In this study, we explore researchers’ attitudes and experiences with generative AI tools in science. To this end, we employ two methods: qualitative expert interviews and a quantitative survey. We further investigate researchers’ procedures, experiences, and the hurdles they encounter while writing through contextual inquiries. Based on these, writing phases and corresponding, suitable AI functions were identified. The findings indicate that researchers generally favor using generative AI, specifically in writing tasks, but criticize the quality of the outputs. Researchers’ writing process was found to be aligned with the cognitive processes model of writing, offering opportunities for AI to assist in distinct writing phases. Researchers generally viewed AI as a collaborative partner, entrusted with tedious tasks such as linguistic improvements, while retaining autonomy over the final product. Furthermore, fragmentation was identified during the writing process, particularly in artifacts such as notes, literature, and manuscripts. AI support could provide additional assistance in managing and aggregating these fragments, thereby enhancing the writer’s focus. Finally, we present design drafts incorporating these findings that could be used in a controlled user study at a later date.