Intentional AI: Why AI can make your content personalization worse

VIRGIL 0:00
Using content personalization to target your audiences better is such a powerful thing, but using AI for it sometimes makes it more complex than a lot of times the sales pitches that everybody gives you makes it sound. You got to set the direction or really AI is going to choose one for you and it might not necessarily be the right one. So let's break this down as we start discussing stupid. Hi everybody, welcome back to the podcast. As with me always is Cole.

COLE 0:44
Hey everyone.

VIRGIL 0:45
We got a new one here for you today, continuing our Intentional AI series. And today, Cole, we're going to be talking about.

COLE 0:54
Personalization.

VIRGIL 0:56
Personalization. What an exciting topic.

COLE 0:59
Yeah, if you're with us for the last episode, towards the end, Seth, our last guest alluded to, once you've had your content created, once you've gone through steps of content management stuff, you might want to go through the steps of personalizing it for different audiences.

VIRGIL 1:17
And what a hot topic that is, because I mean, when you think about it, I mean, you know, when you look at AI, it's all about AI personalization, yada, yada, yada. And that there'll be so much you can do around that, right?

COLE 1:30
Yeah, I mean, there's a lot of talk about it. It's honestly pretty confusing, even to me, but we've had a lot of discussions about this topic of personalization. It seems like there's kind of two lanes of it. You have communication personalization and you have website, or I guess you could say like role-based versus behavioral-based. How would you break it down, Virgil?

VIRGIL 1:50
Yeah, I mean, that's probably fair to say. I mean, obviously for our testing for this episode, we looked at communication based where we, and we'll get into that later about the e-mail, formatting that we did. But I mean, overall, I mean, personalization takes a lot. I kind of personally see it differently and that I see other potential that it could bring things like, but going back, you think of personalizing communication. So, whether it's target audience, a role, maybe the person using individual information out of a CRM or something like that to be able to generate an e-mail to them that makes it feel like more it's to them or, content, to people in that. But the other side is kind of the thought of doing it in your digital mediums, whether it be your websites or maybe your social channels and that, and how you interact with people and, doing A/B testing and multi-variant and all this kind of stuff. And letting AI kind of make those decisions. I honestly, I think personally there's a third one that is probably not as well understood and maybe isn't a real easy use case right now, but hopefully will be, where I think more so in kind of predictive analytics. So otherwise predicting, you know, getting AI, it'd be able to imagine if it could look at your web analytics, your social media analytics, you know, kind of like your podcast analytics, all those kind of things, and kind of look at it and actually give some predictive behavior or analysis on what people are doing and what they anticipate to do and where you maybe have bang for the buck. And I think, you and I both agree is we've kind of used AI tools more and more where we've really seen them give the most amount of power is when you ask them to look at something and give you some kind of analysis back to it versus creating something originally.

COLE 3:48
Yeah, you kind of have to know like what you're asking though. You can't like go into it not really having a compass. I guess that's the best way I can think about saying that right now. But yeah, that's the real advantage of AI with personalization, I think, is when you kind of know like who you're trying to target, who you're trying to personalize for. and the criteria for each group.

VIRGIL 4:13
and honestly, that's been personalization since the beginning of time. That's not a new topic with AI. I mean, the reality is, especially when you talk about like content management systems and kind of marketing companies, that they've all kind of sold everybody this bill of goods that like, oh, you can personalize your site and you can do all this stuff. And we've worked with a lot of content management systems that have great DXP kind of digital experience type aspects to them. But the reality is, when you talk about the percentage of implementations that have been done versus the percentage of people that actually adopt personalization and actually do it, you're talking 100% versus 5 in that. And that's because personalization is such a mind bend. This above and beyond any area is one that if you don't already understand kind of what you want to do and how you're going to measure it. Otherwise, if you're going to do, let's just give a simple example. You're on your website, you're going to do an AB test. Some of the people are going to be showing this content. Some of the people are going to be showing this content. The AI component would be in there is that instead of you saying like 50-50, it monitors it and it actually looks at what people do afterwards by looking at your analytics and that kind of stuff. And it makes judgments on, I'm going to show A more because it seems to resonate more. And I'm going to show B less and that kind of stuff. But anyway, doing that. what's your measurement? What's success out of that? What's success out of that? I mean, that's great. You say, okay, well, we've determined that A works better, but what was it supposed to do in the 1st place? What was the end result? What was the outcome? Was it to get people to go further into the site? Was it to access some kind of promotion? Was it to go and fill out the contact form? Was it to go set up an account? Was it whatever like that? And you have to have some kind of measurement. If you can't measure that and actually say that this affected the outcome you want, then It's worthless. And so it's the same with AI. sure, it might be able to help you make better decisions, but overall, if you don't understand. And that's what's such a mind bend for people is they don't really understand how to make changes to things that actually affect it. And yeah, you know, you and I were talking about on the communication side, how, you know, I receive. an excessive amount more spam e-mail than you do, because I've been doing this for 27 plus years and I'm the business owner and I'm on every e-mail list known to man, unfortunately. Not that I opted into any of them, but I'm on them anyway. And you know, it's funny how you get these things and you just know they're AI generated, that they think they're personalized in two of me and I just kind of hit the delete key actually on those faster than the ones that didn't.

