Journalism has two main functions in democracy: news coverage and the formation of public opinion. This essay develops a theoretical framework for the role of journalism as an epistemic and evaluative intermediary. This framework can be used to discuss the two-part question: how can AI improve the epistemic process of news production, and how can it strengthen the evaluative process of public opinion formation? Veristic social epistemology emphasizes the role of organized skepticism in the process of truth-seeking. One crucial weakness of AI is that it does not meet the necessary condition for collective scrutiny, namely, transparency regarding the process through which it generates truth claims. Consequently, AI can only support the human search for truth. Only a few studies have measured the epistemic quality of AI-generated content and its effects directly and systematically. With regard to public opinion formation, there is a theoretical gap that can be addressed using normative theories of democracy. Values can be derived from these theories, which can serve as yardsticks for democratic opinion formation. Empirical studies also paint a mixed picture. The risks include radicalization and confirmation bias. However, generative AI can also improve the quality of deliberation and enhance the communication skills of disadvantaged groups. The rapid development of AI models requires continuous monitoring and an experimental approach to understand the potential uses and limitations of AI in journalism and the democratic public sphere.