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

Carnino et al.

Page 5

Recommendations for academic journals

Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Based on our findings, we suggest that academic journals consider strategies to address the evolving impact of AI technologies, including a range of LLMs and chatbots beyond just ChatGPT on the content of scholarly articles. Despite existing guidelines surrounding the use of AI tools in research and the call for transparency, it is evident that authors have yet to fully embrace these policies. To address this, academic journals should collaborate with AI experts to pioneer the development and implementation of more refined tools for detecting AI-generated text. This collaborative effort aims to mitigate potential biases and enhance the reliability of assessments. Recognizing the inherent difficulty in creating accurate tools to detect AI-generated text, it is crucial to acknowledge This challenge. However, despite the complexities involved, initiating efforts to develop such tools is imperative for safeguarding academic integrity and ensuring the accuracy of reported scientific findings. Additionally, given the observed variations tied to authors’ country of academic affiliation, journals should exercise caution in interpreting results and refrain from making assumptions about AI usage by authors from non-English speaking countries. Recognizing the imperfect nature of the current system, thorough screening for AI-generated text in academic articles remains crucial to uphold the accuracy of reported scientific findings. In essence, these recommendations seek to strike a harmonious balance between harnessing AI advancements and upholding the integrity and transparency of academic publishing. This study has several limitations should be acknowledged. Firstly, the evaluation focused exclusively on articles published in one high impact academic journal, JAMA - Oto, limiting the generalizability of the results to other medical disciplines or academic journals. Additionally, the exclusion of other article types such as reviews, case reports, opinions, and commentaries due to their varied lengths and structural inconsistencies also restricts the scope of our findings. The study’s reliance on an online application, ZeroGPT.com, for estimating the percentage of AI-generated text introduces a potential source of variability and may not capture nuances in the text’s quality or authenticity. Importantly, applications to detect AI-generated text are still imperfect and thus typically report some level of AI generated text in articles written by humans depending on the author’s style of writing. Furthermore, the study raises questions about the purpose of AI usage, whether for entire manuscript drafting or specific sections, without providing definitive insights into authors’ intentions. The detected variations in AI-generated text based on the country of academic affiliation highlight a potential bias in current AI detectors, necessitating caution in interpreting these results. Overall, these limitations underscore the need for future research to address these constraints and further explore the complex dynamics between AI, scientific discourse, and publication integrity.

Limitations

Conclusion

In conclusion, our investigation into the effects of AI-text generation models, with ChatGPT being one of the many influential tools, has uncovered noteworthy insights. The surge in AI generated text, particularly in abstracts, introductions, and discussions, following ChatGPT’s

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