COLE 7:10
Yeah. Not to talk about elephants in the room every single episode, but I think the elephant in the room for this one is like personalization is kind of creepy. At least it can be, especially with like communication. Like if the e-mail you're getting or like whatever type of communication is like a bit too like personal and you like don't know the organization who's like reaching out to you. That's like a tricky situation. And then also, but you don't want to be too stiff with your personalization either. You want to be able to resonate with your audience. So I kind of thought of, honestly, right before the recording, I was like, okay, there needs to be like a personalization pendulum that you think about when you're thinking about personalization. Like you don't want to be too personal, you don't want to be too stiff. And so it's on you to find that balance.

VIRGIL 7:34
Oh yeah.

COLE 7:35
That's like a tricky situation. And then also, but you don't want to be too stiff with your personalization either. You want to be able to resonate with your audience. So I kind of thought of, honestly, right before the recording, I was like, okay, there needs to be like a personalization pendulum that you think about when you're thinking about personalization. Like you don't want to be too personal, you don't want to be too stiff. And so it's on you to find that balance.

VIRGIL 8:02
Yeah, I'll give you 2 great examples. And I kind of talked about this a little bit when you and I were talking about this episode yesterday, but The first thing that I think of, and it just makes me laugh at, and I said, I get lots of recruiter type emails where they send me these long tables full of their people available all from a particular country. I'm not going to mention what country, but it's all from a particular single country. And it's like, dear sir, gentlemen, you know, something like that. And it's like, here's a list. in that we love the pleasure of working with you all the time. And they haven't worked with me at all and stuff. And that to me is generic. It was something that took them 10 seconds to put together. They have some automated process and I delete those. But on the other side, I get these other emails sometimes. And these tend to be more from a lot of merger and acquisition companies and stuff like that, where it's like, hey, Virgil, it's been a long time since we talked, but boy, this is great. And I'm like, I have no idea who this person is. I have nothing like that. And I delete those equally fast, you know, and you're right, you kind of have to find that middle ground. And I personally think from a personalization standpoint, it's more about can you hit things they care about versus make it seem like you're contacting them. Because if you're reaching out to somebody that you don't know and have no connection with whatsoever, why would you try to acknowledge a connection that is not there? It's just like, it's just like, you know, after I speak at a conference and a month later I get an e-mail from a vendor saying it was great to talk to you at the conference. They didn't, but that's their generic thing in that. So So again, this isn't a new AI problem. This is a problem with personalization and we could probably go off on that forever.

COLE 9:56
We could. Again, it's just, it's on you to, you know, do that work still. Like the work doesn't go away of, you know, figuring out the pain points, figuring out the priorities and being smart with how you reach out to people, maybe. Yeah, so.

VIRGIL 10:10
No, I think that's what you have to do is you have to be they have to be kind of smart with it in that. So, and again, this above and beyond everything, I mean, we've kind of preached it throughout the entire series and we're going to continue to do that, about you have to have a plan and have an understanding. But overall, this one is even more so because, can you get it? and some of the AI tools out there, they advertise about helping you understand customer journeys and that kind of stuff. But overall, you've got to, Years ago, when I first got into the web business, I very much approached it that, my master's degree is in instructional design and curriculum development. And we're not going to go down that path. But anyway, one of the things is from education is you're supposed to 1st look at outcomes. Otherwise, at the end of a class that somebody's taken, what are the outcomes? They're supposed to know this, they're supposed to know that. I've kind of always taken the same approach from a website. It's like, it's not about what you have on your homepage. It's about what you're trying to get them to do and understanding those outcomes. And that's really what personalization is. Most organizations, when you talk to them, they don't have a very good definition of that. They might say, yeah, they want them to sign up for an account. Well, you want them to sign up for an account and do what? do you consider that metric that tells you that what you're doing is successful? I know we've talked with some financial institutions about, is it to start the loan process? Is it to finish the loan process? what does that whole thing, look like in there? So you really have to understand those and then you have to understand what you're going to do. And I think the opportunity from AI is that you have things like, what if it could actually be paying attention per se to somebody on your site and it could sit there and say, it's not only about how long or like what pages they visit, but it's also how long they are. If I click, click, click and go through really fast, you're not actually getting anything meaningful because they're not looking. But if I click and I spend, you know, a minute on this page and then I click into the next page and I spend another minute on this page and I click into the next page, you're actually getting a meaningful experience. So it's It's something that we have a hard time grading that maybe AI ends up having the potential to do is help us look at this and actually do it. And then on that 4th page you visit, you throw up some kind of promo about finishing that process that you almost got them at or something. So there's a lot of things you could do. do there. And the sky's the limit, but overall, you got to have a plan. You have to have an understanding of what you're trying to do in this area.

