Exploring the Adoption of Generative AI in Geoscientific Workflows at INGV: A Structured Qualitative Case Study from the GdL AI Working Group

Generative Artificial Intelligence is rapidly entering scientific practice and poses both an opportunity and a challenge for research institutions. Alongside the rapid diffusion of new tools, researchers are increasingly required to understand where these systems can provide real support, where their limits remain substantial, and under which conditions they can be used responsibly. Within this context, in May 2024 the Istituto Nazionale di Geofisica e Vulcanologia (INGV) established a working group on Generative AI (GdL AI) to explore its possible role in geoscientific workflows. This paper presents a structured account of that experience which lasted one year. The initiative examined a set of representative use cases, including macroseismic interpretation, extraction of information from historical earthquake documents, natural‑language interaction with seismic web services, and AI‑assisted scientific coding for seismological processing. The results suggest that Generative AI can already provide useful support in selected tasks, especially when it helps transform unstructured material into structured intermediate outputs, accelerates access to technical infrastructures, or shortens development and code‑revision cycles in scientific software. At the same time, the experience consistently showed that these systems do not remove the need for expert judgment. Their outputs remained scientifically useful only when embedded in a human‑in‑the‑loop process of validation, correction, and contextual interpretation. More broadly, the INGV case indicates that institutional adoption depends not only on access to tools, but also on training, privacy‑aware governance, and shared practice. These findings support a cautious but constructive conclusion: Generative AI can strengthen selected geoscientific workflows when its use is documented, critically supervised, and integrated into a broader institutional framework of responsible experimentation.