What Good Are AI NPCs?: Lessons from a Large-scale Player Study

The study demonstrates that generative AI‑driven non‑player characters can deliver deeply engaging, emotionally resonant gameplay. In a 122‑hour experiment with 68 participants, the “Dead Meat” demo achieved high immersion scores—97 % UES reward and 94 % focused attention—while keeping mental demand low (NASA‑TLX scores of 64.7 for demand and 52.7 for performance). Qualitative interviews consistently cited the NPCs’ human‑like dialogue and narrative depth as key contributors to player enjoyment.

Quantitative data confirm widespread satisfaction: 96 % of players rated overall enjoyment as high, and 90 % praised the creative freedom afforded by the open‑ended design. Subscale analysis of the GUESS instrument revealed that 60 % achieved a top score for Creative Freedom, 65 % for Personal Gratification, and 80 % for Play Engrossment. Thematic coding identified freedom of expression, challenge‑driven motivation, and immersive conversation as primary drivers of satisfaction, indicating that the game successfully balances agency with sufficient guidance.

Player behavior analysis uncovered seven distinct strategic approaches—such as “Good Cop/Bad Cop” interrogation, “Rule Bender End Justifies the Means,” and “Smart Arse” manipulation—often combined within a single session. Participants responded equally to voiced and text‑based NPCs, and the 20‑minute session length encouraged replayability through role‑playing different characters. Although the brief duration limited long‑term insight, emergent strategies were viewed as a feature rather than a flaw. Future research will explore how authorial adjustments influence player responses across demographic groups, reinforcing the potential of AI NPCs to enrich narrative gameplay on a broad scale.