Assessment of Plagiarism in AI-Generated Responses to Gynecologic Oncology-Related Queries

Originality and attribution of narrative responses in healthcare remain underexamined. Plagiarism carries significant ethical, legal, and professional consequences, undermining trust in information. This study employed a quantitative, comparative design, using iThenticate 2.0, to analyze chatbot plagiarism. This cross-sectional study included 24 questions related to gynecologic-oncology which generated 360 queries submitted to Gemini and Copilot. Narrative responses were generated by Google Gemini (version 1.5 Flash) and Microsoft Copilot (GPT-4 based Prometheus model). The main outcomes were the presence and percentage of plagiarisms in chatbot narratives. Plagiarism was detected in narratives from both Gemini and Copilot. Responses from Gemini contained more plagiarism events per narrative 3.09 ± 0.10 than Copilots 1.59 ± 0.07; P < .05. Gemini cited more sources (4.4 ± 0.11 vs 2.98 ± 0.07; P < .05) and had more plagiarized content from copyrighted or published sources than Copilot (2.57 ± 0.10 1.34 ± 0.07; P < .05). Mean pla-giarism scores were higher for Gemini (26.19% ± 0.88) than for Copilot (19.01% ± 0.88; P < .05). Narrative responses from Gemini and Copilot contain plagiarism and highlight a need for assessing plagiarism in AI-generated content. Proper citation and attribution in AI responses are needed to mitigate plagiarism, reduce hallucination, and enhance the verifiability and credibility of information from AI platforms.