Fact-Checkers Navigating Generative AI: Practices, Boundaries, and Design Implications

The rapid and large-scale spread of misinformation poses a serious challenge for fact-checkers, whose work remains difficult to scale and is often resource-intensive. In response, fact-checking organizations are increasingly adopting generative AI to support verification work. While these tools may improve efficiency and expand capacity, their use also raises concerns about fairness, accountability, transparency, and factuality. Here, we provide an empirically grounded account of how professional fact-checkers integrate generative AI into high-stakes verification work, showing that adoption is heterogeneous, constrained by professional norms and structural conditions, and shaped by conflicts with fairness, accountability, and transparency. We present findings from semi-structured in-depth interviews with 29 professional fact-checkers from 28 organizations operating in 41 countries. Our analysis shows substantial variation in how fact-checkers engage with generative AI, ranging from interaction to the development of in-house systems and public-facing LLM-powered fact-checking chatbots. Across these practices, participants position LLMs as assistive tools embedded in human-led workflows, drawing clear boundaries around automation to preserve core fact-checking principles. At the same time, they report persistent challenges, including hallucinations, uncertainty introduced by probabilistic model outputs, difficulties in source traceability, and uneven performance across languages. Adoption is further shaped by structural constraints, including access restrictions, costs, and organizational capacity. Participants also describe shifts in the verification ecosystem as audiences increasingly rely on AI chatbots to fact-check claims directly. Grounded in fact-checkers’ perspectives, this study contributes an empirically informed account of human-AI interaction in professional fact-checking and derives user-centered design considerations.