We argue that emotionally loaded romantic conversations should not be treated as simple text-generation tasks. In these situations, a user is not only searching for a fluent sentence; they are making a strategic communication decision under emotional pressure. We evaluate Prelude, a domain-specific conversation decision system, against ChatGPT, Claude, and Gemini across 500 emotionally loaded romantic conversation scenarios. Responses were scored using a seven-dimension weighted rubric measuring strategic fit, emotional intelligence, clarity/usability, de-escalation, specificity, conversation progress, and recurrence awareness. In the final evaluated CSV, Prelude achieved the highest average weighted score, 8.27 out of 10, compared with Claude at 7.27, ChatGPT at 5.96, and Gemini at 4.69. Prelude was the sole winner in 421 of 500 scenarios, corresponding to an 84.2% sole-win rate. These results suggest that, under this rubric-based evaluation, Prelude provided stronger strategic support than the evaluated general-purpose assistants on this 500-scenario benchmark. We do not claim that Prelude improves real relationship outcomes; rather, we show that under a structured rubric-based evaluation, Prelude performs strongly on criteria designed for relational communication support.
Mariia Yakovleva (Fri,) studied this question.