This paper examines how generative AI tools can support civil society organizations through a study of academic consulting engagements with five Boston-area nonprofits from January to April 2025. These academic consulting sprints, conducted directly by 15 Northeastern University graduate students in a course setting and the authors of this report, were focused on communications and media work and used an action research approach. A central finding is the concept of" organizational legibility to AI"—the challenge that civil society organizations' unique contexts, values, and relationship-based communication strategies are not immediately interpretable by AI systems, requiring significant human mediation to make organizational realities comprehensible to these tools. Key takeaways include that AI tools became more effective only after consulting teams gained extensive domain knowledge about each organization, that stakeholders sometimes overestimated AI capabilities, and that hybrid approaches combining traditional methods with AI assistance worked best for community-facing work.