Research Trend Observation: How Preprints Empowered by AI Help Secure Academic First Publication Priority
I. Introduction: The “Time Competition” in Academic Communication
The value realization of academic research is highly dependent on the speed at which findings are disseminated, recognized, and cited by peers. In an era where AI technology iterates on a monthly basis, the traditional academic journal publishing cycle—often ranging from several months to more than one year from submission to formal publication—has become a critical bottleneck restricting the efficient dissemination of cutting-edge knowledge. A typical practical case demonstrates that a researcher uploaded a paper to a social science preprint platform, completed AI-based format preliminary review within 2 hours, obtained an official Research Achievement Release and Preservation Certificate within 3 working days, and achieved over 300 views in the first week after release, an efficiency nearly impossible under the traditional journal peer review and publishing system.
Preprints refer to original research works that are uploaded to public digital repositories in advance without formal peer review or official journal publication. They are fundamentally reshaping the logic of academic achievement dissemination and right confirmation. The case of the DeepSeek-R1 model paper, first released as a preprint on arXiv and subsequently featured on the cover of Nature eight months later, sends a clear signal: preprints have evolved from mere manuscript drafts into core positions for high-level research to seize academic first publication priority.
Against this backdrop, this paper focuses on three core research questions: First, how does AI accelerate the production, review, and dissemination of preprint achievements? Second, what new academic research and communication ecosystem is being built through the in-depth collaboration between preprints and AI? Third, what underlying governance challenges does this academic paradigm transformation face, and what are its future development trajectories?
II. How AI Reshapes Research Productivity: Evidence from 2.1 Million Preprints
2.1 First Systematic Quantitative Research
For a long time, academic discussions on AI’s impact on research productivity have relied heavily on subjective judgments and scattered case studies, lacking systematic quantitative empirical evidence. At the end of 2025, a research team from Cornell University published a large-scale empirical study in Science, providing the first comprehensive and practical quantitative basis for this issue. Covering the period from January 2018 to June 2024, the study sampled over 2.1 million papers from three mainstream global preprint platforms—arXiv, bioRxiv, and SSRN—spanning physical sciences, life sciences, and social sciences. With extensive coverage and sufficient data volume, the findings deliver high industrial reference value.
The core quantitative findings indicate that researchers leveraging Large Language Models (LLMs) for academic writing produce approximately one-third more papers on arXiv compared to non-users, while the output growth rate exceeds 50% on bioRxiv and SSRN. This data fully proves that AI is no longer a marginal auxiliary tool for scientific research but a core enabling factor that substantially reshapes the productivity curve and output scale of academic research.
More importantly, AI is restructuring the global spatial pattern of research productivity. Statistics show that researchers from Asian academic institutions achieve a 43.0%–89.3% growth in paper output with LLM assistance, significantly higher than those in other regions worldwide. The research team points out that AI technology has effectively eliminated language barriers in academic research, enabling research communities long constrained by linguistic disadvantages to achieve leapfrog development and break the traditional uneven distribution of global research productivity.
2.2 Cross-Study Verification of Findings
Cornell University’s conclusions are not isolated and have been verified by multiple large-scale cross-platform studies. Another statistical analysis covering 1.12 million papers from arXiv, bioRxiv, and 15 core Nature journals reveals that the adoption rate of AI-assisted revision in paper abstracts and introductions has grown steadily since the popularization of ChatGPT. Computer science leads the growth rate, followed by electrical engineering and system science. Notably, AI application in methodology sections remains relatively low, indicating that current researchers adopt selective and rational AI usage in academic practice.
This differentiated application pattern reflects the academic community’s clear cognition and rational control over intelligent tools. Human researchers maintain absolute dominance in links requiring rigorous logical deduction, in-depth professional judgment, and core technological innovation. Meanwhile, AI is fully leveraged to improve efficiency in repetitive and procedural work such as literature integration, language polishing, background elaboration, and content summarization, realizing targeted efficiency improvement in scientific research.
