Cognitive Intervention and Reconciliation: NPC Believability in Single-Player RPGs

NPCs in single-player RPGs are often experienced as mechanical and limited, a shortfall that erodes player immersion and enjoyment. This work addresses that problem by confronting the lack of a unified definition of "believability" across AI, communication, film theory, and game studies. Rather than adopting any one discipline's framing, it synthesizes these literatures to propose a working definition and to examine the theoretical foundations that govern how players perceive NPCs, including suspension of disbelief, the "illusion of life," and cognitive illusion. These qualities are organized into five categories, with particular attention to appearance, behavior, personality, goals, and emotion.

A central finding is that these qualities function as precursors that shape one another, and that they must be displayed clearly and consistently to players rather than assumed. Appearance that is more expressive than an NPC's internal models produces false impressions, a risk illustrated by the Eliza and Tale-Spin effects. Players also respond automatically to social cues, consistent with the "media equation," so social relationships among characters and between NPCs and the player strengthen perceived believability. Yet automatic social reaction and realism judgment can contradict each other, as the uncanny valley demonstrates, and schema theory explains why simpler characters are sometimes judged more believable than highly detailed ones.

The synthesis concludes that believability arises from the cognitive gap between players' expectations and their actual experience. Distinct qualities can each establish believability, but they are not additive and may conflict. Six design principles follow, including the tendency of players to assume NPCs resemble themselves, the principle that more detail is not always better, and the discomfort people feel with ambiguity. The authors note a significant limitation: the relationship between believability and overall gameplay quality remains unstudied. The reference base spans 1950 to 2013 and draws on foundational work on believable agents, player-character engagement, presence, and social responses to computers.