small_Article_02_720x360.jpg

thumbnail_Article_02_720x360.jpg

Article_02_720x360.jpg

AI may be rewriting research at record speed, but publishers are struggling to keep up. An audit of 162 academic publishers (all members of STM, the international scientific, technical, medical publishers), only 56 had publicly available AI-use policies, meaning,&nbsp;<a href="https://link.springer.com/article/10.1186/s41073-025-00158-y" target="_blank" rel="noopener noreferrer">65.4%</a> of publishers&nbsp;currently offer authors no clear guidance.Researchers are adopting generative AI faster than the publishing ecosystem can regulate it, creating a widening gap in expectations, transparency, and integrity.&nbsp;As generative AI becomes central to how researchers draft, refine, and communicate their work, this policy vacuum is creating uncertainty and raising critical questions about responsible use, disclosure, and accountability. Below, we answer a few key questions that will help researchers navigate the evolving landscape of scholarly publishing.<ol><li>What do Publisher AI Policies Typically Allow or Restrict?</li></ol>Among publishers that do&nbsp;have AI policies, several norms are emerging:<ul style="list-style-type:disc;"><li>AI tools cannot be listed as authors.&nbsp;This is consistent across major publishers, as AI cannot take responsibility for the scientific accuracy of a manuscript.</li><li>Most publishers expect full disclosure of AI use.&nbsp;If a researcher uses AI for drafting, editing, or language support, they are usually required to&nbsp;<a href="https://jasmee.jp/how-to-disclose-the-use-of-ai-in-your-research-paper/" target="_blank" rel="noopener noreferrer">declare it in the manuscript</a> (often in the Methods or Acknowledgements section).</li></ul><a href="https://library.smu.edu.sg/topics-insights/disclosure-use-ai-research-manuscripts-how-are-researchers-doing-it" target="_blank" rel="noopener noreferrer">Sample AI-Use Disclosure Statement&nbsp;</a>“During the preparation of this work, the authors used Chat Generative Pre-Trained Transformer (ChatGPT; OpenAI, San Francisco, CA, USA) to enhance readability and language, aiding in formulating and structuring content. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.”<ul style="list-style-type:disc;"><li>A small number of publishers impose strict bans on generative AI.&nbsp;A minority&nbsp;<a href="https://link.springer.com/article/10.1186/s41073-025-00158-y" target="_blank" rel="noopener noreferrer">prohibit</a> the use of generative AI tools entirely, especially for drafting or rewriting scientific content.</li><li>Permitted use is typically limited to language support.&nbsp;Many policies allow AI for grammar, clarity, or formatting but discourage using AI-generated ideas, interpretations, figures, or citations.</li><li>Policies vary widely in detail and enforcement.&nbsp;Some are clear and comprehensive; others are vague especially about whether AI-generated text, images, or summaries count as acceptable. This inconsistency leaves many researchers unsure where the line is drawn.</li></ul>The emerging consensus is that you may use AI to polish your writing but not to create or interpret scientific content. However, you must be transparent about any AI use. As part of our Responsible AI initiative, we have developed an&nbsp;<a href="https://www.enago.com/responsible-ai-movement/publisher-ai-guidelines" target="_blank" rel="noopener noreferrer">updated, easy-to-navigate list of publisher AI policies</a> to help researchers ensure transparent and compliant submissions.&nbsp;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 2. Why is There So Much Variation or Lack of Policy Across Publishers?<ul style="list-style-type:circle;"><li>AI adoption is outpacing policy development. Researchers have embraced AI tools at extraordinary speed, while publishers are still debating or piloting guidelines. Many policies remain in early stages or unpublished.</li></ul>There are several systemic challenges that explain this inconsistency:<ul style="list-style-type:circle;"><li>The definition of “AI use” is broad, ambiguous, and evolving.&nbsp;AI can be used for tasks as simple as grammar correction or as complex as drafting paragraphs, summarizing literature, checking references, or generating figures. Because these use-cases vary so widely, publishers struggle to draw clear, enforceable boundaries. As The Publication Plan notes,&nbsp;<a href="https://thepublicationplan.com/2024/10/10/publisher-policies-on-ai-use-is-it-time-for-change" target="_blank" rel="noopener noreferrer">this diversity makes standardization difficult</a>.</li><li>Many publishers defer&nbsp;AI-related&nbsp;decisions to journal editors. Instead of adopting a uniform policy across all titles, some publishers leave decisions about AI use to individual journals. This creates significant variations (even within the same publishing house) and can make compliance unpredictable for authors. For instance, Springer has acknowledged that journal-level discretion can lead to&nbsp;<a href="https://www.springernature.com/in/researchers/the-source/blog/blogposts-research-integrity/maintaining-research-integrity/27798254" target="_blank" rel="noopener noreferrer">inconsistent expectations</a>. Therefore, checking target journal guidelines has become essential, because high-level publisher rules rarely capture the nuances that editors apply in practice. 3. How Are Different Regions Approaching AI Use?</li></ul>The absence of global harmonization is widening the compliance gap.&nbsp;<a href="https://www.tandfonline.com/doi/full/10.1080/13876988.2025.2529277" target="_blank" rel="noopener noreferrer">China</a> has implemented some of the strictest generative AI and data governance rules, including mandatory security assessments for AI models. Many Chinese journals require explicit AI-use declarations as part of submission.