Comprehensibility, Coherence, and Engagement of AI-Generated Interactive Fiction for Educational Use: Session Data and Stimuli

This dataset accompanies a pilot study on AI-generated interactive fiction for educational use in higher education. The study examined how short, choice-based learning scenarios generated from STEM content were perceived by participants with respect to comprehensibility, story-content coherence, engagement, and length.

Participants each played one browser-based interactive-fiction scenario and then completed a short post-session questionnaire. The questionnaire covered narrative clarity, coherence between story and educational content, engagement, and acceptance of the scenario length. Additional control items captured participant role, prior participation, subject-matter familiarity, and perceived English-language barrier. The session data also include derived gameplay measures such as session duration, number of selected choices, quiz attempts, wrong-answer clicks, and first-try correctness.

The archive contains anonymised session data from 22 included completed sessions, a reduced manifest for 10 excluded sessions, and the 19 interactive-fiction stimuli referenced by the included sessions. Each stimulus is provided both as a human-readable Ink source file and as the compiled story JSON file used by the runtime. Excluded sessions are documented only by event identifier, timestamp, study identifier, stimulus identifier, and exclusion reason. Participant answer columns from excluded sessions are not distributed.

The material can be used to reproduce the descriptive analyses reported in the associated paper and to inspect the generated learning scenarios themselves. It is intended for non-commercial research, teaching, replication, and methodological comparison in work on AI-generated educational content, interactive fiction, game-based learning, and technology-enhanced learning.