know, strategic marketing assets like your content calendar or your content strategy or whatever. You can write a prompt that lists the best practices for that thing and then the end of the prompt sounds like this. Rate the extent to which the following does or does not meet the best practices listed above. By doing that, the extent to which it does or does not, it's going to get the AI to give you two lists and one shows. Yeah, you check these boxes, good job. But you could work harder on these four, consider changes or improvements. So that idea is like you can audit anything. And I've got a post about this. It's like how to audit anything in your marketing. It's a very big promise of a headline. How do you use AI to audit anything in your marketing? But that's basically it. You know, you give it the Persona and then you give it a prompt that includes best practices for that asset and give it the asset and ask. Rate the extent to which this asset does or does not meet the best practices above and it will do it.
- [26:12]
Michael Stelzner
Do you have a preferred AI platform that you're using to do your analysis right now?
- [26:16]
Andy Crestedina
I do, but it's not preferred for a strong reason. Mike. I'm just using ChatGPT because it's popular and because I want my content to be accessible. I'm not advocating for one over another, mostly.
- [26:25]
Michael Stelzner
Got it. You could do this just as easily with Claude or a Claude project or something like this. Right.
- [26:29]
Andy Crestedina
Claude's probably easier actually, because I don't think you need the paid account to upload things. ChatGPT, you need the $20 per month plus account to be able to upload files.
- [26:37]
Michael Stelzner
Very cool. All right. I think you might need the paid account with Claude to be able to do that. But I'm not 100% certain because I use Claude and I just love Claude. But I know ChatGPT is very powerful as well. Good. So whatever every anybody's using, you can. If you have multiple accounts, you may as well try them both and see which one delivers better output. You got nothing to lose. Okay, so we could just spend the rest of this interview talking about content but we're not. We're going to now move into data. So we're essentially using chat, GPT, whatever, AI platform to help us see things that we cannot see. Right. And we've talked at length about how we can use this to essentially very quickly identify challenges with our existing content and opportunities. Now, let's step into the world of data. And when I say data, I mean numbers, right? I mean, that's mostly what I mean. So unless there's something more to this than just numerical data. But how can AI help us analyze data?
- [27:32]
Andy Crestedina
Well, the outcome of any adventures into analysis and data typically comes out with, these are working well, let's create more of those or get those to work even harder, or these are not working well. How can we fix that or abandon it? Okay, that's mostly the outcome of any type of analysis. You're going to end up saying, these are strong, these are weak. How can we make more strong players and fix the weak players or give.
- [27:53]
Michael Stelzner
Them up real quick before we dive into this, Is it the presumption that you're stating that we kind of are aware of this already because this data is available to us inside of Google Analytics 4 and other kinds of resources, or we've done our own backend analysis and we kind of know what our performers are because we're using this tool or the other tool behind the scenes to kind of indicate what's generating the most leads? Or are we coming to this with not even knowing the answer to that question?
- [28:16]
Andy Crestedina
You probably don't know. There are lots of things that we don't know of yet. And that's my hope, is to discover those and to make that recommendation in the end, like, to find that insight of like, hey, wow, I didn't know that things were doing great. Whenever I talk about that topic, you know, people love it. I didn't really see that from a distance. Now I see it now. I can adapt my content strategy, for example, if we're doing content. But no, that's. Hopefully that's what we're going to discover here is like, who are the outperformers and the 10x contributors? And then who are the very weak performers that we. Why do we keep doing those? We got to stop. Like, that's a waste of time. Those are never getting us anywhere.
- [28:48]
Michael Stelzner
So how do we go about starting with this, using AI to help us analyze raw data?
- [28:54]
Andy Crestedina
Well, as a content marketer, I'll give you a good starting use case. All of us have, in content, have published things, and those things vary wildly in Their performance. Have you ever noticed like every chart in marketing is like an exponential curve and there's like a tiny percentage of a few things that are getting amazing results and most things get nothing?