&nbsp;<a href="https://ids.org.br/en/news-post/chinese-universities-regulate-the-use-of-ai-in-academic-writing/#:~:text=Several%20Chinese%20universities%20are%20regulating,on%20theses%20written%20by%20ghostwriters." target="_blank" rel="noopener noreferrer">Chinese universities</a> and research bodies have issued clear, usage-focused guidelines emphasizing disclosure, human accountability, and restrictions on AI-generated data or interpretations.Europe’s&nbsp;<a href="https://artificialintelligenceact.eu/" target="_blank" rel="noopener noreferrer">EU AI Act</a> directly influences research workflows by emphasizing transparency, traceability, and risk classification. Funders like&nbsp;<a href="https://www.emdesk.com/horizon-2020-horizon-europe-basics-guide/horizon-europe-using-ai-in-grant-proposal-writing-responsibly#:~:text=Horizon%20Europe's%20updated%202025%20proposal,both%20transparency%20and%20evaluation%20confidence." target="_blank" rel="noopener noreferrer">Horizon Europe</a> and agencies across Germany, the Netherlands, and Scandinavia emphasize ethical AI use, documentation, and auditability. European publishers (e.g., Elsevier, Wiley) tend to have stricter, more detailed AI guidelines, aligned with broader EU digital governance priorities.The United States&nbsp;has no central AI-use mandate for research or publishing; policies are fragmented across institutions and publishers. Most US publishers provide principle-based, flexible policies, but clarity is highly variable. Universities offer&nbsp;<a href="https://www.tandfonline.com/doi/full/10.1080/01930826.2025.2560268" target="_blank" rel="noopener noreferrer">diverse interpretations</a>, from encouragement of AI literacy to outright bans in certain departments.&nbsp;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 4. Is there a Demand for Clear Policies? What are the Risks Researches Face and Why it Matters Now?Yes, according to the preliminary findings of a major survey conducted for Wiley’s AI Research Study, <a href="https://www.wiley.com/en-us/about-us/ai-resources/ai-study/publishers-role-ai/" target="_blank" rel="noopener noreferrer">ExplanAItions</a> there is a community-wide demand for clarity, consistency, and training to use AI responsibly in research communication.<ul style="list-style-type:disc;"><li>70% of researchers want publishers to provide clear AI-use guidelines</li><li>62% want concrete best-practice examples</li></ul>The academic community is ready for more consistent and well-defined norms around AI use. This growing expectation signals that the academic community is ready for clearer, more structured norms around AI use. Without clarity, well-intentioned authors risk being accused of misconduct simply for using AI tools differently than an editor expected. Good-faith disclosure becomes a liability without consistent rules. Authors may over-disclose and trigger unnecessary scrutiny, or under-disclose and be flagged for nondisclosure. Furhter, editorial decisions become subjective and unpredictable. Two editors at the same publisher may respond differently to identical AI use. As publishers lack standardized criteria to assess whether AI was used responsibly. This leaves the scholarly record vulnerable precisely when integrity risks are rising.&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;5. Given the Ambiguity in Policies, how Should Researchers Approach Using AI Tools Responsibly to Draft Manuscripts?Here are some good-practice recommendations (based on literature and evolving norms):<ul><li>Did I restrict AI to acceptable use-cases (e.g., language polishing)? Avoid letting AI generate large chunks of text, literature reviews, analyses, or interpretations without careful human oversight and validation.</li><li>Did I verify all facts, references, and claims myself?&nbsp;As human authors, retain full responsibility for facts, citations, interpretations, and overall integrity.</li></ul><ul style="list-style-type:disc;"><li>Did I add a disclosure statement? Even among publishers with policies, disclosure (e.g., in Methods or Acknowledgments) is the most common requirement.</li></ul><ul><li>Are my manuscript drafts saved to demonstrate human authorship?&nbsp;Document your draft history. Keep track of drafts, edits, and your own contributions versus AI contributions. This is especially important if questions about integrity arise later.</li><li>Did I avoid listing AI as a co-author?&nbsp;Most journals now are very clear that AI should not be credited as an author.</li><li>Did I read the journal’s AI policy?&nbsp; If there is no public policy, follow all the above-recommended practices. Additionally, consider contacting the editorial office directly for more details. 6. What Does This Mean for the Future of AI-assisted Research and Publishing?</li></ul>The publishing landscape is evolving dramatically as AI tools become more powerful and accessible. This has nudged all stakeholders to take notice and seriously reconsider research evaluation, editorial and publication strategies.&nbsp;Several shifts are imminent:<ul><li>Cross-publisher standardization is likely. More publishers will be compelled to adopt formal, clear, public policies on AI use. This will lead to setting a standardized process for ethical AI use globally and lead to a complete compliance over time.</li><li>Tools like automated pre-submission integrity checks will become common. Authors will have to treat AI as an assistant and not a co-author, maintaining rigorous oversight, and being transparent about AI’s role in their workflows</li><li>AI literacy will become a core research skill.&nbsp;Researchers and institutions will increasingly need AI literacy to use advanced AI tools responsibly and transparently.</li></ul>As AI accelerates research writing, responsible use has become a core professional responsibility. Editors are already flagging submissions unpredictably, while genuine misuse becomes harder to detect. As global regulations diverge, the absence of publisher clarity creates inequity and uncertainty. This is the moment for decisive, transparent standards—before trust erodes further.&nbsp;The longer publishers take to define boundaries, the more the gap widens between what AI can do and what journals explicitly permit. This gap is already being exploited and honest authors are caught in the crossfire.&nbsp;Until publishers finalize clearer guidelines, the best approach for researchers is simple: be transparent, verify every AI-supported output, and retain full human accountability.&nbsp;Another safe bet would be to get professional editorial help with your manuscript and research workflows to avoid being flagged and to optimize publication timelines. This will also help avoid any integrity issues that could be raised in the future. Following these best practices will not only protect your submission but also strengthen the integrity of the scholarly record.

