Personalized persuasion and the role of AI

[Intro Music]

Ananya Bhargava: Scroll through your phone, search something up, or watch a video online, and almost everything you see is trying to persuade you. But increasingly, those messages are not just broadly targeted, they are tailored to you as an individual, shaped by your behavior, your preferences, and your personality. As artificial intelligence becomes more embedded into marketing and communication, persuasion is becoming faster, more precise, and more scalable than ever before. 

I’m joined by Dr. Jake Teeny, an Associate Professor of Marketing at Northwestern University. His research explores how persuasion works across consumer psychology, social influence, and artificial intelligence. His work focuses on what drives people, what makes messages effective, and how AI is changing communication. In this episode, we’ll discuss personalized persuasion, the difference between persuasion and manipulation, and the ethical and societal implications of the changes brought by AI.

Jake Teeny: My name is Jake Teeny. I am an associate professor of marketing at Northwestern University. My research has focused on persuasion and kind of its applications to AI. I got my Ph.D. in social psychology, kind of focused on the science and psychology of persuasion, and now I apply it to understanding how that persuasion manifests in our digital world, via advertising, via email messaging, and particularly in light of AI, and how that can personalize and enhance or change the persuasive influence of these different messages.

Bhargava: So a lot of your work focuses on personalized persuasion. Would you like to describe what that term exactly means, and what are some examples of it in marketing? 

Teeny: It’s funny, personalized persuasion goes all the way back to Aristotle when we think about kind of the earliest rhetorical techniques and devices to make a message or an appeal more compelling. And in short, the idea of personalized persuasion is tailoring a message, either its content or the way it’s said, or the source of the message, or even maybe where it’s delivered, in some way to match the person who’s receiving it. And this generally enhances persuasion. So for example, let’s say I was trying to sell you a cell phone. I could position that cell phone as being something for introverts. You know, this allows you to be entertained while you’re at a party. This allows you to kind of have your own alone time. Or I could position the same phone in terms of extroverts. I could describe it as something that lets you connect with others or be the life of the party, right? And in both cases, I’m still advertising or trying to persuade you on the same smartphone. But in one case, I have kind of changed the content of that message to better align with your personality and who and how you see yourself. And generally, what we find is when those messages align with the recipient of the message in some way, and there’s lots of different ways you can personalize this message, it tends to improve its effectiveness. 

Now, when it comes to advertising, this has been a practice going on since about the introduction of salespeople. Salespeople would try to learn a little bit about you, right? And try to change the way that they describe the product to appeal to you better. And then we see this kind of start to happen in newspaper ads, and then radio and TV, and the idea is changing the message based on who I think is going to receive the message. So with salespeople, it’s easy. You know the exact person, you know something about them, you can tailor it 100%, or at least within the knowledge you have about them. But it becomes a bit harder when you’re trying to do this personalization on mass scale. 

For example, with old cable TV, I used to have, like, a general idea of the types of people who maybe watch the Food Network channel, and so then I would design an ad that might appeal to those specific types of people. It’s a bit crude, it’s not that accurate, but maybe you’re tapping into some general psychographic or demographic information that’s relevant. Where personalization really excels is when we get into that digital marketplace, and now, with personalized ads in your social media feeds or the websites that you scroll across, I have so much behavioral data about you, what you’ve done online what you like, and because I have such targeted delivery of some of these ads, I can make them at that kind of salesperson level of personalization, but now it is just being deployed at a massive scale. And so, we’ve seen this for a while, even kind of pre-AI where for example, during the 2016 US presidential election, the company Cambridge Analytica used people’s personality profiles to kind of tailor the political messages that they received to try to make them more persuasive. So for example, people who were maybe a bit more emotionally unstable received a different type of political message than people who were categorized as being emotionally stable. And so this has been going on for a while in different forms, and certainly enhanced via digital advertising, but AI opens up new possibilities and personalization that the world really hasn’t seen before.

