Introduction: Gamification and serious games are increasingly used to enhance cognitive, motivational, and affective outcomes in educational contexts. While meta-analytic evidence supports the overall effectiveness of gamified learning compared to traditional learning methods [1], empirical findings also reveal considerable variability [2]. This indicates that not all game elements equally support learning and that poorly designed gamification may even impair performance. One promising yet underexplored element in this context is the narrative or storytelling. While narratives are fundamental to how humans construct meaning and make sense of the world, they are still less frequently examined in gamified learning environments compared to other game elements [3].

Integrating narrative elements may evoke narrative transportation, a psychological state, defined as a distinct mental process involving an integrative melding of attention, mental imagery, and affect, caused by being absorbed into the story. Previous research shows that higher levels of narrative transportation are associated with improved cognitive and affective outcomes and are closely related to flow experiences [4,5]. These findings highlight the potential relevance of narrative design as a mechanism for fostering engagement and supporting learning in gamified educational environments.

The goal of this study is to explore whether adding a more elaborated narrative frame to an existing gamified learning task can enhance learning performance, motivation, narrative transportation, and flow. The current study builds on findings from a previous pilot study that suggested medium-sized positive associations between narrative transportation, flow, and cognitive as well as motivational outcomes. A key change in the current study is the inclusion of a deeper, more detailed narrative compared to the pilot study, designed to more strongly engage participants in the story.

Methods: The study uses an associative learning task, in which participants have to learn twenty symbol-number associations across five levels. In each trial, they select a number for a symbol and receive corrective feedback. The purpose of the first level is to introduce all symbol-number associations as participants have no prior knowledge. In subsequent levels, the same symbols were presented in random order, allowing participants to gradually learn and recall the correct associations through repeated feedback. Participants are randomly assigned to one of two versions of the same task differing only in the depth of the narrative frame (between-subject factor: low vs. high-narrative condition) provided. Both versions incorporate three key game elements: an enhanced visual aesthetics (e.g., an outdoor nature scene), a narrative about a dog searching for bones presented in brief written instructions at the beginning of the task, and an incentive system (e.g., tallying the number of bones found). In the high-narrative version, the task is additionally accompanied by a more detailed story presented as short comics at the start, between levels, and at the end.

Learning performance is assessed as learning efficacy and learning efficiency. Learning efficacy is operationalized as the number of correct responses in the final level, whereas learning efficiency is defined as the sum of correct responses across levels 2 to 5. In addition, motivation, narrative transportation, and flow experience are assessed using standardized self-report scales.

Results and Discussion: Online data collection is ongoing and is expected to be completed by June 2026 with an expected sample of 200 adults based on the effect sizes of the pilot study and a corresponding power analysis. Statistical analyses will, among other things, include multiple regression and mediation analyses. Based on prior findings in a pilot study and theory, we hypothesize that high narrative condition will increase narrative transportation and flow, which are expected to support both learning performance and motivation.

Educational Significance of the research: Gamified learning has been shown to enhance cognitive, affective, and motivational outcomes compared to traditional instructional approaches. However, not all gamification is equally effective, and the mechanisms underlying successful gamified learning are still not completely understood. This study explores the role of narratives in the design and implementation of gamified learning processes and serious games. By identifying how and why certain game elements, such as narratives, promote deeper involvement and better learning, this research provides practical insights for designing more effective gamified educational experiences. Understanding these mechanisms is essential for leveraging the full potential of gamified learning beyond points and badges.

Introduction

Serious games can make media manipulation, social influence, and decision-making under uncertainty experientially accessible (Roozenbeek & van der Linden, 2019). Yet the learning gain depends strongly on debriefing, because players must translate gameplay actions into explicit concepts, criteria, and transferable reasoning (Lederman, 1992; Crookall, 2009). At the same time, generative AI is increasingly present in educational settings (Kasneci et al., 2023), creating a double challenge: learners need to reflect on manipulation and misinformation (Buckingham, 2019), and they need to interact with generative AI critically, without outsourcing judgment (Long & Magerko, 2020; Ng et al., 2021). This contribution proposes a teaching and learning design that combines a serious game experience with a Socratic AI debriefing and reports on its empirical pilot in a university course. The AI scaffolds reflection through targeted questioning that prompts justification, counterfactual thinking, and attention to evidence and values (Paul & Elder, 2007), rather than delivering answers. The aim is to strengthen critical media literacy and AI literacy simultaneously by making both the game mechanics and the AI interaction itself an object of analysis. Two research questions guide the study: (1) How does a Socratic AI debriefing protocol shape the quality and depth of learners' reflective reasoning after a serious game? (2) How do learners perceive and critically evaluate the AI as an interlocutor in the debriefing process?

Methods

The study follows a design-based research approach (Design-Based Research Collective, 2003) and develops a design pattern for a three-phase intervention: gameplay, human-led micro-debrief, and AI-supported Socratic debriefing. After a brief serious game that simulates manipulation strategies, learners document key decisions using a structured log (actions taken, perceived incentives, emotions, perceived social pressure, and perceived success conditions). A short human-led micro-debrief establishes shared vocabulary and norms. Learners then engage with a large language model in a constrained Socratic coach role. The interaction follows a prompt protocol that enforces questioning and prohibits direct solutions, cycling through four moves: eliciting reasons, testing assumptions, exploring alternatives, and linking to transferable criteria. The protocol includes a model critique step where learners identify overgeneralizations, missing evidence, and normative drift in the AI questions. The paper specifies reusable artifacts: a decision log template, a Socratic prompt script, a rubric for quality of justification, and safety and ethics constraints for classroom use. The design is implemented and evaluated in a university course during the summer semester 2026. Data collection includes decision logs, AI interaction transcripts, reflective writing, and a post-session questionnaire on perceived learning, critical incidents, and attitudes toward AI as a learning partner. Empirical results will be available for presentation at the conference.

