AI-generated high-intensity interval training information: a cross-sectional comparative evaluation of quality and readability
- 1Department of Sports Science, Yancheng Institute of Technology
- 2Department of Sports Science, Kyonggi University
- 3Police PE Teaching & Research Section, Jiangsu Justice Police College
Abstract
Background: AI chatbots are increasingly used for health and exercise information. High-intensity interval training (HIIT) is a common exercise topic, yet the quality, transparency, and readability of chatbot-generated HIIT information have not been systematically compared under a standardized testing protocol.
Objective: To compare selected dimensions of quality, transparency, and readability in HIIT-related responses generated by ChatGPT, Copilot, Gemini, and Perplexity, and to evaluate readability against a prespecified sixth-grade benchmark for public-facing health information.
Methods: This cross-sectional comparative content analysis was conducted in August 2026. Google Trends was used to identify HIIT-related search terms worldwide over the preceding 5 years. Of 29 candidate terms, 14 were retained after predefined screening, and each term was submitted once to each of four chatbots, yielding 56 responses. Two reviewers, both holding doctoral degrees in sports science, independently evaluated all responses using DISCERN, EQIP, the JAMA benchmark criteria, and GQS, which assessed complementary dimensions of information quality, usability, transparency, and overall usefulness. Inter-rater reliability was assessed before adjudication. Six readability indices were calculated. Because the same 14 queries were evaluated across all four chatbots, overall differences were tested using Friedman tests; significant results were followed by paired Wilcoxon signed-rank tests with Holm correction. Kendall’s W was reported as the overall effect-size measure.
Results: Perplexity-generated responses had the highest median scores for DISCERN [51.50 (47.50, 53.75)], EQIP [75.00 (70.00, 80.00)], JAMA [1.00 (1.00, 1.00)], and GQS [4.00 (4.00, 4.00)]. Friedman tests showed significant differences across chatbots for DISCERN (χ ² = 27.98, p < 0.001, W = 0.666), EQIP (χ ² = 25.74, p < 0.001, W = 0.613), JAMA (χ ² = 35.40, p < 0.001, W = 0.843), and GQS (χ ² = 21.15, p < 0.001, W = 0.504). None of the four chatbots met the prespecified sixth-grade readability benchmark across the evaluated indices.
Conclusion: Under this specific testing protocol, HIIT-related responses generated by the four chatbots differed in the evaluated dimensions of information quality and transparency, with Perplexity-generated responses receiving higher median scores on all four assessment instruments. These scores should not be interpreted as direct evidence of factual accuracy, citation accuracy, or clinical safety, which were not independently assessed. None of the evaluated chatbot-generated responses met the prespecified readability benchmark.
Related
- Quality, readability, and information fidelity of large language model–generated versus physician-authored informed-consent content for targeted therapies: a blinded, paired comparative study
- Quality, Readability, and Content Deficiencies of Three Large Language Models in Bee Sting Health Information: A Comparative Evaluation with Patient Perspectives
- Large language models as sources of diabetic retinopathy patient education: a comparative evaluation of safety, health information quality, and readability
- Quality and readability of responses generated by popular large language models to frequently asked questions about menopause: A comparative study
- Quality, Readability, Understandability, and Actionability of AI Chatbots’ Responses to Dental Patient Questions: A Comparative Evaluation
BibTeX
@article{zhang2026aigenerated,
title = {{AI-generated high-intensity interval training information: a cross-sectional comparative evaluation of quality and readability}},
author = {Zheng Zhang and Xiang-Yan Ruan and Kai-Tuo Zhu},
journal = {Frontiers in Public Health},
year = {2026},
month = oct,
doi = {10.3389/fpubh.2026.1976676},
url = {https://www.frontiersin.org/articles/10.3389/fpubh.2026.1976676/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - AI-generated high-intensity interval training information: a cross-sectional comparative evaluation of quality and readability
AU - Zheng Zhang
AU - Xiang-Yan Ruan
AU - Kai-Tuo Zhu
PY - 2026
DA - 2026/10/23
JO - Frontiers in Public Health
DO - 10.3389/fpubh.2026.1976676
UR - https://www.frontiersin.org/articles/10.3389/fpubh.2026.1976676/full
AB - Background: AI chatbots are increasingly used for health and exercise information. High-intensity interval training (HIIT) is a common exercise topic, yet the quality, transparency, and readability of chatbot-generated HIIT information have not been systematically compared under a standardized testing protocol. Objective: To compare selected dimensions of quality, transparency, and readability in HIIT-related responses generated by ChatGPT, Copilot, Gemini, and Perplexity, and to evaluate readability against a prespecified sixth-grade benchmark for public-facing health information. Methods: This cross-sectional comparative content analysis was conducted in August 2026. Google Trends was used to identify HIIT-related search terms worldwide over the preceding 5 years. Of 29 candidate terms, 14 were retained after predefined screening, and each term was submitted once to each of four chatbots, yielding 56 responses. Two reviewers, both holding doctoral degrees in sports science, independently evaluated all responses using DISCERN, EQIP, the JAMA benchmark criteria, and GQS, which assessed complementary dimensions of information quality, usability, transparency, and overall usefulness. Inter-rater reliability was assessed before adjudication. Six readability indices were calculated. Because the same 14 queries were evaluated across all four chatbots, overall differences were tested using Friedman tests; significant results were followed by paired Wilcoxon signed-rank tests with Holm correction. Kendall’s W was reported as the overall effect-size measure. Results: Perplexity-generated responses had the highest median scores for DISCERN [51.50 (47.50, 53.75)], EQIP [75.00 (70.00, 80.00)], JAMA [1.00 (1.00, 1.00)], and GQS [4.00 (4.00, 4.00)]. Friedman tests showed significant differences across chatbots for DISCERN (χ ² = 27.98, p < 0.001, W = 0.666), EQIP (χ ² = 25.74, p < 0.001, W = 0.613), JAMA (χ ² = 35.40, p < 0.001, W = 0.843), and GQS (χ ² = 21.15, p < 0.001, W = 0.504). None of the four chatbots met the prespecified sixth-grade readability benchmark across the evaluated indices. Conclusion: Under this specific testing protocol, HIIT-related responses generated by the four chatbots differed in the evaluated dimensions of information quality and transparency, with Perplexity-generated responses receiving higher median scores on all four assessment instruments. These scores should not be interpreted as direct evidence of factual accuracy, citation accuracy, or clinical safety, which were not independently assessed. None of the evaluated chatbot-generated responses met the prespecified readability benchmark.
SN - 2296-2565
ER -