- [29:10]
Michael Stelzner
Of course.
- [29:10]
Andy Crestedina
Yeah, I know. Everyone kind of knows that. But the strategic insight from that maybe isn't as obvious. Let's stop making medium quality things and focus more on those high performers. That's what I try to conclude. So in GA4, probably the most popular report is just called the Pages report. And if you go to the Pages report, you can export a file that shows the performance of all of your articles in all of the channels and then give it to AI to find out what's working. Here's how to get that report out of GA4. In the pages Report, we're going to change the first column. There's a dropdown above the very first column which says page path. Change that to page title. Now it's got the title tags of all of your content. Great. But I want to see where each of these things performed, not just generally. I want to, because some of these probably did great in search, some of these probably did great in social. And what works in Search vs Social is Opposites. Right. The things that work in search are things that sort of meet expectations. The things that work in social are sometimes unexpected. Right. They're actually opposite channels. So I like to add the second column of the source of traffic before giving it to AI. To do that, you click on the blue plus next to that first dropdown. It's to add another column. It's called a secondary dimension. The dimension you're going to choose is called Source Medium. That's it. I know that sounded complicated. That's the whole. That's all you got to do. Click the export button. So I've got a report that has all the title tags in column A. All of the source mediums, LinkedIn, social or Google Search or you know, email campaigns in the second column. And then to the right, all of the metrics. Right. Engagement rate, the total sessions, the key conversion rates. The Export button in GA4 is. It looks like a little share button. It's in the top right. You click that export. You can choose different formats. I just use CSV. Wow. Open it up.
- [31:03]
Michael Stelzner
Wait real quick. Before we export this, do you recommend a certain time duration? Do you recommend like all time? Do you recommend weekly, monthly? What's your thoughts? Because that's the one variable that we haven't discussed.
- [31:14]
Andy Crestedina
Thank you. Yes, I would normally just grab a giant date range. I mean, you And I have talked about this once before and you were talking about excluding the recent period because of that's when you are using your, your marketing horsepower to drive certain traffic to certain things.
- [31:25]
Michael Stelzner
Right.
- [31:26]
Andy Crestedina
I think you're a fan of that, right, Mike, you include the recent date range?
- [31:28]
Michael Stelzner
I mean, honestly, I don't even know because I don't even do it anymore, to be very frank with you. I've got somebody on my team doing it. But what would you like look at a year or something like that, presumably?
- [31:36]
Andy Crestedina
Yep.
- [31:37]
Michael Stelzner
Okay.
- [31:37]
Andy Crestedina
Last prior 12 months is one of the choices. If you choose the date selector, scroll down, you can just Click on prior 12 months.
- [31:43]
Michael Stelzner
Okay, so we're getting this file, presumably a CSV or an Excel file. What do we do with this thing once we have it?
- [31:50]
Andy Crestedina
The next perspective is before uploading data to AI, you need to clean it up. There are prompts where you can tell it to clean up its own data. It doesn't, but I'd highly recommend doing it by hand because it's going to choke on that. It's going to struggle. You're going to ask it to make edits to the file and it might mess it up. I sort of don't trust AI to edit CSV files. I've had too many problems. Like, you ask it to like remove all the non English title tags and then it just removes almost everything. Like what just happened? Why did it do that? You know, I only have three rows anymore, so I would open up the file. The first nine rows are comments. Delete those. That's GA4. Adding some notes to the top of your CSV file. And also, you know, just scroll down and you're going to see just tons of weird outliers with very low amounts of data. Just delete all those too. Probably all your, you know, non first language rows, you know, because the title tag in GA4, it might be whatever the person's browser setting changed, you know.
- [32:44]
Michael Stelzner
Oh, I see what you're saying. Okay, this is fascinating. So just to clarify, Google will change the language whether you wrote it in another language or not. So you can just get rid of all that stuff if you don't care about that. That's fascinating. Okay.