Results and Discussion

The paper presents the design rationale together with empirical findings from the university pilot. On the conceptual level, the Socratic AI debriefing adds value beyond conventional debriefing in three ways. First, it increases the density of reflective prompts and ensures that every learner receives sustained questioning, even in large groups. Second, it foregrounds epistemic practices central to resilience against manipulation: distinguishing claims from evidence, seeking alternative explanations, and articulating criteria for trust and credibility. Third, it creates a learning situation in which the AI is not a neutral authority but a fallible interlocutor, supporting metacognitive awareness of how language models can steer attention, simplify complex causality, or implicitly moralize. The contribution discusses key risks and design safeguards. Risks include shallow compliance, reification of the AI as evaluator, and the possibility that AI questions reproduce bias or culturally narrow assumptions. Safeguards include strict role instructions, transparency about limitations, triangulation with human discussion, and the requirement that learners ground statements in their gameplay log and external sources.

Educational Significance

The design pattern contributes a concrete, transferable framework for integrating serious games and generative AI in ethically and pedagogically responsible ways. It supports critical media literacy by making manipulation strategies and incentive structures explicit, and it supports AI literacy by modeling disciplined, skeptical interaction with a language model. The empirical pilot provides evidence on feasibility and learner response, strengthening the design's practical relevance. The contribution discusses materials and decision points for adaptation, including assessment options that prioritize reasoning processes and reflective justification over product correctness.

1 Introduction

Lasting knowledge is pivotal for students' academic success and their ability to participate in today's knowledge society. However, recent large-scale assessments indicated that students often struggle to acquire and retain lasting knowledge. Generative activities facilitate the construction of coherent representations by prompting students to select, organize, and integrate information, which fosters (meta-)cognitive learning. Contrarily, retrieval practice consolidates knowledge and supports long-term retention. Given their distinct functions, generative and retrieval activities may complement each other, but evidence on their combined effects is limited. How such combinations affect learning in authentic school settings—immediately and lastingly—remains open.

Against this background and following an interdisciplinary approach integrating educational psychology, instructional design, and subject-matter education, we investigated how sequentially combining technology-enhanced generative activities and retrieval practice affects students’ conceptual knowledge and monitoring accuracy in authentic school settings for immediate and lasting learning after eight weeks. We preregistered hypotheses on immediate and lasting conceptual knowledge and monitoring accuracy: First, a generative activity (non-interactive teaching) should outperform the non-generative activity (restudy; generation hypothesis). Second, retrieval (retrieval quiz) should outperform non-retrieval (restudy; retrieval hypothesis). Third, the sequential combination of generation and retrieval should yield synergistic effects (interaction hypothesis). Moreover, we explored potential interaction effects between generation, retrieval, and explanation quality (completeness, elaboration, correctness).

2 Methods

We conducted a classroom experiment with secondary students (N = 344; power > .90; M = 12.58 years; 49.20% female) attending a curriculum-aligned, inquiry-based physics unit on converging lenses. We applied a 2×2 between-subjects design crossing technology-enhanced generative learning (non-generation, generation) and retrieval practice (non-retrieval, retrieval). Students either restudied the contents or taught them to a fictitious peer, and subsequently either restudied or completed a quiz. Time on task was kept constant, and the generative task was open-book. We assessed students' conceptual knowledge using the validated ROC-CI test (α = .72). Monitoring accuracy was calculated as difference between estimated and actual performance. To test our preregistered hypotheses, we used structural equation modeling (SEM) with the lavaan package in R (version 4.4.3), controlling for students' prior conceptual knowledge or prior monitoring accuracy. We also explored whether completeness, elaboration, and correctness of explanations or restudy notes moderated the effects.

3 Findings and Discussion

Analyses showed no significant differences between conditions regarding prerequisites (prior conceptual knowledge, prior monitoring accuracy, age, gender, native language, p ≥ .107). Contrary to our hypotheses, conceptual knowledge and monitoring accuracy did not significantly differ between generation (–.06 ≤ β ≤ .25; .093 ≤ p ≤ .969), retrieval (–.08 ≤ β ≤ .11; .307 ≤ p ≤ .783), or their sequential combination (–.18 ≤ β ≤ .12; .268 ≤ p ≤ .726) at either test. However, exploratory analyses revealed that retrieval practice supported lasting learning when prior generative processing was of low quality—that is, when explanations were incomplete, less elaborated, or inaccurate (–.28 ≤ β ≤ –.18; .002 ≤ p ≤ .007).

The results demonstrate that combining generation and retrieval does not universally enhance learning. Its effectiveness appears to depend on the quality of prior generative processing. In this sense, retrieval practice may serve a happy role—compensating for fragile knowledge—or a redundant one, offering no added benefit when prior understanding is already strong. These insights highlight the importance of considering students' processing and recognizing that instructional effectiveness is not one-size-fits-all, underscoring the need for continued multidisciplinary perspectives to better understand and optimize students' learning in authentic classroom contexts.

4 Educational Significance of the Research

The present findings indicate that the sequential combination of generative activities and retrieval practice should be applied selectively rather than as a uniform instructional approach. For classroom teaching, this suggests that teachers should continuously attend to how students enact generative activities and consider extending retrieval activities primarily when the quality of generative processing remains low. Importantly, these implications should not be understood as prescriptive rules. Teachers’ ongoing professional judgment remains essential, as the effectiveness of such instructional combinations may vary across students and contextual conditions.