COLE 12:56
Totally agree. I think AI provides a lot of opportunity to kind of slim down the workload on that process. And yeah, I think it's kind of just about establishing your parameters, like what you want to measure. And then, yeah, just kind of letting AI assist you in that process.

VIRGIL 13:14
So in speaking of that process, obviously I did a non-plan process with our testing this time around and just kind of randomly asked it based off of our kind of more readable article version and asked it to read exactly or, suggest it. So in the test, like always, we tested 3 tools and The prompt itself that I used was what we tried to target, was it basically getting it to generate different customized emails towards some target groups and that and see what it did. And again, I kept it basic, just to remind everybody, I kept it basic because that's our goal here is to do it how most people are going to do it, which is just to try AI and see that and see what happens. So I said, based off the attached article written, generate 3 personalized emails to promote this article to three distinct audiences. The head of a web marketing team, the director of information technology, and a C-level business executive. Now, the one we used was the one that we did with Grammarly, which we did two episodes ago. So if you have not listened to that episode, we'd recommend in that. And we got a lot of great episodes, so I'd always recommend that you look at subscribing and liking and also getting notifications by following our channel.

COLE 14:46
100%, yeah. So yeah, I think we should definitely get into the results of the tool test here now.

VIRGIL 14:55
Yeah. So we looked at three tools. We used Perplexity, We use Copilot and we use Claude. And this one, I just thought it would be more interesting to use some of the general ones. I thought about using some of the kind of more sales marketing ones, but a lot of those, you know, have a lot of process to set up and I didn't want to do those ones that were specific for that. But, you know, and honestly, most of these ones use one of these, you know, the engines behind these. So, you know, Perplexity using, Claude and Claude using another one. Actually, in Perplexity, I take that back. I actually used Gemini, which is Google's learning language model. And then Copilot, I used GPT-4.

COLE 15:47
Yeah. So, yeah, I took a look at all of the answers from each tool. And wait, so you say you didn't use Perplexity or you used Gemini instead of Perplexity?

VIRGIL 15:57
No, Perplexity, but I used the Gemini learning language model inside of it. So you can choose different models inside these ones. So I just tried the Gemini because I've heard some good things about its newer version. So I decided to use it.

COLE 16:12
Well, it's funny, the order that you listed them in is kind of the order of worst to best.

VIRGIL 16:20
Perfect.

COLE 16:21
Yeah. So starting off with perplexity, I thought that these emails were definitely like the most indirect ones out of all of them. They just kind of spoke pretty distanced from the reader. Like they're kind of on that other end of the pendulum that I was referring to. And then also they kind of referenced the article as if we didn't were the ones who wrote it. So it was almost like, oh, I saw this article. I thought you might enjoy it. And then there's also no call to action in these emails. So I thought those were some reasons why this was the kind of weaker side of the emails. They're also like a bit too wordy. Again, these are things that are fixable in the prompting process, but you know, your average person, how much are they going to engage that heavily in the prompting process? And I just thought that this was kind of the weaker set, so

VIRGIL 17:14
It's interesting because you say, Well, was that perplexity or was that the Google A.I. engine, you know, the learning language model behind it? And it is kind of interesting because a lot of these tools, you can select different models and they all learn different strengths, which is kind of great, but also very frustrating because it's like, well, not only do I need to find the right tool to use and all these tools do things differently, but on top of that, I've also got to decide what language model and some of them are thinking, some of them are doing and that kind of stuff. And again, it kind of just goes back to our whole thing of the AI industry just being so convoluted for most people to understand. which what they need is that you need like a podcast series that tells you all about all the different things.

COLE 18:03
If only...

VIRGIL 18:06
Yeah, right.

COLE 18:07
But kind of moving on to Copilot, I thought that what Copilot produced for us was kind of a step up from Perplexity. It was a bit, you know, still on the stiff and overly formal side. You could still kind of tell it was AI written. But it spoke a little closer to the reader. It had kind of more of a call to action at the end. And I almost thought that, between not to get into Claude already, but between Claude and Copilot, like it's kind of two different sides of like emails that you might want to use. Like the Claude one's like super personal. It's like there's like a PS note written at the end. You know, it's talking, but it's like talking to you as if like you or as if It's like writing to like a friend. So I don't know, it's just kind of two sides of personalization, like how formal you want to get. It's kind of up to you, like, depends on who you're targeting and stuff. But yeah, I thought the Claude one was, I thought it was phenomenal, honestly. But you do run the risk of being overly personal, but that's something that's on you to work through and edit and kind of cater to your audience.

