Effective emotional-support dialogue requires more than fluent empathy: responses must adapt to the support seeker’s evolving state. Existing emotional-support systems are largely optimized at the response level and may therefore learn the form of supportive counseling without reliably adapting to client-side change. We introduce ARIA, Affect-Responsive Interaction Alignment, a client-centered alignment framework that constructs DPO preference pairs from the support seeker’s subsequent multimodal affective trajectories observed in human interactions. ARIA distills audio- and visual-derived valence trajectories into text-based predictors and uses cross-modal consensus and uncertainty-aware filtering to identify high-confidence preferences associated with delayed client-side valence improvement. Experiments show that ARIA improves process-oriented measures of client progress beyond surface compliance and produces support that is more timely and calibrated to the seeker’s evolving openness. Behavioral analyses further show a phase-sensitive shift from restrained pre-open support toward more active support after openness emerges. Human and counseling-expert evaluations provide evidence that ARIA’s responses are more timely, targeted, and supportive of the client’s own process.

Large Language Models (LLMs) are increasingly deployed in socially complex, interaction-driven tasks, yet their ability to mirror human behavior in emotionally and strategically complex contexts remains underexplored. This study assesses the behavioral alignment of personality-prompted LLMs in adversarial dispute resolution by simulating multi-turn conflict dialogues that incorporate negotiation. Each LLM is guided by a matched Five-Factor personality profile to control for individual variation and enhance realism. We evaluate alignment across three dimensions: linguistic style, emotional expression (e.g., anger dynamics), and strategic behavior. GPT-4.1 achieves the closest alignment with humans in linguistic style and emotional dynamics, while Claude-3.7-Sonnet best reflects strategic behavior. Nonetheless, substantial alignment gaps persist. Our findings establish a benchmark for alignment between LLMs and humans in socially complex interactions, underscoring both the promise and the limitations of personality conditioning in dialogue modeling.