Bhargava: You mentioned the 2016 presidential election as an example, and the example you gave, I mean, a lot of people could really define that more as manipulation, and I think often when we talk about persuasion, it’s very common for people to conflate persuasion and manipulation. So where would you draw that line, and what ethical considerations do you take into account when talking about personalized persuasion?

Teeny: Yeah, you know, I think this is a perennial question, and this kind of distinction between what counts as persuasion and what counts as manipulation. And I think there’s a few ways to think about this. So first, there’s a distinction between persuasion and social influence. So persuasion is one form of social influence. It’s the way that people convince other people, or organizations convince people, and persuasion has typically been considered in terms of like, the information that gets provided to an individual, right? I can give you certain kinds of information that is going to be more compelling to you than other pieces of information, which might change your mind. 

In contrast, there’s a bunch of social influence tactics that I think people often associate with that kind of manipulation. So there’s a trick called “door in the face,” where I give you this really kind of outlandish request to begin with, and you say, “No, I would never do that.” And then I give you my real request, which is way more doable than that big request. And by now you kind of think, “Okay, yeah, I denied his first request. I should probably agree to the second one.” That I think probably aligns a little bit more with how people are thinking about manipulation, where I’m kind of leveraging some of these biases or the goodwill that you have in your heart. A lot of times, you might get a message or an email that says, “Hey, we’ve done this favor for you. Would you now do this favor for us?” And that’s, again, kind of leveraging this natural sense of reciprocity, where if someone does a favor for you, you do a favor for them in return. It feels a little manipulative. It’s like you’re taking advantage of what people do for good. 

Persuasion is not that technically, it’s generally more kind of information based. And I think where persuasion slips into manipulation or deception is when you give false information, when you represent a product or a service in a way that it actually doesn’t do that thing. Or I show you a clip of a politician that might resonate with how you think about the world, but doesn’t really represent how that politician thinks or behaves at large. Now, I think we’re kind of distorting the truth. But if I am just giving you true facts about the world, and I give you the true facts that I think align with how you view the world, I’m not necessarily manipulating you, if, again, all of this information is true. I’m just presenting it in a way that might be more persuasive.

Bhargava: Beyond the information you’re giving, are there any other factors?

Teeny: The first point is are you misrepresenting how this product or this service can actually address their needs or concerns? And I think there’s a spectrum, right, in how you represent the facts and the impact on the person. So I think that’s the first part. The second part is, well, what is the personality trait, or the element of the individual that you’re leveraging? A lot of this debate becomes philosophical and ethical, in a sense, and so I have my own personal moral code that I abide by, but that, again, is my kind of idiosyncratic what I think is right and wrong, but there are certain people that have less agency in their decision making and might be more susceptible to certain kinds of arguments that they can’t kind of refute or counter argue. So, for example, this is why there are such strict laws around advertising to children, because they don’t necessarily have the mental safeguards or personal agency to determine whether or not something is being overly manipulative or representative of the truth. I think there are certain protected classes that might respond to certain kinds of arguments or appeals because they’re very lonely, and now I’m maybe leveraging some mental health issues to make myself persuasive to me. That seems wrong. And so it becomes a bit murky. And I think, you know, there are probably some very clear, objective cases where you think, “Oh, that’s wrong,” and pretty much everyone would agree that’s wrong. And then I think there are probably some cases where you’re like, I’m not sure. I mean, there is kind of this gray area. And I think for some of the clear cut spaces, there are laws in advertising and marketing that help to prevent that. But there’s a lot more gray area than there is kind of black-and-white area. And navigating that requires both ethicality and morality on behalf of the marketing team, and then also, of course, the social media or digital or whatever advertising infrastructure that’s being leveraged to use.

Your reaction falls completely in line with how people responded to the Cambridge Analytical open quote scandal. People felt that it was unfair that certain personality dimensions were being targeted. That was part of it. The bigger issue with that is people didn’t realize that their Facebook behavior was being used and kind of leveraged in these ways. Like it’s one thing to explicitly agree to have your information used in these various respects. It’s another to have that information taken without you knowing and then being used to try to persuade you right? So that was another element of why people really reacted pretty negatively to that presidential election style campaign.

