Human-AI Co-creation for Narrative Game Design: Evaluating Generative AI Tools for Iterative Narrative Design
Year: 2026
Type: dissertation
Abstract: Introduction: This thesis examines the alignment between contemporary generative AI tools and the iterative workflows of narrative game design. Narrative design in games requires continuous revision and incremental refinement as designers negotiate between story elements and evolving gameplay systems. However, many current generative AI tools operate through prompt-based, completion-oriented interaction models that prioritize finished outputs over iterative collaboration. Research Questions: This thesis addresses two research questions: 1. Which system-level heuristics can be derived from prior research to support iterative narrative design workflows in game development? 2. To what extent are the derived heuristics able to describe how publicly available generative AI tools support iterative narrative design workflows? Method: Through thematic analysis of 23 academic sources on mixedinitiative systems, creativity support tools, and iterative game design, six systems-level heuristics were derived for evaluating AI tools in narrative design contexts. These heuristics address authorial control, output fidelity, co-creative generation, initiative distribution, workflow continuity, and systems-level integration. The heuristics were then applied to evaluate three publicly available tools through heuristic evaluation: ChatGPT, Claude, and articy:draft X. Results: No tool met all six heuristics. All three met authorial control, but none met co-creative generation or initiative distribution. articy:draft X was the only tool to meet output fidelity and systems-level integration, reflecting its design as a domain-specific narrative tool with game engine integration. ChatGPT and Claude received identical results for the heuristics but showed meaningfully different behaviors in cross-task continuity, revision style, and creative initiative. These differences were captured through the qualitative descriptions that accompanied each heuristic judgment. Discussion: The findings suggest that the derived heuristics are capable of revealing meaningful differences between tools of different types and identifying where current tools fall short of supporting iterative narrative design. The study functions as a pilot study, establishing that a literaturederived heuristic framework can meaningfully describe how generative AI tools relate to the demands of narrative design practice within game design. Future work should involve practitioner validation, a broader tool selection, and a graded evaluation scale to capture finer nuances.
Source: DiVA (Stockholm University)
Author: Thorell, Alexander
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7165831622
