The emergence of large language models (LLMs) has sparked significant challenges within the global research community, leading to a surge of low-quality academic papers generated with the help of AI. This phenomenon raises important questions about the future of academic publishing.
We are no longer in the realm of hypothetical concerns; extensive analyses from leading scholarly databases such as Elsevier’s Scopus, Web of Science, and arXiv have unveiled a dramatic trend. By early 2025, researchers have observed that close to 20 percent of preprints in computer science and over 10 percent of papers in STEM databases exhibit linguistic characteristics suggesting LLM involvement.
Some estimates propose that scientists utilizing these tools have increased their publication output by as much as one-third, resulting in a flood of submissions overwhelming a system that was not designed to handle such volume.
Submission Acceleration
This issue extends beyond mere editorial refinement; AI technology has significantly lowered the barriers to entry for industrial-scale paper mills, now capable of producing fabricated but seemingly credible studies at an alarming rate. This proliferation risks undermining the foundational trust essential for scientific progress. Nonetheless, the spike in publications reflecting LLM involvement should not solely be viewed as a decline in research quality; it also signifies the deeper integration of AI into standard scientific practices, which may enhance rigor and productivity.
This overwhelming increase in AI-driven output is pushing the peer review system to the brink of collapse. Since the mid-17th century with the Royal Society’s introduction of Philosophical Transactions, peer review has served as the cornerstone of research quality.
However, the sheer volume of AI-enhanced submissions leads to a pressing question: is the traditional peer-reviewed paper still the most effective platform for sharing research findings? The challenge lies both in supply and demand; as submissions surge, the pool of qualified reviewers—already stretched thin and often unrewarded—is dwindling, leaving those responsible for scientific integrity grappling with the influx of machine-generated content.
Publishers are adapting to this landscape. Organizations like Springer Nature and Elsevier are investing in AI technologies and human expertise to identify machine-generated materials, streamlining the process to enable more effective desk rejections before peer review begins. While these technological measures are essential and beneficial, they are not a comprehensive solution. The battle against low-quality publications represents a race for detection capabilities, and reliance on technology alone may further entrench biases where non-native English speakers are disproportionately flagged for employing conventional structures that resemble AI outputs.
Cultural Shift
To transcend a purely reactive approach, everyone involved in academia has a role in advocating for a transition from a culture focused on quantity to one emphasizing quality. This involves reevaluating the incentives that shape research behavior, acknowledging that optimal strategies will differ significantly across academic fields. The responsible use of generative AI in STEM disciplines will differ greatly from the more interpretative approaches found in the humanities. Nonetheless, several pivotal changes can be collectively championed:
Supervisors and Mentors: We should shift away from fixed publication targets for PhD candidates to alleviate the pressure of a “publish or perish” environment. In the humanities, this could mean emphasizing the depth of a chapter or monograph; in STEM, it might prioritize methodological soundness and research impact over sheer volume.
This change is not intended to hinder students’ publishing opportunities as they establish their careers. Instead, it aims to ensure their work’s quality is acknowledged and effectively demonstrated. By emphasizing the merit of research over the quantity of output, the system can better support early-career researchers in making valuable contributions to their fields.
Examiners: Peer-reviewed publications and outputs should not serve as the primary metric for assessing a PhD candidate in their viva. Although such publications provide fundamental validation and assurance of a candidate’s originality, the candidate’s domain expertise, the influence of their research results, and the integrity of their contributions should remain paramount.
Funders and Reviewers: There is a promising trend toward narrative CVs and comprehensive assessments, but this shift needs to accelerate. A decisive move away from merely counting outputs toward evaluating the long-term scientific and societal impact of research and broader contributions is essential.
University Leadership: The Research Excellence Framework has made significant strides in emphasizing quality outputs over quantity, and university leaders should follow suit by prioritizing quality in evaluations, promotions, and hiring practices. This approach may involve developing workload models that allow academics the time for deep thinking and impactful work, ensuring that institutional standards value the rigor of individual publications over the rapid dissemination enabled by AI.
Publishers and Editorial Boards: They must evolve from passive recipients of content to proactive guardians of research integrity. This requires professional mentorship as well as a strong mandate. Journals should leverage international conferences and workshops as key platforms for establishing a global standard on publication culture. By taking a leadership role in these discussions, journals can move from merely enforcing rules to genuinely promoting transparent disclosure of AI usage and maintaining high standards of integrity.
At the same time, editorial boards must implement stricter penalties for systemic malpractices. By combining global outreach with rigorous enforcement, they can ensure that publication opportunities remain reserved for those who adhere to the standards upheld by the international research community.
A Different Future
Additionally, we must equip graduate students and early career researchers with the tools to develop a healthy, transparent relationship with LLMs. This isn’t about prohibition; it’s about fostering an intentional partnership grounded in domain expertise, critical thinking, and intellectual integrity.
Ultimately, these recommendations alone won’t address every challenge. The solution to the challenges posed by AI lies in a complex interplay of cultural adjustments, responsible technology use, and a fundamental realignment of incentives.
As we look toward the future, we must remain open to transformative changes. Just as AI is revolutionizing how research is initiated and conducted, it will inevitably alter the way we communicate findings. We may see a shift toward dynamic digital documents and open data repositories rather than static PDFs. The traditional academic paper, which has served us remarkably well since the 1600s, must evolve. Embracing the potential for change in scholarly communication is essential, ensuring that our methods of knowledge sharing progress alongside the intelligent tools we employ to create it.
