The Risks of AI-Generated, Hyper-Personalized Digital Advertisements

Generative AI is set to transform digital advertising, which remains a revenue juggernaut for much of the internet. In the current model, advertisers use personal data to target users with pre-made content. But with generative AI systems, they will soon be able to create novel advertisements, tailored in real time using individual behavioral and demographic profiles. Early studies suggest that these AI-generated, hyper-personalized ads are significantly more persuasive than traditional ones---even with today’s relatively limited models. As the technology advances, such ads are likely to become the dominant form of digital persuasion. I argue that this development threatens to undermine longstanding constraints that help keep advertising roughly honest. In particular, generative systems can learn to exploit users’ epistemic blind spots---areas where they are especially prone to believe false or misleading claims. While this poses familiar risks in commercial contexts, the stakes are even higher in political advertising, where personalized generative content may produce fragmented "political underworlds" and new forms of targeted propagandizing. I conclude by assessing both technological and regulatory options for mitigating these risks.