Behavioral Science & Policy

Online media shapes public opinion, with messages from opposing information sources deepening polarization. How content is framed and presented determines its meaning and impact on readers. To date, studies on framing have focused on finding certain phrases, topics, or ideas characterizing messages from particular sources, a strategy providing only limited results. In this study, we used a broader, agnostic approach that involves extracting the narrative from text, including its characters, plot, setting, and moral of the story, and the relationships between these elements. We translated text from tens of thousands of health documents in the language of conspiracy corpus (LOCO) into representations called semantic graphs. We then compared these graphs across documents from conspiracy and mainstream sources. We found that conspiracy media framed health through belief, emphasizing immediacy and individual impact, whereas mainstream media used scientific framing with a long-term, institutional focus. Shared words carried divergent narratives. As one example, in documents related to COVID-19, conspiracy media emphasized preventing violence rather than preventing infection (the mainstream narrative). Understanding conspiracy narratives is crucial for policymakers and media platforms seeking to curb their spread. This understanding can inform efforts to boost media literacy and reduce the number of posts peddling misinformation. We suggest using automated tools and AI as an aid in both efforts.

By rewarding engagement over accuracy, social media platforms foster the spread of misinformation. Likes and similar engagement reactions are a central form of feedback on most platforms and shape what users share. On the platforms, users quickly learn that sharing interesting, attention-grabbing content garners positive feedback, even when it is inaccurate. With repeated exposure to these social rewards, sharing interesting content—including interesting but inaccurate content—can become a habit. We review evidence for this reward-based learning and propose a simple redesign of platform rewards: adding a Trust button so users can reward accurate, reliable posts. Experimental evidence supports this approach: Users give Trusts to accurate posts more than inaccurate posts. Then, when they receive trust feedback, users increasingly share accurate content (even when less interesting) and reduce sharing of inaccurate but highly interesting posts. Because this intervention changes incentives, it is scalable, preserves user choice, and aligns with people’s stated goal of sharing accurate information. Misinformation interventions that overlook the role of social media incentives are unlikely to produce lasting results.

Psychological and neuroscience research reveals how motivational systems in the brain interact with the structure of social media platforms to enhance the propagation of misinformation. Features such as likes activate the brain’s reward system, encouraging people to share more emotionally engaging content, especially that which provokes outrage against opposing groups. False information is more likely to evoke outrage than is truthful information. Furthermore, exposure to misinformation increases its perceived credibility, and later corrections often fail to override its influence. To reduce impacts of false information, regulation of social media platforms should focus partly on the algorithms that prioritize content. Additionally, investing in media literacy interventions and grassroots efforts can bolster people’s motivation and ability to resist misleading content.

Misinformation can shape beliefs and undermine democracy, making effective mitigation essential. Generative AI (gen AI) presents risks and opportunities in this space—it can generate dubious content but also detect and counter misinformation at scale. AI’s ability to support and persuade people, facilitate discourse, and enhance media literacy underscores its potential. However, risks such as hallucinations, bias reinforcement, and manipulation highlight the need for responsible implementation.

In this paper, we explore gen AI’s role as an informer, guardian, persuader, integrator, collaborator, teacher, and playmaker, examining strengths, weaknesses, opportunities, and threats via strengths, weaknesses, opportunities, and threats (SWOT) analysis. For policymakers and technology leaders, we highlight the importance of regulations, transparency, human oversight, and AI literacy to ensure that gen AI can serve as a tool for truth rather than deception.

Gun violence, like many self-destructive behaviors (suicide, drug overdose, etc.), has proven hard to change. Conventional wisdom assumes these behaviors stem from people making deliberate, rational benefit–cost calculations, which has led to a policy focus on incentives (sticks and carrots). But most shootings are not premeditated or motivated by economic considerations; they’re in-the-moment arguments that spiral and end in tragedy because someone has a gun. In such settings, behavior seems more automatic than deliberate. Behavioral science gives us a way to understand and change that behavior. A growing body of research shows that this seemingly intractable problem is remarkably responsive to the right policies, including changes to the choice environment and even efforts to “change the chooser.”

Personalized nudging (PeN) promises greater intervention effectiveness, especially for heterogenous populations. However, developments in PeN are hindered due to a lack of conceptual clarity and high methodological variability. We present a framework for PeN to tackle these challenges. We argue that personalization is contingent on personal data availability and choice environment malleability. Applying these factors to a nudge’s content, design, and underlying mechanism, we suggest that various levels of PeN exist, from simple name changes to more technologically sophisticated adaptive approaches. These levels highlight various novel methodological considerations, which we split into theory-driven (top down) and data-driven (bottom up) approaches. Finally, we discuss how our framework supports practitioner goals and reveals future research directions.