- [32:56]
Andy Crestedina
Yeah. And I used to do this with prompts. I'd say, you know, remove all the rows with ultra low number of sessions or remove all the rows with non English title tags or no, I just scroll down and I chop the bottom two thirds of this giant report with just tons of low data rows. Okay, I don't need it. We're not actually looking for perfect accuracy or perfection, right? We're just giving it a bunch of data and asking for insights and then getting some ideas. And hopefully I'm avoiding all the sociological concerns, the tech concerns about accuracy and bias and labor market impact and everything that I'm doing here, AI for analysis is just me getting ideas and then I'm going to filter through those and make my own decisions. I don't really trust the AI, so I only need sufficient data to get some insights and then I will just go through those by hand. But yeah, we clean out the data, right? It's not structured data. And GA4 isn't even that bad. There's lots of tools that you can export data from, like LinkedIn where the data is a mess, it doesn't really make sense. So unstructured data exported from your favorite technology probably needs a little help before you should upload it to AI.
- [34:01]
Michael Stelzner
Okay, so we're cleaning up all the stuff that doesn't matter and deleting all the columns and rows and stuff that is just extraneous information re exporting as a CSV. It's on our desktop now. What do we do next?
- [34:12]
Andy Crestedina
We're going to drag that into ChatGPT or Claude or whatever your LLM of choice is, and we're going to start asking it questions to find insights. I could pull up the prompts and read them verbatim, but basically what you're saying is, I'm giving you a GA4 report that shows all of my content and its performance across channels, so you can ask any question you want. Now, for example, find correlations between topics and channels, which topics tend to work in which channels. Or perform a semantic distance analysis and show me which articles I have never written, but almost written.
- [34:45]
Michael Stelzner
What does that mean?
- [34:45]
Andy Crestedina
Well, once it has all your title tags, it can infer the topics from the title tags. And so you've written 50 articles, 500 articles, but there's many things that you've never written in that data set and it can check the distance of, between the meanings of all the things you've done and say, ah, you know, there's a gap here. You've surrounded this topic, but you've never made that piece right there on that topic. So it'll help you find the gaps in your content marketing.
- [35:08]
Michael Stelzner
So I would imagine we could say this in a different kind of natural language kind of way, right? Like analyze all the things that I've written in the last year and identify titles. I mean, I would imagine either one right that would be relevant for my ideal customer audience that I have not yet written about. Something along those lines would accomplish the same thing.
- [35:26]
Andy Crestedina
Yep. I use certain words because I know them to be effective at getting AI to think about these things. And this is actually one of the benefits of getting prompt ideas from those who have been in the trenches for years. So when you use certain words like semantic distance analysis, that's not a word everyone uses every day. But SEOs like me often think about that. Let's say you're researching thought leadership. You know, help me find counter narratives that would be interesting and compelling for my Persona. Counter narratives isn't a word that most people use every day, but the social media marketer might, or the content strategist might. So yeah, you can do these in plain English, but I find that there's certain words that will trigger the AI to go through certain types of analysis.
- [36:02]
Michael Stelzner
I love the idea that you're using words that infer it is acting in a certain kind of way on this data rather than a generic way. Are you also starting out your prompt with something along the lines of you are a world class data analyst for content marketing and Google Analytics for data? Something along those lines just to kind of like tell it that this is how you want it to behave specifically with. And I could tell you brought up the prompt. I mean, do you have something like that that you're starting it with?
- [36:34]
Andy Crestedina
Yeah, that's the classic format for prompting. For me and so many other marketers or just users of AI, you are a job title. Yep. Skilled in skills. So those two fill in the blank pieces at the, at the beginning of the prompt, put the AI into that corner of its giant probability matrix.
- [36:51]
Michael Stelzner
Right.
- [36:51]
Andy Crestedina
To come back and give you the insights. So that's a classic approach. I'll find one of those.
