Artificial intelligence chatbots provide inadequately transparent and readable patient information on triangular fibrocartilage complex injuries: a cross-sectional study
Abstract
BackgroundPatients increasingly use consumer artificial intelligence (AI) chatbots to obtain health information, yet the quality and readability of their musculoskeletal advice remain uncertain. Communicating triangular fibrocartilage complex (TFCC) injuries is particularly difficult because magnetic resonance imaging abnormalities may be incidental and their clinical relevance depends on history and targeted physical examination.MethodsIn this cross-sectional online study, 11 unmodified English-language TFCC-related queries selected from worldwide Google Trends data were submitted once to ChatGPT 5.5, Microsoft Copilot, Gemini 3.5 Flash, and Perplexity on August 3, 2026, yielding 44 responses. Two orthopedic physicians scored coded responses with platform names removed using DISCERN, a 20-item binary EQIP rubric, the Global Quality Score, and Journal of the American Medical Association transparency criteria; a response-level clinical content assessment also evaluated clinically inaccurate or misleading information and potentially harmful advice. Six readability indices were calculated. Model comparisons used matched Friedman tests with Kendall’s W, within-query permutation checks, paired Wilcoxon tests when indicated, and Holm adjustment.ResultsIndependent pre-consensus reliability was good for DISCERN [ICC(2, 1) = 0.857] and EQIP [ICC(2, 1) = 0.793], with weighted kappa values of 0.756 for GQS and 0.954 for JAMA. JAMA transparency scores differed among models (Holm-adjusted p < 0.001); DISCERN did not reach significance in the asymptotic family-adjusted analysis (p = 0.056) but did in the permutation sensitivity analysis (p = 0.045), whereas EQIP and GQS showed no overall differences. Perplexity had higher JAMA scores than ChatGPT 5.5 and Gemini 3.5 Flash in paired comparisons. Eight of 44 responses (18.2%) contained clinically inaccurate or misleading statements, while none met the criterion for potentially harmful advice. All six readability indices differed among models after Holm correction (adjusted P ≤ 0.002); every response scored above 6 on all five grade-level indices and below 80 on FRES.ConclusionChatbot-generated TFCC information showed limitations in treatment-related information quality, source transparency, readability, and some claim-level clinical content. Although chatbots may provide useful introductory information, their responses should complement rather than replace individualized clinical assessment, particularly when interpreting imaging findings or considering treatment options.
Related
- WITHDRAWN: Relevance Based Prediction: A Transparent, Non-Artificial Intelligence, Mathematical Solution to Personalized Opioid Treatment
- Developing and evaluating patient information leaflets for artificial intelligence use in healthcare
- Artificial intelligence chatbots as information sources for herpes zoster: an evaluation of patient-education suitability, overall quality, and readability
- Efficacy and Safety of Crisugabalin for Inadequately Controlled Diabetic Peripheral Neuropathic Pain: A Randomized, Active‐Controlled, Phase 2 Trial
- A focused PubMed artificial intelligence search program creates better literature reviews than general purpose artificial intelligence and is comparable to trainees in shoulder and elbow surgery
BibTeX
@article{ying2026artificial,
title = {{Artificial intelligence chatbots provide inadequately transparent and readable patient information on triangular fibrocartilage complex injuries: a cross-sectional study}},
author = {Dawen Ying and Qifang Zeng and Minzheng Ying and Min Yi and Jiongbiao Zhong},
journal = {Frontiers in Public Health},
year = {2026},
month = nov,
doi = {10.3389/fpubh.2026.1986047},
url = {https://www.frontiersin.org/articles/10.3389/fpubh.2026.1986047},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Artificial intelligence chatbots provide inadequately transparent and readable patient information on triangular fibrocartilage complex injuries: a cross-sectional study
AU - Dawen Ying
AU - Qifang Zeng
AU - Minzheng Ying
AU - Min Yi
AU - Jiongbiao Zhong
PY - 2026
DA - 2026/11/03
JO - Frontiers in Public Health
DO - 10.3389/fpubh.2026.1986047
UR - https://www.frontiersin.org/articles/10.3389/fpubh.2026.1986047
AB - BackgroundPatients increasingly use consumer artificial intelligence (AI) chatbots to obtain health information, yet the quality and readability of their musculoskeletal advice remain uncertain. Communicating triangular fibrocartilage complex (TFCC) injuries is particularly difficult because magnetic resonance imaging abnormalities may be incidental and their clinical relevance depends on history and targeted physical examination.MethodsIn this cross-sectional online study, 11 unmodified English-language TFCC-related queries selected from worldwide Google Trends data were submitted once to ChatGPT 5.5, Microsoft Copilot, Gemini 3.5 Flash, and Perplexity on August 3, 2026, yielding 44 responses. Two orthopedic physicians scored coded responses with platform names removed using DISCERN, a 20-item binary EQIP rubric, the Global Quality Score, and Journal of the American Medical Association transparency criteria; a response-level clinical content assessment also evaluated clinically inaccurate or misleading information and potentially harmful advice. Six readability indices were calculated. Model comparisons used matched Friedman tests with Kendall’s W, within-query permutation checks, paired Wilcoxon tests when indicated, and Holm adjustment.ResultsIndependent pre-consensus reliability was good for DISCERN [ICC(2, 1) = 0.857] and EQIP [ICC(2, 1) = 0.793], with weighted kappa values of 0.756 for GQS and 0.954 for JAMA. JAMA transparency scores differed among models (Holm-adjusted p < 0.001); DISCERN did not reach significance in the asymptotic family-adjusted analysis (p = 0.056) but did in the permutation sensitivity analysis (p = 0.045), whereas EQIP and GQS showed no overall differences. Perplexity had higher JAMA scores than ChatGPT 5.5 and Gemini 3.5 Flash in paired comparisons. Eight of 44 responses (18.2%) contained clinically inaccurate or misleading statements, while none met the criterion for potentially harmful advice. All six readability indices differed among models after Holm correction (adjusted P ≤ 0.002); every response scored above 6 on all five grade-level indices and below 80 on FRES.ConclusionChatbot-generated TFCC information showed limitations in treatment-related information quality, source transparency, readability, and some claim-level clinical content. Although chatbots may provide useful introductory information, their responses should complement rather than replace individualized clinical assessment, particularly when interpreting imaging findings or considering treatment options.
SN - 2296-2565
ER -