2.3 The Quantity-Quality Paradox
It is crucial to recognize that the output growth driven by AI is not entirely beneficial, and efficiency improvement is accompanied by prominent quality paradoxes. Multiple empirical studies confirm that some AI-assisted papers exhibit better linguistic fluency and formatting standardization yet have significantly lower journal acceptance rates. Peer review feedback consistently highlights a common flaw: such papers feature polished wording but empty content and insufficient innovation, lacking substantive academic contributions and research increments.
This finding delivers vital industry implications: the popularization of AI tools has decoupled superficial writing quality from core academic quality, increasing the difficulty for journal editors and reviewers to identify high-value research. Pure publication volume can no longer objectively and accurately reflect researchers’ academic contributions. As emphasized by the research team, the core academic concern has shifted from “whether AI is used” to “how to use AI in a standardized, efficient, and high-quality manner”.
From the perspective of journal operation, the surging number of well-polished but low-innovation AI-assisted submissions has greatly increased the workload and screening cost of academic review. While empowering research productivity, AI also amplifies academic noise and the proliferation of low-quality achievements. This core paradox between efficiency and quality has become an urgent governance challenge for the current preprint academic ecosystem.
2.4 Disciplinary Heterogeneity in AI’s Impact
AI’s productivity-enabling effect shows significant disciplinary heterogeneity. Cornell’s quantitative data indicates that disciplines such as computer science and electrical engineering witness the most substantial output growth with AI assistance, while the productivity gains are relatively limited in some subfields of life sciences and social sciences.
The heterogeneity stems from differences in research paradigms, writing logic, and workflow across disciplines. AI delivers remarkable efficiency gains in disciplines relying heavily on literature review, data sorting, experimental description, and textual induction. In contrast, its application scenarios and enabling value are constrained in disciplines centered on field investigation, physical experiments, qualitative in-depth analysis, and original theoretical construction.
This unbalanced disciplinary impact suggests that AI’s disruption to the traditional academic publishing system is not an across-the-board transformation but a gradual, differentiated, and field-specific iteration. Corresponding academic governance, industrial norms, and supervision strategies must abandon unified standards and adopt refined, discipline-specific, and scenario-based regulation.
III. Paradigm Shift in Academic Publishing: From CRP to PRC
3.1 Bottlenecks of the Traditional CRP Model
Traditional academic publishing has long followed the classic CRP (Commission-Review-Publish) model: journal editors receive unsolicited submissions or commission papers, organize closed peer review, determine acceptance based on reviewer comments, and complete final typesetting, proofreading, and official publication. In the rapidly iterating academic environment, this traditional model has exposed fundamental flaws and can no longer adapt to the dissemination needs of cutting-edge research.
First, publishing power is highly centralized, leading to an imbalance between supply and demand. A paper’s access to public dissemination and academic exposure largely depends on editors’ subjective judgments and journal preferences, leaving authors and readers in a passive position and causing high-quality frontier research to be easily overlooked due to artificial screening bias. Second, the publishing cycle is excessively long, resulting in delayed knowledge dissemination. The average period from submission to initial review decision ranges from 3 to 6 months, with additional months required for final publication after acceptance, often exceeding one year in total. This seriously lags behind the monthly iteration rhythm of AI-era research and hinders the rapid circulation and innovative iteration of frontier knowledge. Third, limited journal page space creates high barriers to publication. A large number of incrementally valuable research findings, especially original works by early-career scholars and researchers from developing regions, struggle to be published, restricting the release of academic value.
Professor Su Xinning from Nanjing University pointed out that the fragmented, scattered, and weak publishing pattern of traditional journals and the one-way academic communication model violate the global mainstream trend of intensive, integrated, and networked academic publishing, making institutional reform imperative.
3.2 Rise and Logic of the PRC Model
The large-scale popularization of preprint platforms has given rise to the innovative PRC (Post-Review-Commission) academic publishing paradigm. Breaking the rigid process of traditional publishing, this model decomposes academic communication into three independent modular stages—posting, review, and commissioning—reconstructing the core logic and subject responsibilities of academic publishing.
