AI-Generated 'Ghost Citations' Increasingly Appearing in Scientific Papers
The use of artificial intelligence is leading to a growing problem of fabricated citations, known as 'ghost citations,' appearing in scientific articles. Research by De Groene Amsterdammer reveals that Dutch academia is not immune to this phenomenon. This follows a February report in the journal Nature that identified at least 10,000 publications from 2025 containing AI-invented academic references. Over the past three years, hundreds of scientists affiliated with Dutch universities have also published articles featuring these non-existent references. The Dutch publication, in collaboration with Utrecht University's Data School, analyzed over 100,000 scientific journal publications. They found that the proportion of articles with ghost citations has increased sevenfold in three years, rising from one in 1,400 articles in 2023 to one in 200 in early 2026. In total, De Groene identified over 748 references to non-existent studies across 208 peer-reviewed articles.
Scientists involved in the research indicated that they used AI to adjust the alphabetical order or style of their reference lists, which inadvertently led to these hallucinations. In other instances, researchers sought AI-generated sources to support their hypotheses, failing to adequately verify if the suggested references actually existed. The danger of these fake references is their potential to create a snowball effect, where a fabricated citation begins to be cited by others, potentially misleading research, such as a study on brain cancer treatment that relied on both real and invented sources. This can create a false consensus based on a fragile foundation, leading to a distorted view of existing literature with serious consequences.
Universities like Wageningen University and Erasmus MC have been identified as having published articles with ghost citations. Wageningen University expressed concern and stated they would review internal control mechanisms. While traditional peer review is meant to catch errors, it can be time-consuming and sometimes lacks thoroughness. Issues with citation accuracy predate generative AI, partly due to publication pressure on scientists. Publishers are now developing AI detection systems, and some, like Springer Nature and Elsevier, have expanded their integrity teams, acknowledging that traditional checks are insufficient. New global reporting standards for AI in science are being developed, emphasizing the need for transparency from scientists regarding their AI usage and verification processes.
AI-driven content generation tools, while offering potential efficiencies, introduce significant risks to academic integrity through the proliferation of 'ghost citations.' This phenomenon highlights a systemic vulnerability where the pursuit of publication metrics, amplified by the ease of AI use, can override rigorous verification processes. The increasing prevalence of fabricated references suggests a critical need for enhanced AI detection capabilities and a re-evaluation of academic reward structures that may inadvertently incentivize speed over accuracy. Future research integrity frameworks must integrate robust AI usage guidelines, mandating transparency and verification protocols to ensure that scientific discourse remains grounded in verifiable evidence, thereby safeguarding against the erosion of trust in scholarly communication.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.