LeagueBot: A Voice LLM Companion of Cognitive and Emotional Support for Novice Players in Competitive Games

Competitive games pose steep learning curves and strong social pressures, often discouraging novice players and limiting sustained engagement. To address these challenges, this study introduces LeagueBot, a large language model-based voice chatbot designed to provide both informational and emotional support during live gameplay in league of legends, one of the most competitive multiplayer online battle arena games. In a within-subjects experiment with 33 novice players, LeagueBot was found to reduce cognitive challenge, performative challenge, and perceived tension. Qualitative analysis further identified three themes: enhanced access to game information, relief from cognitive burden, and practical limitations. Participants noted that LeagueBot offered context-appropriate guidance and emotional support, helping ease the steep learning curve and psychological pressures of competitive gaming. Together, these findings underscore the potential of voice-based LLM companions to assist novice players in competitive environments and highlight their broader applicability for real-time support in other high-pressure contexts.

💡 Research Summary

The paper presents LeagueBot, a voice‑based large language model (LLM) companion designed to provide both informational and emotional support to novice players during live League of Legends (LoL) matches. Competitive multiplayer online battle arena (MOBA) games are notorious for steep learning curves, complex decision‑making, and intense social pressure, which together discourage newcomers and increase churn. While prior assistance tools (tutorials, visual overlays, matchmaking policies) have focused largely on technical performance, they rarely address the psychological strain that players experience in real time.

To fill this gap, the authors built a system that integrates a state‑of‑the‑art LLM with automatic speech recognition (ASR) and text‑to‑speech (TTS) components. The game client streams contextual data (champion level, cooldowns, item purchases, map events) to a backend service. This data is formatted into prompts that guide the LLM to generate context‑appropriate advice, item‑build recommendations, strategic cues, and brief emotional encouragement (e.g., “You’re doing great, stay calm”). The generated text is then vocalized and delivered to the player through a headset, allowing hands‑free interaction. Players can also ask open‑ended questions (“What should I buy next?”) or request a brief pause (“I need a moment”). The architecture emphasizes low latency through asynchronous processing and local caching, while employing multi‑candidate selection to mitigate ASR errors.

The effectiveness of LeagueBot was evaluated with a within‑subjects experiment involving 33 novice LoL players. Each participant played two matches: one with LeagueBot active and one without, in a counterbalanced order. After each match, participants completed three validated questionnaires: (1) NASA‑TLX‑derived cognitive load, (2) perceived performance challenge, and (3) subjective tension. Quantitative analysis revealed statistically significant reductions across all three metrics when LeagueBot was present (p < 0.05), with the most pronounced effect on cognitive load.

In addition to the quantitative data, semi‑structured post‑match interviews were conducted and subjected to thematic analysis. Three dominant themes emerged:

1. Enhanced Access to Game Information – Players appreciated being able to retrieve real‑time data (cooldowns, optimal item timings, enemy positioning) without diverting visual attention from the screen.

2. Relief from Cognitive Burden – By offloading tactical reasoning to the LLM, participants reported higher focus, fewer mistakes, and a smoother flow of play.

3. Practical Limitations – Participants noted occasional speech‑recognition mistakes, latency spikes, and moments when the assistant’s suggestions were either too frequent or insufficiently specific, occasionally disrupting immersion.

The discussion interprets these findings in light of HCI principles for high‑pressure, fast‑paced environments. The authors argue that voice‑based LLMs can uniquely satisfy the dual need for rapid, context‑sensitive guidance and affective reassurance, something text‑only systems cannot achieve without breaking gameplay flow. Design recommendations include: (a) ensuring non‑intrusiveness by allowing users to set advice frequency and granularity, (b) maintaining strong context awareness through robust prompt engineering, and (c) providing transparent confidence indicators so users can judge the reliability of suggestions.

Limitations are acknowledged: the prototype currently operates only in English, the ASR component still suffers from occasional errors, and the latency, while acceptable for the experimental setting, may need further optimization for competitive play at the highest tiers. Future work is outlined to incorporate multilingual support, lightweight on‑device LLM inference, adaptive personalization based on player skill progression, and integration with existing in‑game UI elements.

Finally, the authors extrapolate the broader applicability of voice‑LLM companions beyond gaming, suggesting potential in domains such as medical training, aviation, and sports coaching, where real‑time cognitive assistance and emotional regulation are equally critical. The study thus positions LeagueBot as a proof‑of‑concept that bridges the gap between AI‑driven technical tutoring and affective support, opening new research avenues for AI‑enhanced human performance in high‑stakes, real‑time contexts.