ChatGPT road running training prescription: An expert validation analysis

Abstract

The use of online tools for training guidance, especially ChatGPT, is a common practice among road runners. However, there is no information on the quality of the information provided. This study aimed to analyse the guidance and prescriptions obtained by ChatGPT for road running, under validation by professional coaches. This is an exploratory descriptive study, carried out in two stages: 1) presentation of prompts for evaluation of ChatGPT regarding training guidelines in three different cases (beginner runner; intermediate runner; advanced runner); 2) evaluation, using a Likert scale (1:completely correct; 4:completely incorrect) of the responses generated by ChatGPT, by road running coaches. Most of the responses were considered “completely correct.” For the first simulated case, it was noted that some experts classified questions three (training volume) and four (training intensity) as “more incorrect than correct.” In comparison to the simulated cases two and three, most experts classified the generated responses as “more incorrect than correct.” The highest frequencies of the “more incorrect than correct” option were observed for questions two and five at the intermediate level, and questions one, two, and four at the advanced level. For the questions related to the third case (advanced level), it was observed that question five was considered “completely incorrect” by 6.3% of the experts. The coaches’ suggestions were directed to the race strategy, hydration and supplementation. ChatGPT provided realistic running training information based on user prompts, but the use of AI tools to prescribe specific road running activities has both potential and limitations.