Posting Stage: Papers can be publicly released on preprint platforms after basic format verification, compliance screening, and minimum quality checks. This stage fully decentralizes publishing initiative to authors, who independently determine the maturity and release timing of their research. Without waiting for journal review, findings are globally accessible in real time, enabling the rapid locking of academic first publication priority.
Review Stage: Global academic communities conduct open and diversified evaluations on published preprints. Unlike the closed small-scale peer review of traditional journals, the PRC model supports parallel evaluations of a single achievement by multiple research communities, delivering more transparent processes, comprehensive evaluation dimensions, and dynamic and objective feedback.
Commissioning Stage: Journals and academic publishing platforms screen high-quality preprint achievements with core innovation and academic value for professional proofreading, compilation, and official publication. In this model, journals transform from “gatekeepers” that control publication eligibility to “certifiers” and “promoters” of high-quality research, with their core function upgraded from judging publication qualification to discovering valuable achievements and empowering academic dissemination.
The core essence of the PRC model is “posting priority and pre-emptive right confirmation”. Researchers are freed from lengthy journal review cycles to publicly release findings and secure academic priority via preprints, then undergo open community peer evaluation, and finally obtain official academic certification through traditional journals, forming a closed loop of rapid first release, dynamic optimization, and authoritative recognition.
In 2022, the top international journal eLife fully adopted the PRC publishing model, abolishing the traditional binary accept/reject decision mechanism. All reviewed papers are published as “reviewed preprints” with complete public peer review comments and evaluation results. This disruptive reform has become a landmark event in the global transformation of academic publishing paradigms and provides a core reference for industrial innovation.
3.3 Global Infrastructure Layout and China’s Catch-Up Progress
The large-scale implementation of the PRC model relies on sound preprint platform infrastructure. Globally, there are more than 90 compliant preprint platforms with a cumulative document inventory exceeding 4.5 million, forming a mature open academic communication system. Major international academic publishers have actively deployed preprint services: SpringerNature’s ResearchSquare has released over 350,000 preprints; bioRxiv’s Back-to-Journal (B2J) service enables seamless manuscript transfer to cooperative journals, bridging preprint first release and formal journal publication.
China has accelerated the construction of domestic preprint infrastructure and achieved remarkable catch-up progress. In April 2024, Renmin University of China launched China’s first comprehensive preprint platform for philosophy and social sciences, filling the infrastructure gap in domestic humanities and social science open publishing. To date, the platform has established strategic cooperation with 800 journals, attracting over 40,000 registered users, releasing 66,000 papers covering 30 first-level disciplines, and accumulating more than 13 million total visits. With an ultra-fast release cycle of 6 hours to 7 working days, the platform has completely reshaped the knowledge dissemination rhythm of China’s philosophy and social sciences research.
In terms of institutional guarantees, the irrevocable licensing mechanism of preprint platforms provides solid legal support for academic priority confirmation. Researchers can adopt permanent open licenses such as CC-BY-4.0 when uploading preprints to mainstream platforms including arXiv. These licenses are irrevocable; even after formal journal publication, the original preprint content remains permanently, freely, and legally accessible to the public, institutionally establishing preprints as valid credentials for academic first publication priority.
IV. AI-Driven New Paradigm of Quality Control
4.1 Diversion Strategy: Practical Exploration of AiraXiv
The explosive growth of preprint outputs driven by AI has rendered traditional manual quality screening incapable of coping with massive daily submissions, upgrading quality control from routine work to the core of ecological governance. The industry has adopted a new governance philosophy of “countering AI with AI”, realizing hierarchical governance and precise screening of AI-generated academic achievements through intelligent technology.
In 2025, the Natural Language Processing Laboratory of Westlake University launched AiraXiv, the world’s first open preprint platform dedicated to the archiving, exhibition, and intelligent review of AI-generated academic achievements. Adopting a refined “classified diversion and differentiated governance” strategy instead of simplistic prohibition or indiscriminate acceptance, the platform builds exclusive community archives for AI-generated and AI-assisted papers. It isolates low-quality AI-generated trivial papers from disrupting traditional academic review systems while providing professional dissemination and right confirmation channels for high-quality AI-empowered research. AiraXiv has been successfully deployed in the paper submission system of ICAIS 2025, verifying its effectiveness in real industrial scenarios.
