French journalist Romain Challand targeted a 1-hour, 45-minute finish at the Lyon half-marathon on October 4 after letting ChatGPT design his six-month training regimen, index.hu reported. The experiment exposed the limits of artificial intelligence in athletic coaching, revealing that automated running plans often lack subjective adaptation, environmental awareness, and individualized physiological monitoring.
Challand began preparing for the October race in March, following a weekly schedule generated by the OpenAI chatbot that consisted of an easy run on Tuesdays, a specialized workout on Thursdays, and a long run on Sundays. Although Challand was an experienced runner with an athletic background, the plan quickly unraveled on race day. By the third kilometer, he realized his target pace was unsustainable and was forced to incorporate early walking breaks to avoid physical collapse.
Around the sixteenth kilometer, Challand abandoned hope of hitting his 1-hour, 45-minute goal and blamed the chatbot for the miscalculation. Seeking answers, he consulted a long-time friend who operates a trail-running club and works as a running coach. According to the coach, Thibault, the artificial intelligence failed to account for environmental variables such as intense summer heatwaves, during which the chatbot continued to prescribe high-intensity sessions that Challand had to manually modify.
The coaching analysis highlighted a fundamental flaw in automated training platforms. Thibault noted that the artificial intelligence trains a clock rather than a human being, completely omitting subjective factors and personalization. While human coaches prefer daily monitoring and encourage athletes to trust their bodily signals, Challand followed the machine’s directives blindly, accumulating fatigue through long interval blocks and runs exceeding 15 kilometers that culminated in exhaustion on race day.
Why the Chatbot Pushed an Unrealistic Goal
Throughout the six-month preparation, ChatGPT evaluated every completed and validated workout as fresh evidence that the 1-hour, 45-minute target remained achievable. The software functioned by aggregating positive signals while largely ignoring weaker negative indicators, creating a persistent bias toward optimism. Studies examining automated training programming, including research published in the Biology of Sport, indicate that algorithmic plans frequently prioritize broad safety parameters over precise training efficiency and often lack true state-specificity.
While artificial intelligence models perform well when delivering practical, scientifically sound answers based on rigid data inputs, they cannot replicate the dynamic interaction of a human coach. The system did not ask clarifying questions about Challand’s nutrition or recovery habits, nor did it proactively adjust the schedule in response to subjective fatigue levels.
The Future of Real-Time Motion Tracking
As athletes and technologists explore the intersection of artificial intelligence and physical training, new tools are emerging to bridge the gap between static schedules and dynamic human movement. An experimental study published in Healthcare in February tested MediaPipe Pose, a machine learning model developed by Google capable of analyzing body posture and real-time movement via a standard smartphone camera to provide instant feedback on exercise execution.
October 4 marked the conclusion of Challand’s half-marathon experiment in Lyon.
Keep reading
