Imagine scrolling through a social-media landscape curated just for you—subtly and perfectly shaped to your tastes, speech patterns, and core values.
That kind of hyper-personalized future is not here yet, says Jake Teeny, an associate professor of marketing at Kellogg. But as generative AI supercharges companies’ ability to use their customers’ personal information to craft targeted messages, it’s becoming more realistic.
“I think it’s going to change how we interact with the web and the world at large,” Teeny says.
Tailoring a message to its receiver’s personality, preferences, and needs—what psychologists refer to as “personalized persuasion”—dates as far back as Aristotle. Today, this targeting is central not only to marketing and branding campaigns but all manner of political and social initiatives as well.
Now, advances in generative AI are supercharging this approach, helping companies figure out what matters to the people they seek to persuade and then use that information to create custom content. What was once costly and time-consuming for people to perform has become cheap and relatively quick with AI tools.
“AI has changed the game,” Teeny says. “It has essentially allowed for the industrialization of influence, because we can [potentially] take a single person and give them a message as precisely tailored to them as if I were an individual seller with all the information possible.”
With tech companies ratcheting up AI performance, Teeny and colleague Sandra Matz of Columbia University created a framework to help us better understand the ways generative AI is being used for personalized persuasion—and the gaps in knowledge that remain.
“This framework gives both researchers and practitioners a way to more adequately direct [their attention] toward the factors that are going to matter most,” Teeny says.
Defining a new field
If there’s one common theme of existing research on using generative AI for personalized persuasion, it’s that individualized messages perform better than generic ones.
But these studies often differ in key aspects of their design, such as whether they target a person based on their demographics, personality, or morals.
Teeny and Matz’s framework helps make sense of this research by establishing four categories for research on AI and personalization.
The first category covers studies focused on how to gather personal information. This may range from surveys about how people see themselves to data drawn from people’s digital footprints to AI tools that combine self-reported information with online activity.
Of note, chatbots may provide a new stream for this kind of data. People often write to an AI chatbot as though communicating with a personal confidant, with chat histories serving as a window into their private thoughts and feelings, Teeny says. A chatbot could even prompt a person to reveal information that might be useful in targeting them with ads or messages.
Next is the category about which different types of personal information are used in personalization. Demographic information, for instance, may be too broad to be very useful in personalized persuasion, while behavioral information provides insight into an individual’s actions, such as which ads they click on or what they write in social-media posts.
Insight in your inbox
Receive our newsletters to keep up with the latest research and ideas from faculty at the Kellogg School of Management.
https://insight.kellogg.northwestern.edu/newsletter/insightful-leader
AI tools could use psychological information, too, including people’s beliefs, values, political ideology, and personality traits. Research suggests that this more-precise information on an individual’s values and preferences will lead to more effective personalization. Teeny gives the example of how a traditional advertiser might target a science-fiction fan with a commercial for a fantasy film, thinking the genres are similar. But some viewers might have a personal distaste for the dragons and elves common to fantasy. If generative AI can identify these nuances, it could help advertisers better target their audience.
Honing the delivery
The third category involves translating that personal information into strategies for personalization.
So far, most research about personalized persuasion has focused on messages targeting one personality trait, like only looking at an individual’s openness to new experiences. Generative AI can work with a much larger amount of information than human researchers can, which presents an opportunity to try tuning messages to multiple traits.
Perhaps marketers could use personal information to create an ad that successfully appeals to complementary traits, such as a person’s tendency to consider negative outcomes and to prioritize preventing harm in life. At the same time, too much personalization could backfire if a person feels it’s invasive or creepy.
The final category is for studies that consider how these personalized messages are delivered. This includes the medium such as audio, text, or video. It also includes whether messages are interactive and whether they are repeated.
AI chatbots could be responsive to a person’s remarks and tweak the tone or even the arguments themselves. “[In contrast,] humans even in one-on-one conversations would have a difficult time adjusting on the fly,” Teeny says.
Teeny believes that all the insights gathered from using the team’s new framework could help reveal best practices and key factors in using generative AI for personalized persuasion.
Ethical questions
But Teeny and Matz’s study also highlights ethical concerns about using AI tools to hyper-personalize messages. AI-enabled personalized persuasion could limit human choice, isolate people from each other, and potentially expose people to misinformation curated to their preferences.
With generative AI, Teeny envisions the possibility of a precarious future in which an AI-curated browser exclusively shows content it has determined a person wants to see. A product search, for instance, could turn up just a few options out of the hundreds available. And the framing—how those options are presented—would also be controlled by AI. It may, for instance, present a sponsored option and use its most-convincing tactics to sweeten its appeal.
In that scenario, AI is taking full control while giving you the illusion of choice, Teeny says. “And at the highest level, it brings up this debate about autonomy and free will … and about how much your choices are your own.”
These questions are particularly worrisome to Teeny because of how large-language models play into the very human desire to be understood. “It’s like a tireless therapist that is there to reaffirm your own interests and beliefs,” Teeny says.
That allows AI to build trust and make recommendations that some people may follow without questioning.
The implications extend beyond shopping to politics and community life. A politician may have made hundreds of statements in reality, but an AI tool could deliver to the masses only the statement most aligned with their values.
Preparing for persuasion
For now, AI is more useful for targeting people with advertisements than personalizing messages. “But if AI continues along even a similar path and speed as we’re seeing now, then this becomes less of a Black Mirror episode and more of reality,” Teeny says.
Researchers, tech companies, governments, and individuals all have a role to play in gearing up for the potential flood of personalized content. Researchers need to investigate which interventions would be most effective in protecting people from undue influence, Teeny says. That could include raising people’s guard by warning them of arguments a persuader might use.
Regulation could potentially help, but Teeny says that countries have been wary of placing limits on generative AI for fear of stymying innovation. “For the big tech companies, I think they have to actually hold themselves accountable and do things that are less profit-driven in order to help protect people,” he says. “That’s obviously a pretty tall order.” Instead, Teeny predicts that what will emerge are third-party apps that protect people by analyzing posts and issuing warnings about tailored content.
But in a world where content is manufactured for people and its truth uncertain, it’s really everyone’s responsibility to be more skeptical. “It’s about having a constant awareness that if something sounds too good to be true, it probably is,” Teeny says.