Using AI to Tailor the News

Newsrooms have been exploring ways to incorporate AI, many have experimented with how to integrate it into the reporting process. But what about audience-facing uses? Could AI be used to tailor the news for specific audiences, or to edit reporting to increase its resonance with diverse audiences? This talk will discuss research being done at the Center for Media Engagement that speaks to these questions. The first study, drawing from research on tailoring, the machine heuristic, and communication accommodation theory (CAT), analyzes whether news tailored by AI to cater to young adult audiences affects people’s evaluations and learning. Based on a between-subjects experiment with U.S. residents (n = 1,007) that manipulates the article style (original article, AI rewrite for Gen Z, or AI rewrite for 18-to-21-year-olds), we find that tailoring the news for younger audiences has little effect on article evaluations or learning, regardless of how old participants are. In most cases, people estimate similar AI use across the original article and AI-tailored versions; however, reading an article AI-tailored for their generational cohort leads younger audiences to detect greater AI use, which indirectly lowers evaluations, decreases engagement intentions, and reduces learning from the article. The second series of studies, conducted in partnership with two newsrooms, involved evaluating the efficacy of AI in reducing perceptions of bias in the news. In these studies, each newsroom chose an article that was then annotated for bias by around 200 people from their state (evenly divided between Democrats and Republicans). The original articles were then revised in collaboration with the newsroom based on the audience feedback. At the same time, we used AI to rewrite the original articles based on a prompt asking AI to remove signs of bias. We then conducted two experiments (n ~ 2,500 in total) to evaluate how people reacted to (a) the original article, (b) the article revised based on audience feedback, and (c) the AI-rewritten article. Results suggest that both (b) and (c) improved audience perceptions. Implications of AI use for newsrooms are discussed. The Misinformation Speaker Series is co-sponsored by the Shorenstein Center for Media, Politics and Public Policy at Harvard’s Kennedy School and the Institute for Information, the Internet & Democracy (IIID) at Northeastern University.