- [36:56]
Michael Stelzner
No, that's totally cool. I mean, you don't even need to worry about it. I think that's really important that people understand that it's a little different than what we've talked about previously because you're going to get when you tell it to act in this kind of way, it's going to know when you drop these words that you'd have to scroll back the recording to, you know, it's going to understand the context in which we're asking these words. So what do we do with some of the stuff that comes out of this? Is it going to be wrong at first? Do we need to do a deeper analysis on this? What's your thoughts on this?
- [37:25]
Andy Crestedina
Well, if the prompt works you'll see insights right away. So you upload this file and it sounds like something like, you know, you are a Google Analytics expert and content strategist skilled at aligning topics with marketing channels. I'm giving you a GA4 report that shows all of my title tags and topics along with the source of traffic and the engagement within that channel. Make recommendations for my content strategy to improve engagement, improve efficiency. You know, show me my top performers. Write a headline that is the or title tag that is the most likely to perform in a certain channel. What topic should I stop publishing on completely? What do my top performers in Channel X have in common? What is the expected engagement on this topic? If I published that, what would you expect to see as the traffic and engagement in social versus search? So it can actually start to predict like here's a title, here's how well it might work. These things work well there. These things don't work well here. It's actually fascinating to just have this subsequent conversation about it and see what's working because it now knows all of this. A year's worth of content, what worked.
- [38:32]
Michael Stelzner
And where are you providing in this GA4 export a link to the original source so that if it wants to, it can go query the original content. Or are you simply just providing it the high level data on the title and the numerical data.
- [38:46]
Andy Crestedina
It's not going to go back and look at those and there's no link in the report anyway for it to go find that. But if you did want to take the time to enhance this data, you'd call it probably if you want to enhance this data with one thing. I would and I've done this and it's a bit magical and one of my favorite things to do, even though it's quite time consuming, make another column in the report and then in that column put in the topic for that.
- [39:08]
Michael Stelzner
URL and that you could already have topic tags on your website that you could just look at to get that data right.
- [39:13]
Andy Crestedina
You could. I don't know how to automatically add that to the GA4 report, but it's.
- [39:17]
Michael Stelzner
Manual but it might help you is what I'm trying to say.
- [39:20]
Andy Crestedina
If you had that in. Yep, yep. If that's all. Like if there's, you know, these things are all tagged in your CMS and there's a certain thing and you can go back and maybe a vlookup will combine these things so it can export them from WordPress and you know, there's probably ways to do it. I've done it manually and let's say I spend an hour and that's worth it to me to go through the top 50 or 100 posts or articles and then add topics, AI, content strategy, SEO, content strategy, UX, web design. Because now once it has all topics for that, you have greater confidence when you ask it to find correlations between topic and channel performance, which is very important as a content strategist. Every top on the Internet has, you know, a fit or a misfit in its channel. Some things are great for search and some things are great for social and they are often not the same thing at all. If I write an article about how to create a great FAQ page, people who search for that want it badly and they want it now. Nobody in a social stream is suddenly interested in how to make a better FAQ page, right? So there's certain things, if a thing is kind of a counter narrative or a strong perspective or highly collaborative or highly visual that's going to perform in social. If a thing is in fact just satisfying someone's information needs, answering questions, it's a long form text based how to thing, right? Not very visual. And it's a topic people look for that's going to work in search. So I've always had to do it manually until now. But finding the topic channel fit is an important moment for a content strategist when they realize like you can begin to build your content for performance in a channel instead of just making one thing and flinging it everywhere. Some things you don't. You could forget about keyword research. It's no one's looking for it at all. Other things, you know, this is absolutely a good keyword opportunity. But 0% of people on a social stream care about it today. So AI can help you find those correlations much better if you tag as another column the topics of each of your articles.
- [41:17]
Michael Stelzner
Talk to me about how you're using AI also to give you a visual representation of, of some of this stuff because I think that's kind of magical.
