Abstract
The growing ubiquity of conversational AI highlights the need for frameworksthat capture not only users' instrumental goals but also the situated,adaptive, and social practices through which they achieve them. Existingtaxonomies of conversational behavior either overgeneralize, remaindomain-specific, or reduce interactions to narrow dialogue functions. Toaddress this gap, we introduce the Taxonomy of User Needs and Actions (TUNA),an empirically grounded framework developed through iterative qualitativeanalysis of 1193 human-AI conversations, supplemented by theoretical review andvalidation across diverse contexts. TUNA organizes user actions into athree-level hierarchy encompassing behaviors associated with informationseeking, synthesis, procedural guidance, content creation, social interaction,and meta-conversation. By centering user agency and appropriation practices,TUNA enables multi-scale evaluation, supports policy harmonization acrossproducts, and provides a backbone for layering domain-specific taxonomies. Thiswork contributes a systematic vocabulary for describing AI use, advancing bothscholarly understanding and practical design of safer, more responsive, andmore accountable conversational systems.