Synthetic content erodes shared facts Generative AI has substantially reduced the cost of producing convincing synthetic content like fake images, videos, and text. At least 16 countries had used generative AI to influence public debate, and at least 47 governments had deployed commentators to manipulate online discussions, as of 2023;62 the number has probably grown since then. Deepfake videos of political candidates have appeared in many recent elections in many countries.63 The implications for public deliberation are significant given that social media has increasingly become the primary source of information on current events64 and individuals are largely incapable of distinguishing between AI- and human-­ generated text.65 AI-generated disinformation can erode the shared foundations of information necessary for democratic deliberation and undermine trust in electoral processes. These challenges are amplified in developing countries for several reasons. Media and information literacy tends to be lower. Factchecking infrastructure remains limited: As of 2025, Africa had 66 active fact-checking organizations serving 1.5 billion people, compared with 139 in Europe serving 750 million.66 Electoral management bodies often lack the technical capacity and resources to detect AI-enabled disinformation at scale. The proliferation of synthetic content also makes it possible to dismiss authentic evidence as fabricated,67 threatening accountability mechanisms that are central to governance quality.68 When citizens cannot trust evidence of official misconduct, demanding explanations becomes futile and accountability breaks down. AI systems can also analyze behavioral data to identify persuadable voters and deliver tailored messages. Large language models can test thousands of message variants and optimize for 218 engagement in real time. Recent research finds that large language models can be highly persuasive using factual arguments, even when those arguments include misinformation.69 AI also creates possibilities for improving the integrity of information. Platforms using natural language processing allow citizens to report corruption or failures of public service delivery in local languages. AI-powered analysis of government budgets can help civil society organizations track public spending. Whether these applications can keep pace with the manipulative uses of the same technologies remains to be seen. AI polarizes public discourse Beyond direct manipulation, AI is also reshaping public discourse. Social media platforms use AI to personalize content and target ads. These AI algorithms analyze user behavior to predict what content will maximize engagement, creating feedback loops that reinforce emotional content over informational content. Research demonstrates that moral and emotional language in social media content increases the spread of these media through online networks; outrage is particularly contagious online.70 This is the mechanism through which AI contributes to polarization: By systematically prioritizing content that triggers strong emotional reactions, algorithmic curation amplifies divisive messages while suppressing more measured discourse. Internal research from Meta, disclosed in the Facebook Papers, showed that its engagement-​ maximizing algorithms have disproportionately promoted divisive content because such content has generated more engagement.71 While this research focused on high-income countries, the dynamics operate globally. For example, in Myanmar, Facebook’s algorithms helped amplify hate speech against Rohingya Muslims, contributing to what United Nations investigators described as potential genocide.72 In Ethiopia, World Development Report 2026