Whitepaper

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

Published by University of Bristol, Meaning Machine

Summary

Lessons from a Large-scale Player Study This study provides new insight into the debate around the use It provides evidence of what players actually feel about Al. Not in theory, but when they experience it in a game firsthand. The study - conducted by the University of Bristol - is distinctive due to its depth. It involved 122 player sessions over 122 hours.

From the report

MEANING University of MACHINE BRISTOL What good are Al NPCs? Lessons from a Large-scale Player Study

This study provides new insight into the debate around the use of generative Al in games. It provides evidence of what players actually feel about Al. Not in theory, but when they experience it in a game firsthand. The study - conducted by the University of Bristol - is distinctive due to its depth. It involved 122 player sessions over 122 hours. It used a combination of well-validated quantitive measures, as well as in-depth thematic analysis of interviews games logs, and "think aloud" play sessions. Participants played two games featuring Al characters, both powered by Meaning Machine's technology.

The first part of the study focused on an early demo of the game "Dead Meat" (created in 2023, researched in Q1 2025). The second part of the study focused on a Dead Meat - Early Demo game called "Blood Will Out" (created in Q3 2025, researched in Q4 2025). Both games are developed by Meaning Machine, and are directed by their "Authored Al" control technology. Both games are "First Person Talkers", and share the following characteristics: murder mysteries; interrogation format; open FEARFUL text chat (ask anything); player can read character's mind. Blood Will Out Due to the time gap between the two games, they represent different generations of Meaning Machine's Al control technology. The Dead Meat demo has lower authorial direction, whereas Blood Will Out has higher authorial direction. The results here relate to the first part of the study, and to the Dead Meat demo only. The final results, including "Blood Will Out", will come later in 2026.

DISCLAIMER #1 DISCLAIMER #2 These results relate to using These are preliminary LLMs at runtime to create Al results, which form part of powered characters an ongoing study DISCLAIMER #3 DISCLAIMER #4 Final results will be The research was publicly funded, published later this year - and conducted independently by watch this space the University

D Dr Richard Cole Dr Chris Bevan Vishal Joshi Lu Han Tim Holland Senior Lecturer in Lecturer in Immersive Research Assistant PhD Student PhD Student Digital Futures Technologies

How players play games with Al-powered characters How they reason about these games in real time How they feel after playing these games.

Pages: 41
Words: 3,591
Type: Whitepaper

Topics

View this document on Game Industry Library.

Related Documents

Report

The Future of NPCs

This analysis explores the evolving role of non-playable characters (NPCs) in the video game industry, arguing that while graphical and narrative technologies have advanced, NPC interactivity has remained stagnant. The central thesis posits that integrating advanced AI—specifically large language models and machine learning—is essential for maintaining player immersion and driving commercial growth. By moving away from static dialogue trees toward dynamic, situational awareness, developers can address long-standing player frustrations regarding repetitive behaviors and scripted limitations. Key findings indicate a significant market demand for sophisticated AI integration. Data shows that 84% of gamers view NPCs as vital to their experience, yet 52% are frustrated by repetitive dialogue. The transition to advanced AI NPCs appears to offer a clear financial incentive: 81% of players expressed a willingness to pay more for games featuring intelligent characters, and 79% stated they would be more likely to purchase a title with such features. Furthermore, 88% of respondents believe advanced AI would improve overall gameplay and immersion, suggesting that these technologies could increase player retention and session length. The scope of the research focuses on 1,002 U.S.-based gamers aged 16 to 50 who play at least five to eight hours per week across PC, console, mobile, and VR platforms. Methodology involved a 20-minute survey conducted by Bryter Research, which included a demonstration of generative AI NPC capabilities to gauge participant reactions. The findings suggest that while role-playing and sandbox genres stand to benefit most, there is a near-unanimous consensus (99%) among surveyed gamers that advanced AI NPCs represent a positive value add for the industry.

ai player-behavior game-design+1

Inworld AIJan 2023

Report

The Future of Consumer Apps: How AI and Game Design Principles Are Reshaping Every Category

H O W A I + G A M E D E S IG N THEWORLD OF CONSUMER TECH IS CHANGING 1990 2007 2023 2026 INTERNET MOBILE AI APPS AI APPS ERA ERA 1 .0 2 .0 The web goes Computing moves Chat w rappers. A A gents, interfaces, public. Everything into the pocket. new wave of and entirely new gets connected.

market-analysis ai game-design+2

BITKRAFT VenturesMar 2026

Report

Why Chance-Based Mechanics Keep Us Hooked

Chance-based mechanics drive player engagement by prioritizing the psychological thrill of anticipation over the actual value of rewards. This engagement is rooted in the release of dopamine during the period of uncertainty, where the wait for a result creates more neurological stimulation than the prize itself. By utilizing unpredictable reward schedules and the "near-miss" effect, developers foster a persistent belief that a significant win is imminent. This strategy is exemplified by the commercial success of Monopoly GO!, which generates between $100 million and $125 million in monthly revenue through a "saw-tooth" gameplay loop that oscillates between resource depletion and sudden, event-driven recovery. The effectiveness of these systems relies on a "pressure and release" cycle designed to maintain emotional tension without causing player burnout. High-volatility mechanics, such as digital wheels and randomized heists, are tuned to prioritize emotional impact over mathematical fairness. For instance, probabilities are often manipulated to limit low-tier prizes—sometimes to as little as 13%—while visually emphasizing jackpots to maximize excitement. Even traditionally negative outcomes are reframed as positive opportunities; in certain high-performing titles, escape rates from penalty mechanics like "Jail" are set as high as 80% to ensure the player remains within the rewarding flow of the game. Ultimately, long-term retention is achieved through the careful management of sensory-rich animations and gacha-style collection systems that create frequent "emotional spikes." By blending live events with boosters that temporarily alter the odds, developers create a dynamic environment where the player feels a constant sense of progression. This sophisticated orchestration of risk, hope, and visual feedback ensures that the psychological journey toward a potential reward remains compelling enough to sustain high levels of monetization and daily active usage across the mobile gaming landscape.

monetization player-behavior game-design+1

Sensor TowerJan 2026

Report

Annual Review: 2025

As one era of the UK When we set out our ambition to supercharge the UK video games industry at the beginning of last year, we could never games industry comes to have predicted how the industry would rise to the challenge! an end, a new and even Throughout this year, we have been proud and emerging markets across the Middle to be the leading trade body supporting East, Asia Pacific and Europe.

market-analysis game-development employment+2

UkieDec 2025