AiraXiv’s diversion governance model delivers significant institutional innovation value for the industry. Abandoning the binary opposition of “completely banning AI research” or “fully allowing AI-generated achievements”, its refined design of spatial partitioning, procedural stratification, and evaluative classification enables AI-empowered works and traditional original research to develop independently and complementarily under tailored evaluation systems, offering a novel solution for academic ecological governance in the AI era.
4.2 DeepReview: Chain-of-Thought Simulated AI Reviewer
Developed to support the AiraXiv platform, the DeepReview intelligent review system is the world’s first AI academic review tool simulating human experts’ chain-of-thought reasoning. It overcomes the defects of shallow matching and single-dimensional evaluation inherent in traditional machine review, achieving in-depth judgment comparable to manual review. Its standardized review workflow consists of three progressive and closed-loop verification links.
First, innovation verification: it conducts intelligent full-text literature retrieval and multi-dimensional content comparison to accurately identify research increments, innovative points, and differential value, verifying the academic novelty of papers. Second, multi-dimensional comprehensive evaluation: it conducts all-round structured scoring and assessment covering research methodology rationality, expression clarity, academic contribution, and logical integrity. Third, reliability verification: it inspects internal logical consistency, data-conclusion correlation, and experimental design rigor to identify logical loopholes and argumentation defects.
The efficiency gap between human and machine review is extremely prominent. Human experts require weeks or even months to complete in-depth paper review, while DeepReview outputs complete structured review comments and revision suggestions within minutes. Trained on the DeepReview-13K dataset containing 13,378 valid academic review samples, the DeepReviewer-14B model outperforms GPT-o1 and DeepSeek-R1 with winning rates of 88.21% and 80.20% respectively under standard test conditions, leading the industry in intelligent review performance.
Notably, DeepReview is positioned as an auxiliary tool rather than a replacement for human reviewers. Its core value lies in rapidly completing preliminary screening, compliance verification, and basic structured evaluation of massive preprints to filter low-quality papers, allowing human experts to focus their core cognitive resources on innovation judgment, academic dispute analysis, and in-depth academic value assessment, realizing a new human-machine collaborative efficient review model.
4.3 Credibility of Preprints: Counter-Intuitive Findings
There is a prevalent academic stereotype that preprints, lacking formal peer review, are inferior in quality to officially published journal papers. However, multiple large-scale empirical studies have drawn counter-intuitive and credible conclusions. A large-scale statistical study published on bioRxiv, covering 72,644 biomedical preprints from 2018 to 2025, systematically verifies the real quality level of preprint achievements.
The research data shows that 39.9% of preprints retain unchanged core research conclusions, 50.0% only undergo minor revisions without core viewpoint alterations, and merely 10.2% require major content modifications. More convincingly, papers submitted directly to journals without prior preprint publication have a retraction rate twice that of preprint papers.
This conclusion subverts traditional cognition, proving that preprints are not synonymous with low-quality academic achievements but can provide more reliable early frontier research information in biomedicine. Its core advantage stems from open review and global community supervision. After public release, preprints are subject to real-time questioning, error correction, supplementation, and optimization by global peers, forming a dynamic, continuous, and public quality verification system. Compared with the closed, one-time, small-scale peer review of traditional journals, this mechanism delivers stronger error correction capability and quality assurance.
4.4 UniResearch: Full-Link Support from Literature Reading to Achievement Output
Within the complete AI-assisted research tool ecosystem, preprint platforms focus on the public release, priority confirmation, and dissemination of research achievements. Full-process research efficiency improvement requires integrated tools covering literature interpretation, knowledge accumulation, and achievement creation, giving rise to UniResearch, an AI-powered research platform that builds a full-link empowerment system from literature absorption to result output.
