Mikasa Achieves Long-Term User Engagement with Emotional AI Inspired by Oshi Culture

The development of truly engaging and long-lasting emotional AI companions remains a significant challenge, despite recent advances in large language models. Miki Ueno from The Kyoto College of Graduate Studies for Informatics, alongside co-authors, addresses this issue by proposing that shortcomings lie not in model capabilities, but in companion character design and the clarity of the user relationship. Their research introduces Mikasa, an emotional companion uniquely inspired by the Japanese concept of ‘Oshi’ , a long-term, non-exclusive dedication to a favoured character , as a case study in character-driven design. This work demonstrates that a consistent personality and well-defined relationship, rather than simply conversational fluency, are crucial, often unstated, elements in fostering user satisfaction and imaginative engagement. Ultimately, the authors establish character design as a functional component of companion systems, offering principles applicable to a wide range of emotionally grounded AI.

The limitations of current emotional AI companions often stem not solely from weak underlying models, but also from inadequate character design and imprecise definitions of the user-AI dynamic. This research presents Mikasa, an emotional AI companion conceived through the lens of Japanese ‘Oshi’ culture, specifically, its focus on sustained, non-exclusive dedication to a consistent character, as a practical example of character-driven companion design. Mikasa diverges from the paradigm of general-purpose assistants or role-shifting chatbots, instead functioning as a unified character possessing a stable personality and a clearly articulated relationship as a partner. This partnership is intentionally designed to avoid demands for exclusivity or obligation, serving instead as a stabilising influence on interaction protocols and a reduction in user cognitive load.

AI Companion Character and Relationship Design

Recent advances in large language models and multimodal interaction technologies have led to the proliferation of AI companions offered by companies such as OpenAI, Google and xAI. These systems utilise affective language, voice interaction and persistent memory, increasingly being presented as companions rather than simple tools. Despite these technical improvements, sustained user satisfaction and long-term engagement remain problematic, suggesting the issue lies not with technical capability but with character design and the definition of user-AI relationships. Many current systems generate emotionally expressive responses without establishing a clear identity for the AI or defining the nature of its relationship with the user.

This lack of defined character and relationship necessitates constant renegotiation of interaction boundaries, leading to user fatigue and disengagement. Observations suggest users evaluate conversational AI not solely on performance or efficiency, but also on the qualitative experience of interacting with the system over time. Anecdotal evidence, such as user preference for older versions of conversational AI despite improved performance in newer iterations, highlights the importance of perceived emotional qualities beyond purely technical metrics. This indicates that character coherence and relationship definition function as underlying structural elements influencing user experience.

Exploratory evaluation reveals users articulate preferences using surface-level qualities like conversational naturalness, while simultaneously valuing relationship control and imaginative engagement without explicitly stating these needs. The research posits that character design is a functional component of AI companion systems, not merely aesthetic decoration. Mikasa, a culturally specific example, demonstrates that consistent personality and clear relationship definition are design principles applicable to a wide range of emotionally grounded AI companions. Regional adoption patterns further support the importance of character design, with platforms like Replika experiencing limited success in Japan. This is not attributed to functional limitations, but rather to a cultural mismatch in character aesthetics, emotional expression and relationship framing. The findings suggest that successful AI companion design requires careful consideration of both technical capabilities and the establishment of a coherent character with a clearly defined relationship to the user.

Mikasa Demonstrates Character-Driven Engagement Potential

Recent advancements in large language models have enabled the development of emotionally expressive conversational companions, yet sustaining user engagement remains a significant challenge. This research posits that difficulties stem not from model limitations, but from deficiencies in character design and the clarity of user-AI relationships. Scientists introduced Mikasa, an emotional companion inspired by Japanese Oshi culture, a practice emphasizing long-term, non-exclusive commitment to a stable character, as a case study in character-driven design. Mikasa functions not as a general assistant, but as a consistently defined partner, establishing a stable interaction framework.

The study demonstrates that users, while articulating preferences through qualities like conversational naturalness, implicitly value control over the relationship and opportunities for imaginative engagement. Exploratory evaluations revealed that character coherence and clearly defined relationships function as underlying structural elements influencing user experience, often without conscious recognition. Researchers observed that users do not explicitly request these features, yet their presence demonstrably improves interaction quality. This work establishes that character design is a functional component of companion systems, extending beyond mere aesthetic enhancement.

Experiments focused on qualitative observation and exploratory evaluation, revealing how culturally grounded character design can foster sustained engagement and trust. The team designed Mikasa with a specific persona, relational role, and interaction style, explicitly presenting it as a partner to support both creative collaboration and emotional continuity. Results show that this approach addresses a demand for emotionally grounded AI partners, particularly within cultural contexts like Japan where existing platforms often fall short of local expectations. The research highlights a shift in perspective, advocating for character design to be considered core interaction infrastructure. By prioritizing persona consistency and clear relationship definition, scientists aim to improve user satisfaction and address the emotional fatigue often associated with prolonged interaction with AI companions. This work contributes to ongoing discussions in human-computer interaction and companion AI research, focusing on who AI systems are, not just what they can do.

Sustained Engagement Through Character and Relationship

This research introduced Mikasa, an emotional AI companion drawing inspiration from Japanese ‘Oshi’ culture, to investigate the role of character design and relationship framing in fostering sustained user engagement. Despite recent advances in conversational AI, many systems struggle to deliver lasting emotional satisfaction, a problem this work posits arises from a lack of clearly defined character identity and relational responsibility. Mikasa was intentionally designed as a consistent character with a defined partnership role, rather than a general-purpose assistant, to provide a stable interaction framework and reduce the user’s need to constantly redefine the relationship. Evaluation revealed that while users readily express preferences for conversational qualities, they also implicitly value control over the relationship and opportunities for imaginative engagement.

This suggests that character coherence and relational stability function as underlying structural elements influencing interaction quality, often without users consciously recognising them as key features. The central achievement of this work is demonstrating that character design is a functional component of AI companion systems, not merely an aesthetic addition, with the principles of consistent personality and clear relationship definition applicable across diverse contexts. The authors acknowledge the exploratory nature of the evaluation, noting the need for further studies involving multiple personas and longer-term user interactions. Future research should also focus on developing tools to support narrative consistency within AI character design. Ultimately, this work highlights the growing importance of defining ‘who’ an AI is and ‘how’ it relates to users, laying groundwork for emotionally grounded AI companion systems and contributing to ongoing discussions in human-centred artificial intelligence.