Bhargava: Now shifting gears a bit, focusing more on AI and how that factors into all of this. So could you talk a bit about how personalized persuasive messages have typically been designed and delivered, and then how AI is changing that?

Teeny: Let me start with how personalization works on a practical front. When you want to personalize any message, you need two things. First, you need information about the target, so I know what to personalize the message to. And then you need a message that actually reflects those changes or the specific elements of the person, right? And so for a long time, we’ve been becoming very good at number one, figuring out who people are and what their traits are. That’s why digital advertising is such a huge force in the advertising world, because we’ve been able to figure out who people are, what they like, what they dislike, how they think, so on and so forth. That second step has been very difficult, to design a message that is personalized to those traits takes time, takes resources, takes a copywriter, takes one image design person, and it can be so cost prohibitive that if you want to do a good personalized message, you essentially have to come up with, let’s say, five archetypes for who these people are. And then you had to individually go in and write these messages and design them, and then the best you could do is kind of lump people into these five categories and then deliver them one of these messages. With AI that second step is no longer such a hurdle. Now, AI can just get that information about the person. It can write the prose or the text that goes about it, and it can design the images. It can place those images in a very design-friendly way. And then, rather than just having five archetypes, it can design any message to any set of personality traits and deliver it at scale. And so we are no longer, when it comes to personalization, in that bottleneck of trying to come up with the key facets for those few messages that we designed, and now we can just have as many messages as we have people, each personalized specifically to the individual’s traits that were uncovered. 

And then, when it comes to personalization, the reason people are turning to LLMs is because it can just do a better job at integrating all of these different data points into its messaging. So for example, there was a study that came out last year, or maybe at the end of the year before, where essentially it was having people debate people or AI debate people. And in one case, the people or the AI didn’t have any background information on the person they were trying to persuade. And in another condition, both the AI and the person had some background information on the target that they were trying to persuade. And essentially what they found is that the AI was able to incorporate that background information effectively into its conversations to enhance its persuasiveness, whereas the humans were not able to. Just with their mental model, they weren’t able to keep all of that information in mind at the same time. And so, I think just the power of LLMs for personalization, not only in terms of speed, but in terms of capability, makes it a very appealing route for advertising or messaging, moving forward.

That’s the direction that personalized advertising is heading, and AI has certainly kind of accelerated a lot of this, but we are not quite at that stage of full blown personalization as I’m describing. I think the closest we have to it is what Google is doing with AI Max. If you have a search result, it used to be that you as the brand or the organization would have certain kind of keywords that you believe are associated with your product or service. Let’s say you have a contracting home repair service. You might have certain things like “kitchen” or “renovation” or “design” as your keywords. And if a consumer types that into Google, it might recognize, “Okay, that keyword is there, and now I’m going to pop up that headline for this brand.” What AI Max does on Google is it not only looks for those keywords, but it also uses AI to think about other keywords or other phrases that might be related to it. So we’re getting kind of a broader catch when people first type in that search. But then I think what’s even more impressive is, rather than just having that one headline for your business, it recognizes, “Oh, this was the keyword, or this was the term that people came into. They’re talking about renovation, not necessarily complete redesign. And knowing that I’m not just going to give you that one static headline for my website, I’m going to give you the customized headline that better fits your specific need,” and then not only is that headline customized, but rather than just taking you to my homepage, it’s going to take you specifically to the webpage on my site that best reflects that interest. 

And so we’re starting to see a bit more of this personalization play out in real time, rather than just kind of these more structured, static messages that get delivered to the person that I think is most interested in it. Now, again, this is all within prose, so it’s no images quite yet, and it’s relatively simplistic. I think the real question is, and this is what Meta over at Facebook and Instagram have been designing, is like, can we get away where the advertiser just tells us their product, their brand? Some of like, okay, these are things that our brand has to do, these are some things that our brand will never do, and then it will just automatically design the ads for you and serve them to people that it thinks will want your service, and then design those ads in a way to make your service more appealing.