- [41:24]
Andy Crestedina
It is magical. It feels like magic. It's like this shocking moment, like you're like this intake of breath. What did that? Look at that. Because it can make a report for you that shows, for example, that prompt sounds like this and it's the second prompt right after the last. It says draw a color coded heat map matrix that shows the normalized. You have to put that in there because we want it to be balanced. The normalized correlation between topics and marketing metrics and it will draw a chart for you that's going to be beautiful and colorful. That has in one axis all the topics that you publish on, and in the other axis all the marketing metrics, sessions, engagement, conversion rate, whatever. Then that's just GA4. You could do this for email, you could do it for social, but that chart is going to. You're going to go have a great meeting, put that in a PowerPoint, bring it to a meeting. You're going to look like a wizard when you open up that slide. Where do you get this? Oh, I just wrote a few prompts. You know, it's a fun aha moment for everybody.
- [42:26]
Michael Stelzner
As we're recording this, ChatGPT Model 01 has rolled out to everyone. Have you started experimenting with this new reasoning model to do analysis and have you found it to be any better than 4.0, which is what everybody was probably using just a few weeks ago?
- [42:41]
Andy Crestedina
I'm sure it's better. I've heard it's much better. Or rather it does. The training on it goes to another level and that the processing power required for each prompt is much greater than before. So O1 is going to be a breakthrough. I have not experimented with it yet. The other thing that I want to play with more, which I think a lot of people are going to be opening their eyes to, you know, as they hear. This is the Canvas ChatGPT Canvas, which is this new UX where the thing you're working on, the writing project is in one frame and the conversation with AI is in another. So you're sort of going back and forth in the same tab where up until now a lot of people have been moving content back and forth. And I'm in a browser, I got a Google Doc open over here and AI over there.
- [43:22]
Michael Stelzner
Exciting frontiers and projects just rolled out too, right? The ability for you to kind of move threads into different little sub projects and data sources. We are in a rapidly evolving world. By the time everybody listens to this, I guarantee you there'll be things that we didn't even know when we're recording this, they're going to be coming out. This is why you need to listen to this show. This is why you need to follow people like Andy. Andy, first of all, thank you for making it easy for people to wrap their heads around how they can use AI to analyze their data. There's going to be plenty of people that want to learn more about you. What's your preferred social platform for them to connect with you on? And if they want to learn more about your business, where do you want to send them.
- [44:02]
Andy Crestedina
LinkedIn is my best network by far. Anyone listening to this can skip the blue follow button and find the three dots and scroll down and go to the connect button and then we can DM each other. If you want to send me a message, Orbitmedia.com is where I post an article every two weeks and that you can learn more about our business there. But also the book which you mentioned up front has just been updated. Version 7. I have to keep writing this thing to keep it up to date. Version 7 of Content Chemistry has all the GA4 reports and the AI prompts we just discussed, so it should all be there.
- [44:29]
Michael Stelzner
Very cool. Andy, thank you so much for coming on the show today. Today.
- [44:32]
Andy Crestedina
Thank you. This was so fun and I'm so glad we've stayed in touch after all these years.
- [44:36]
Michael Stelzner
Hey, if you missed anything, we took all the notes for you over@socialmediaexaminer.com a35. Also, be sure to follow this show on your favorite podcasting app. And if you've been a longtime listener, would you do me a favor? Can you give us a review on whatever platform you're listening on and maybe let your friends know about the show? I'm Telzner on Facebook, tellsner on LinkedIn, and ikestelsner on X. And do check out our other shows, the Social Media Marketing Podcast and this 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.
- [45:19]
Andy Crestedina
The AI Explored Podcast is a production of Social Media Examiner.
- [45:28]
Michael Stelzner
Don't forget to get your AI ticket to Social Media Marketing World2025. Become an AI Enhanced Marketer. Grab your tickets now at socialmediaexaminer. Com aicon.