xRead - Incorporating Artificial Intelligence into Clinical Practice (March 2026)

Carnino et al.

Page 4

when assessing whether the country of academic affiliation of the principal investigator (PI) was primarily English-speaking, as indicated in Table 5.

Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Discussion

This study aimed to assess the impact of AI-generated text on the content of articles published in JAMA – Oto, focusing on the period before and after the release of ChatGPT in November 2022. The findings of this analysis align with the recent policy updates from JAMA and the JAMA Network journals, which advocate for transparent and responsible AI tool use in both manuscript preparation and research. As detailed in their guidelines first issued in 2023 and expanded upon in 2024 [21, 22], the emphasis on disclosing AI involvement underscores the critical nature of transparency and accountability in scholarly publishing. This study, revealing a notable uptick in AI-generated content post-ChatGPT’s introduction, prompts a broader reflection on the academic community’s adaptation to these policies and highlights the ongoing need for dialogue and research to effectively navigate AI’s challenges and opportunities within scientific dissemination. These results suggest that researchers are likely utilizing ChatGPT for assistance in composing these specific article sections. However, it remains unclear whether authors employ AI to draft entire manuscripts or if these chatbots are utilized primarily for proofreading and editing. Regardless of the purpose, there is a potential risk associated with ChatGPT, given its propensity to generate false statements, often referred to as “AI hallucinations” [17, 23]. Interestingly, authors from non-English speaking countries were found to have a higher percentage of AI-generated text than authors from English-speaking countries (Table 5). Interpreting this finding poses challenges due to the imperfect nature of current applications designed to detect AI-generated text. On one hand, it could suggest that authors from non-English speaking countries are more inclined to use AI chatbots for tasks such as translation or drafting manuscripts. However, it’s crucial to acknowledge that AI detectors have been reported to falsely accuse non-native English speakers [24]. This discrepancy may arise from the stylistic similarity between the writing style of individuals learning English as a second language and text generated by AI. Consequently, the interpretation of this finding remains uncertain, emphasizing the need for further research before considering widespread adoption of applications for detecting AI-generated text. Moreover, the unexpectedly high detection of AI-generated text in the results sections can be partly explained by the performance characteristics of ZeroGPT.com. The tool’s accuracy is compromised by the shorter length and the straightforward, data-focused language typical of results sections, which resembles AI-generated text. This finding is particularly surprising since one would expect the highest percentages of AI-generated text in the more narrative sections such as the abstract, introduction, or discussion, not in the results which typically involve direct data reporting. This highlights the need for refined methodologies in AI text detection to ensure accuracy and reliability.

Eur Arch Otorhinolaryngol . Author manuscript; available in PMC 2025 November 01.