Breaking the fragmentation and singleness of traditional research tools, UniResearch integrates massive preprint literature analysis, structured academic knowledge management, intelligent paper writing, and iterative content optimization into a unified operating system, effectively reducing researchers’ tool switching costs and repetitive textual work. Researchers can leverage the platform to quickly dissect frontier preprint findings, accumulate core knowledge points, build personal knowledge networks, efficiently absorb cutting-edge research, iterate their own works, and complete rapid preprint first release.
UniResearch forms a highly complementary collaborative relationship with preprint platforms including AiraXiv and DarXiv. It focuses on improving full-process research efficiency to solve the problem of high-quality achievement output, while preprint platforms focus on public dissemination and priority confirmation to realize academic value recognition. Together, they build a complete research innovation chain of “accelerated frontier absorption — improved achievement output — preemptive academic first publication”.
V. Challenges and Future Prospects
5.1 Ethical Controversies: The Impact of aiXiv
While the in-depth integration of preprints and AI empowers scientific innovation, it also triggers profound academic ethical and integrity controversies. Launched at the end of 2025, the aiXiv platform has become a typical controversial case in the industry. It supports full-process academic paper writing, review, and publication independently completed by AI without human author or reviewer participation, forming a closed loop of “AI writing, AI reviewing, AI publishing” that strongly impacts the core rules of traditional academic systems.
This model raises acute ethical questions for the academic community: Do fully AI-written and AI-reviewed academic achievements possess legitimate academic value, and can they be included in mainstream academic databases and cited in formal papers? Does the human-independent intelligent academic closed loop trigger systematic academic integrity risks and undermine the original nature of academic research?
Many scholars predict that the academic community will gradually split into two camps: innovators who support the rational application of AI in research, and conservatives who oppose in-depth AI intervention in academic creation and review. Current international mainstream academic publishing ethical guidelines such as COPE and ICMJE only permit appropriate AI use for auxiliary work including language polishing and formatting optimization with full disclosure; undisclosed AI application will lead to paper retraction once identified. The fully AI-operated closed-loop model of aiXiv completely breaks existing ethical boundaries, plunging core issues including author identity definition, review responsibility division, and academic integrity bottom lines into continuous debate.
5.2 Urgent Demand and Feasible Path of Governance Framework
Cornell University’s research team points out that AI’s impact on academic publishing and research ecosystems is “real yet uneven”, with significant differences in enabling effects and potential risks across disciplines, research links, and application scenarios. Uniform one-size-fits-all supervision policies cannot meet industrial development needs, making refined, differentiated, and scenario-based governance frameworks urgently necessary.
The team proposes a structured and implementable industrial governance framework: first, clarify scenario boundaries by strictly distinguishing between “AI-assisted writing” and “AI-led writing”, and formulate differentiated disclosure, review, and evaluation rules for each scenario; second, establish a transparent AI usage disclosure system, requiring authors to accurately and fully report AI application scenarios, scope, and methods during preprint release and journal submission to ensure traceable academic processes; third, innovate academic evaluation criteria, building a new evaluation system centered on core innovation, research increment, and academic value, and avoiding equating superficial indicators such as linguistic fluency and formatting standardization with academic quality.
In industrial practice, AiraXiv’s achievement diversion governance model and DeepReview’s human-machine collaborative review mechanism have formed mature and feasible governance paradigms. The core logic is to clarify human-machine division of labor: AI undertakes standardized, procedural, and repetitive work including format verification, language polishing, data screening, and preliminary review, while human researchers dominate core links including innovation judgment, value evaluation, theoretical construction, and dispute analysis. This model effectively balances AI efficiency empowerment and academic quality bottom lines, alleviates AI-induced academic governance dilemmas, and provides a viable path for standardized industrial development.
5.3 Future Development Trends
In the long run, the in-depth iterative collaboration between preprints and AI will drive three fundamental and disruptive transformations in academic research and publishing ecosystems.
First, preprint platforms will evolve from static document repositories to dynamic research accelerators. Current AiraXiv has realized real-time intelligent review and high-quality achievement recommendation for preprints; cutting-edge experimental frameworks such as AgentRxiv further enable AI agents to share, reference, and iterate research findings autonomously, promoting automated upgrading of research innovation. Future preprint platforms will completely break away from static archiving functions and evolve into AI-driven dynamic knowledge circulation networks that continuously empower global scientific research innovation.
