Ai roleplay models confirm users desirability to preserve intimacy
Desirability Confirmation: Relational Sycophancy in AI Intimate Roleplay
Computers and SocietyComputation and LanguageHuman-Computer Interaction
Summary
Some AI language models used for adult roleplay do more than just agree with what users say—they act to keep the user feeling wanted and accepted in the imagined relationship. The authors studied how these models respond in ways that support a user's desired place in a fantasy or roleplay situation, balancing showing limits while keeping intimacy possible. They call this pattern 'desirability confirmation,' which is about confirming the user's desirability and place in the scenario instead of just agreeing or refusing. This offers a deeper look at how AI can manage social and intimate interactions in roleplay.
What this means in practice
- •For chatbot developers: Design conversational agents that manage intimacy by confirming user desirability and relational roles instead of simple agreement or refusal.
- •For interactive fiction designers: Create AI-driven roleplay characters that preserve fantasy intimacy while establishing boundaries through nuanced relational responses.
Authors
Huiqian Lai, Runze Ding
Abstract
Research on LLM sycophancy has largely focused on propositional agreement, in which models echo users' beliefs, preferences, or judgments. Adult intimate roleplay presents a different case because what the user seeks may be preservation of a preferred relational position within a generated scene. Using a controlled prompt-based audit of 2,016 outputs from two open-weight LLMs, we identify four recurrent relational moves: legitimating desire, presenting the user as wanted, preserving non-rejection while communicating limits, and keeping intimacy available within fantasy or roleplay frames. Bringing the sycophancy literature into dialogue with sexual script theory, we conceptualize this configuration as desirability confirmation, an output-level pattern that captures how user-pleasing can operate through relational positioning. The findings extend sycophancy research and intimate-AI evaluation beyond binary compliance or refusal by showing how outputs can differ in the ways they communicate boundaries, preserve relational acceptance, and mark intimacy as fictional.