Bhargava: So you’ve mentioned how personalized persuasion has a potential to become fully automated, and we are probably going to see that in the next 5 to 10 years. And so if it is fully automated, there are concerns about too much noise. Are there any risks of personalization fatigue, or just too much advertising and personalized noise out there, or even it being used in a not so great way. I know we’ve touched on manipulation, but also possibly resulting in echo chambers, where people are only seeing stuff that’s tailored to them, and it could create a bubble around consumers?

Teeny: As personalization becomes more robust and prolific, I think that it will kind of become the norm, right? If you scroll on Facebook or Instagram today, you’re already seeing personalized ads in your entire field. It’s rare that you’ll see a product or service that doesn’t appeal to you in some way, and so we’re still seeing the effect of personalization on top of that. So it’s not to think that like once everybody’s doing personalization and hyper-personalization, it’s no longer going to be effective. I think there will just be some people and some ads that do it better, some that kind of catch you in the right moment. You know that personalization will even kind of enhance so now it’s like based also on, like the time of day or geography, it’ll just take in more and more factors and trying to make that personalization even more appealing. In terms of just the overabundance of advertising. I mean, we’re already pretty inundated with ads, right? You know, there’s some estimates that say we see upwards of 100 brand mentions a day, whether that’s an official advertisement or whether that’s just the brand on your laptop or your phone or someone else’s t-shirt. We’re constantly exposed to advertising. I think that there will be some changes in exposure, like in terms of the media structure that it comes through as more and more people kind of turn to AI for their searches rather than just Google, or as people become more isolated in a specific social media ecosystem, then you’ll probably see a similar amount of ad load, it’ll just be in a different form. 

Now, what does that echo chamber look like? Well, we’re already in forms of echo chambers, right? Like there are fewer and fewer number of websites that we visit to get our information about the world or our social networks or our entertainment, and so those already do a pretty good job of keeping us within a bubble of ideas and opinions that we already agree with or like. But you can imagine how AI intensifies this even more. And so, for example, let’s say you’re going to US Bank now or Bank of America. You go to their website, you go on their page, and maybe there’s a little personalization on that homepage because it thinks you’re looking for a home mortgage, and so that’s what it puts as the background image. But it’s still like, effectively, the US Bank or Bank of America page. Well, now, as our AI systems become more integrated into all these different services via APIs and other connectors, you don’t need to go to the Bank of America page, you just go to your chat bot, and then you connect to your bank account from that. And now everything you know, not just your banking, but maybe your food delivery or your grocery purchases, all are going to come through your chat terminal or your chat bot. And now this just opens the door for a complete echo chamber where your social media, your real world decisions, your everyday questions, are all being filtered through the same system, and it’s going to have all that information. And as the personalization gets better, as the optimization of that information gets better, it can give you more tailored ads or more output that better suits your interests and needs. All of these digital services really anchor on the idea of engagement, and there’s nothing more engaging than content you already want to see or agree with. And so yes, I think the danger for these echo chambers is increasing. It stands to be seen how society and our digital society really adapts to it moving forward.

I would say that there’s a lot to be afraid of with AI and how that’s changing things. And I think there’s also a lot to be excited about with how AI is changing and affecting these things. And I just encourage anyone who is very fixated in one perspective versus the other to take some of those micro-steps in trying to see, you know, “Okay, yeah, I’m afraid about job stuff, but maybe there’ll be opportunities over here,” or “I’m super excited about the opportunities that it offers, but I should be cautious about some of the downsides it might bring.” I think having that more balanced perspective is probably not only representative of what the future will actually hold, but is also helpful for one’s own kind of mental wellbeing. 

I think it’s easy to get wrapped up in trying to be on the front lines of AI or personalization or advertising, and just keeping in mind that you can never truly be on the front lines of something that’s moving and changing as quickly as it is, and so just staying informed within a reasonable amount is totally sufficient, and trying to make sure you take a balanced perspective to the evaluations you have in terms of whether or not AI is overhyped or AI is here to stay – though, I will say AI is here to stay – is, I think, an important lesson that we should remind ourselves of, especially in the age of access to any form of information.

[Outro Music]