Exploring Large Language Model as an Interactive Sports Coach: Lessons from a Single-Subject Half Marathon Preparation

TL;DR

This study investigates whether a general-purpose large language model can function as a longitudinal virtual coach for endurance training in a two-month, single-subject half-marathon preparation. It documents how the LLM acted as planner, explainer, and motivator, guiding periodized training that culminated in finishing 21.1 km at a pace of approximately 6'30"/km after starting from 2 km at 7'54"/km, with concurrent improvements in cadence and cardiovascular efficiency. The work also identifies critical gaps—absence of real-time sensing, reliance on text-only feedback, variable motivational support, and limited personalization—while offering a design agenda centered on multimodal sensing, on-device feedback, proactive motivation, and privacy-preserving personalization. Collectively, the findings demonstrate the feasibility of LLM-assisted coaching for real-world endurance training while outlining concrete requirements to scale and secure such systems across larger cohorts and longer horizons.

Abstract

Large language models (LLMs) are emerging as everyday assistants, but their role as longitudinal virtual coaches is underexplored. This two-month single subject case study documents LLM guided half marathon preparation (July-September 2025). Using text based interactions and consumer app logs, the LLM acted as planner, explainer, and occasional motivator. Performance improved from sustaining 2 km at 7min 54sec per km to completing 21.1 km at 6min 30sec per km, with gains in cadence, pace HR coupling, and efficiency index trends. While causal attribution is limited without a control, outcomes demonstrate safe, measurable progress. At the same time, gaps were evident, no realtime sensor integration, text only feedback, motivation support that was user initiated, and limited personalization or safety guardrails. We propose design requirements for next generation systems, persistent athlete models with explicit guardrails, multimodal on device sensing, audio, haptic, visual feedback, proactive motivation scaffolds, and privacy-preserving personalization. This study offers grounded evidence and a design agenda for evolving LLMs from retrospective advisors to closed-loop coaching companions.

Figures (21)

• Figure 1: Representative Nike Run Club screenshots. Core fields (date, distance, average pace, route) were consistently available and shared with the LLM.
• Figure 2: Representative Zepp screenshots paired with Amazfit Balance 2. Compared with Nike Run Club, Zepp offered richer physiological and training signals (continuous HR, cadence, stride, estimated $\mathrm{VO_2max}$, recovery indices).
• Figure 3: Initial instruction (system prompt) supplied to the LLM to standardize its coaching role.
• Figure 4: Illustrative participant--LLM exchange showing initial assessment and a condensed nine-week roadmap.
• Figure 5: Daily running mileage across the two-month build. Bars represent per-session distance, beginning with the baseline assessment on July 25, 2025.