Generative AI in Scientific Writing : A Conceptual Analysis of Hallucinated References and Implications for Research Data Integrity

Annisa Humairo (Universitas Telkom)

Media Informatika
Vol 25 No 2 (2026)

The evolution of the generative artificial intelligence (AI) has transformed scientific writing practices, especially in the process literature synthesis and referencing. The use of Generative AI offering efficiency and cognitive support for researchers, but that it also causes the risk in reference hallucinations, such as citations that looks credible but that is just fictive. This phenomenon becomes serious challenges for validity of scientific claims, the reproducibility of research, and trust to the the academic publication system. Previous studies have discussed about the use of generative AI in ethical perspective, human–AI collaboration, and AI-based content detection. However, most studies still put the referencing errors as a technical or ethical issue, without explicitly linking it to the research data integrity framework that considering the referencing as a crucial scientific data artifact. This study aims to fill this gap by analyzing the hallucinated references as a systemic research data integrity problem. This study uses a qualitative approach based on conceptual analysis and cross-domain thematic synthesis and understand about the generative AI literature, scientific writing, data integrity research , and governance and scientific publications auditing. The analysis focuses on mapping the relationships between the AI models technical limitations, user behavior, and institutional mechanisms in the scholarly publishing ecosystem. The research results indicated that hallucinated references are a socio-technical phenomenon that is strengthened by the interaction between generation AI probabilistic, user bias automation, and the weaknesses in policy and editorial control. The impacts include erosion of research data integrity, contamination of accumulated knowledge, increased peer review pressure, and decreased the trust in scientific publications. This study emphasizes the need for a paradigm shift from an individual approach to a systemic governance-based approach, which the references are treated as objects of research data that must be verified, documented, and auditable. These findings provide conceptual contributions to policy development, editorial practices, and a continuing research agenda to maintain scientific integrity in the era of generative AI.