<p>The increasing prevalence of mental health issues has created a significant gap between the demand for mental health services and available resources, resulting in prolonged wait times for therapy. Virtual agents capable of conducting Motivational Interviews (MI) offer a promising solution by providing immediate support and interventions during the waiting period. Our aim is to build such an agent, a task which first requires analyzing human-human data so we can build a computational model to drive the agent’s verbal and nonverbal behaviors. This study presents the Empathic Multimodal Motivational Interviews (EMMI) features set, composed of multimodal features extracted from two existing MI corpora, AnnoMI and the Motivational Interviewing Dataset (MID). Our analyses focus on therapists’ and patients’ verbal and nonverbal behaviors to understand their interactions better. We identify three distinct patient types, “Open to change,” “Resistant to change,” and “Receptive”, and examine how therapists adapt their behaviors to their clients. Our findings reveal that therapists adjust their verbal and non-verbal behavior based on patients’ types. This study highlights the importance of considering verbal and nonverbal cues when developing virtual agents for MI, emphasizing the need for these agents to adapt dynamically to different patients’ types and behaviors. Our insights provide a foundation for creating adaptive virtual MI interviewers to personalize interventions, thus enhancing their effectiveness and supporting mental health care delivery.</p>