EUVIMEDEUVIMED entdecken
Übersichtsarbeit

AI hallucinations in academic writing: implications for research integrity

Louie Giray, Md Emdadul Islam, John Arvin Glo

Naunyn-Schmiedeberg's Archives of Pharmacology · 2026-08 · 399 (11) · 16827-16839

DOI: 10.1007/s00210-026-05494-4 ↗

Kurz gefasst

Übersichtsarbeit in der Zeitschrift Naunyn-Schmiedeberg's Archives of Pharmacology, online erschienen am 30.05.2026. Verfasst von Louie Giray und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Liberal Arts, School of Foundational Studies and Education, Mapua University, Manila, Philippines. Hauptthemen laut MEDLINE: Writing, Artificial Intelligence. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 76 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As artificial intelligence becomes deeply embedded in scholarly practice, a critical and underexamined threat to research integrity has emerged: AI hallucinations. AI hallucinations refer to outputs generated by large language models that are factually incorrect, fabricated, or logically inconsistent with verifiable knowledge, produced not through genuine understanding but through statistical pattern prediction. This paper examines how these hallucinations manifest in academic writing and analyzes their specific consequences for scholarly work. The paper identifies the principal causes of hallucination, including training data gaps, monofacts, named entity errors, and prompt design failures, and documents their most common forms in academic contexts, among them citation fabrication, factual distortion, logical inconsistency, and propagation errors. The paper then examines how these failures undermine the accuracy and reliability of research and disrupt peer review processes in ways that traditional editorial mechanisms are not equipped to detect. Ethical dimensions are addressed directly: AI hallucinations create conditions of distributed epistemic responsibility that complicate established definitions of research misconduct and authorship, yet accountability remains fully with human authors. Also, it discusses mitigation strategies, including detection tools, institutional policy frameworks, and AI literacy curricula, arguing that human verification is irreplaceable and that literacy-based approaches offer the most sustainable institutional response. Equity concerns are highlighted throughout, as hallucination rates vary by language and disciplinary domain. The paper concludes that responsible AI use in scholarship requires transparency, systematic verification, and human oversight as non-negotiable ethical obligations.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

WritingArtificial IntelligenceHumansAcademiaScientific Misconduct

Stichwörter der Autor:innen

AI hallucinationsAcademic writingLarge language modelsResearch ethicsResearch integrityScientific publishing

Autor:innen und Institutionen

  • Department of Liberal Arts, School of Foundational Studies and Education, Mapua University, Manila, Philippines.
  • Research Center for International and Global Higher Education, Khazar University, Baku, Azerbaijan.
  • Institute of Information Sciences, Noakhali Science and Technology University, Noakhali, Bangladesh.
  • National University, Manila, Philippines. jaaglo@nu-laspinas.edu.ph.

Zitieren

APA

Giray, L., Islam, M. E., & Glo, J. A. (2026). AI hallucinations in academic writing: implications for research integrity. Naunyn-Schmiedeberg's Archives of Pharmacology, 399(11), 16827-16839. https://doi.org/10.1007/s00210-026-05494-4

Vancouver

Giray L, Islam ME, Glo JA. AI hallucinations in academic writing: implications for research integrity. Naunyn-Schmiedeberg's Archives of Pharmacology. 2026;399(11):16827-16839. doi:10.1007/s00210-026-05494-4

BibTeX
@article{Giray2026,
  author = {Giray, Louie and Islam, Md Emdadul and Glo, John Arvin},
  title = {AI hallucinations in academic writing: implications for research integrity},
  journal = {Naunyn-Schmiedeberg's Archives of Pharmacology},
  year = {2026},
  volume = {399},
  number = {11},
  pages = {16827--16839},
  doi = {10.1007/s00210-026-05494-4}
}
Kostenlos starten →