When users seek social support from chatbots, they typically disclose their situation gradually. Yet most evaluations of supportive LLMs rely on single-turn, fully specified prompts. We introduce a multi-turn simulation framework to better match this interaction pattern. Drawing on support-seeking narratives from five Reddit communities, we decompose posts into ordered fragments and reveal them to a language model one turn at a time. Each response is coded with the Social Support Behavior Code (SSBC), an established multi-label taxonomy that captures the composition of support, rather than a single quality score. To test whether support strategies track the model's own representation of user distress, we use linear probes over hidden states to estimate this internal signal without altering the generation context. Across two mid-sized models (Llama-3.1-8B, OLMo-3-7B) and more than 6,200 turns, support composition shifts systematically with perceived distress: teaching declines as perceived distress rises (replicating across architectures), while increases in affective strategies are model-specific. Community context independently shapes behavior, tracking topic and discourse norms rather than demographic categories. These trajectory-level dynamics, invisible to single-turn evaluation, motivate multi-turn auditing frameworks for socially sensitive applications.

PDF