VIRGIL 19:20
Yeah. Well, and you and I were talking about it because, you know, you, I know we weren't going to talk about it during this episode, but your kind of AI thinking document that you put together right now.

COLE 19:31
Yeah.

VIRGIL 19:31
But it is very true. I mean, it's why we'll probably end up adding a prompts episode kind of towards the end and kind of extending it by one more episode because, you know, prompts are so, you know, so critical and there's so much difficulty to them. Don't want to go real in depth into it now, but overall, it is part of the process is kind of understanding that prompts and how to word them and that kind of stuff. And again, I'm not giving a lot of instructions to him. I'm giving very basic. But if you were to do this, he might sit there and say, you know, for a C-level executive, you know, hit these pain points. And for a IT person, hit these pain points or something like that. And you know, and give them a little bit more about your voice. And you kind of talked about, kind of like we created our voice document. for High Monkey, you can do that and use that as a reference that using this voice document and that, and there's some great props that you can do with that. So there's a lot of things that you can do, but overall, what it does is it makes the process much, much more complex for you to figure out, which means you're going to take more time. And the reason we're staying simple is because most people aren't going to do that in there. And you know, it's not even just if you're a large marketing team, you probably have people that have time and the effort to be able to do that. If you're a smaller company or a really small company where maybe you only have one or two people doing this, you're going to tend to take these things at face value a lot more and you're probably doing yourself a disservice. But overall, most small companies, you know, don't have a dedicated marketing person. They have a marketing slash 500 other activity type job and doing that personalization. I mean, you might understand that a little bit, Cole.

COLE 21:19
Yeah, I might. Yeah, no, but I mean, I was going to say, if you're going to use AI to fill the role of what a person will be doing, you should be very, very clear with like what that person would be doing. Right. So I feel like I've said establishing parameters like 8 times in this episode, but really that is what it's about. Like, and then using AI more as like, an electric scooter for you to accomplish your goals with your organization instead of, in that like think piece that I wrote for Virgil, I compare it to like those hover chairs of Wall-E, like you don't want to be just like kind of floating around with like AI, oh yeah, AI is doing this for me, like I'm good. No, you know.

VIRGIL 22:07
And honestly, there's a lot of hype right now about how AI is replacing jobs. And the reality, and that's why so many layoffs and that kind of stuff. And the reality, there's a lot more nuances in there. But again, I've said this before, I think there's going to be huge backlash at some point. That's honestly already starting because they're finding that AI can't quite do what people did. Process automation, things like looking at things after the fact and giving you thoughts on that, AI seems to really excel at that as long as you've kind of done the original work, you have the understanding in it giving you more of a basis to do something better. But from the standpoint of original creation, because this kind of is the same as the content creation episode, we just kind of see it lacking. It just doesn't have the ability without a lot more instructions and information from you to really kind of help you well.

COLE 23:00
Yeah, man, the content creation side of AI is a very slippery slope. And but I thought that the content creation in this episode for the emails went a lot better than the content creation for the article in episode three of this series, because we had so much kind of predetermined details that we gave the AI versus like with the content creation, it was kind of just directly based off our research document. It just kind of like regurgitated that document. And yeah, we went with the best of the three from the content creation tool testing that we did, but it still wasn't original. It wasn't, it didn't come from our our brains, we didn't really add a ton of, I mean, we did incorporate some of our brand voice, but it's just not the same as if you like are to write something yourself. There's just a lot better uses of AI than content creation, in my opinion.

VIRGIL 24:05
Right. No, totally agree. Totally agree. I think from a personalization standpoint, using it more around helping you create campaigns or using it more around looking at your analytics and coming back with some predictions about that. I think there's a lot more value in that area. That may be something we'll explore in future episodes down the road.

COLE 24:26
Just be intentional. Just be intentional.

VIRGIL 24:28
Just be intentional. Just be intentional. Exactly. Well, I think we covered that topic. So thank you everybody for joining. We look forward to seeing you in the next episode.

COLE 24:38
Yeah, thanks everyone.

VIRGIL 24:44
Just a reminder, we'll be dropping new episodes every two weeks. If you enjoyed the discussion today, we would appreciate it if you hit the like button and leave us a review or comment below. And to listen to past episodes or be notified when future episodes are released, visit our website at www.discussingstupid.com
and sign up for our e-mail updates. Not only will we share when each new episode drops, but also we'll be including a ton of good content to help you in discussing stupid in your own organization. Of course, you can also follow us on YouTube, Apple Podcasts, Spotify, or SoundCloud, or really any of the other favorite podcast platforms we might use. Thanks again for joining, and we'll see you next time.

‍