Second, the focus of academic evaluation will shift from journal brand certification to intrinsic content value. Under the new PRC publishing model, journals’ certification and commissioning functions only serve as screening and endorsement tools for high-quality achievements, rather than determinants of academic value. Industrial resources and evaluation criteria are gradually returning to content essence. A typical example is that the Bill & Melinda Gates Foundation explicitly requires funded researchers to prioritize preprint publication and terminates payment for high traditional journal article processing charges, driving the restructuring of the academic evaluation system through practical resource orientation.
Third, human-machine collaboration will become normalized infrastructure for academic research. Future research division of labor will be highly refined: AI will fully undertake repetitive and procedural work including literature screening, format calibration, data verification, and preliminary review, greatly reducing researchers’ inefficient labor costs. Human researchers will focus on high-value core innovation links, including problem definition, methodological originality, theoretical breakthrough, and in-depth academic criticism. The implementation of integrated AI research platforms such as UniResearch is continuously transforming this human-machine collaborative research paradigm into a popular industrial norm, guiding scientific research to return to the essence of innovation.
VI. Conclusion
The two-way empowerment and in-depth integration of preprints and AI are driving two core transformations in the academic communication system: from the traditional journal-centric model to an author-centric model, and from closed screening monopoly to open and inclusive competition. The core competition logic for academic first publication priority has been fundamentally reshaped. It no longer depends on journal review efficiency and publishing resources, but on researchers’ innovative efficiency, knowledge accumulation capability, and standardized AI tool application proficiency.
In this industrial transformation, AI presents dual attributes: it acts as a core accelerator for scientific innovation and communication efficiency optimization, while also amplifying academic noise and inducing low-quality achievement proliferation; it is both a source of industrial challenges and a provider of governance solutions. The core task of the global academic community is to build a refined governance system adapted to the AI era and preprint development rules, fully releasing the innovative empowerment value of new technologies while strictly controlling academic quality and integrity. This avoids damaging the academic ecosystem through technology abuse and hindering industrial innovation due to conservative inertia.
The ultimate goal of academic publishing and communication has never been mere paper publication, but the efficient circulation, wide application, and continuous iteration of high-quality research findings. The collaborative innovation of preprints and AI essentially leverages technological empowerment and model reform to enable more valuable original achievements to be seen and applied efficiently and extensively by the academic community, continuously promoting the accumulation, iteration, and progress of human knowledge systems.
Data Sources
[1] Renmin University of China News. Building a “Chinese Solution” for Knowledge Sharing: The “Academic World” Consolidates the Digital Foundation for Constructing an Independent Chinese Knowledge System[N]. 2025-11-10.
[2] Qi M, Cao Z, Wang Q, et al. Does GenAI Rewrite How We Write? An Empirical Study on Two-Million Preprints[J]. arXiv, 2025.
[3] Natural Language Processing Laboratory of Westlake University. AiraXiv and DeepReview: Open Platform and Intelligent Review System for AI-Generated Academic Achievements[EB/OL]. Zhiyuan Community, 2025-09-15.
[4] Teixeira da Silva J A. Will AI-written and AI-reviewed preprints from aiXiv be bibliometrically accepted?[J]. Health Affairs Scholar, 2026, 4(3): qxag064.
[5] Xu R (Trans.). Surge in Low-Value Academic Submissions Enabled by AI[N]. China Science Daily, 2025-12-26.
[6] Lei X, Yao C Q, Zhao W, et al. Transformation and Countermeasures of Academic Publishing Models in the Open Science Context[J]. Chinese Journal of Scientific and Technical Periodicals, 2025.
[7] CONCERT. Quantitative Analysis of LLM Application in Academic Papers[EB/OL]. 2025-09-16.
[8] Lin C X. Competition and Application of arXiv CC Licensed Preprints After Journal Publication[EB/OL]. Feng Chia University Library Blog, 2026-01-01.