Why 65% of Publishers Still Have No Clear AI Policy and What It Means for Researchers? – 6 Important Points to Consider

An audit reveals that 65 percent of academic publishers still have no clear AI-use policy. Learn what this means for researchers and how to navigate AI disclosure, ethics, and compliance.

Why 65 Percent of Publishers Lack AI Policies – 6 Insights for Researchers

Dr. Rebecca D'souza

thumbnail_Rebecca.png

Rebecca.png

A clear breakdown of why most publishers still lack formal AI-use policies, the risks this creates for researchers, and the practical steps scholars must take to protect their work and reputation.

thumbnail_Article_02_340x200.png

Article_02_340x200.png

small_Article_01_720x360.jpg

thumbnail_Article_01_720x360.jpg

Article_01_720x360.jpg

The rapid integration of generative AI into research workflows has sparked both optimism and growing alarm. A recent analysis in the Journal of Korean Medical Science (JKMS) found that <a href="https://jkms.org/DOIx.php?id=10.3346%2Fjkms.2025.40.e280" target="_blank" rel="noopener noreferrer">nearly one-third</a> of retractions in AI-related publications were attributed to AI misuse, signaling that integrity risks are rising rather than stabilizing.This finding echoes a broader concern highlighted by our global <a href="https://www.enago.com/responsible-ai-movement?gad_source=2&amp;gad_campaignid=23166855774" target="_blank" rel="noopener noreferrer">Responsible Use of AI (RUAI) Movement</a>, which urges researchers and publishers to balance efficiency gains with human judgment, transparency, and accountability. The integrity landscape is shifting faster than current policies can accommodate. As researchers rely more heavily on AI to accelerate writing, data handling, or literature synthesis, journals have begun tightening scrutiny. This can to lead to either rejections or worse retractions. The later not only undermines public trust but also compromise the scholarly record. Once faulty work enters the literature, it can continue to influence citations long after it is withdrawn, compounding the damage.<h2>What Types of AI Misuse are Actually Leading to Retractions?</h2>The JKMS study highlighted several recurring patterns of misuse. The most common include:<ul><li>Undisclosed AI-generated text, where authors rely on AI to produce sections of the manuscript without transparency. For example, authors submitting literature reviews written largely with generative AI, without disclosure may result in factual inaccuracies, oversimplified interpretations, or missing nuance. Editors often detect inconsistencies in tone or sudden shifts in technical depth.</li><li>Fabricated or manipulated data/images produced using AI tools. Some papers have been flagged as they had clinical images enhanced or altered with AI-based editing tools. <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC4702694/" target="_blank" rel="noopener noreferrer">In one case</a>, contrast-enhancement algorithms unintentionally removed crucial artifacts, making the data appear more conclusive than it actually was.</li><li>Hallucinated references, where AI generates plausible-looking but non-existent citations. Common in early-career submissions, students and researchers relying on AI to generate citations have submitted references to <a href="https://www.sciencedirect.com/science/article/abs/pii/S0099133325000618" target="_blank" rel="noopener noreferrer">non-existent trials or incorrectly attributed studies</a>.</li></ul>Retractions were also triggered by <a href="https://onlinelibrary.wiley.com/doi/abs/10.5694/j.1326-5377.2006.tb00504" target="_blank" rel="noopener noreferrer">methodological flaws</a> introduced or amplified by AI-generated content. In essence, the issue is that AI use is opaque, unverified, and sometimes misrepresented as genuine human scholarship.<h2>Are AI detectors Creating New Integrity Risks Through False Positives?</h2>Surprisingly, yes. Multiple independent studies have now shown that AI-text detectors produce both high false positives and false negatives at concerning rates. A <a href="https://edintegrity.biomedcentral.com/articles/10.1007/s40979-024-00155-6">2024 analysis</a> in Research Integrity and Peer Review found that several leading AI-detection tools misclassified human-written academic text as AI-generated, sometimes with high confidence, and no tool achieved consistently reliable accuracy across disciplines. Additionally, ESL and international scholars are&nbsp;<a href="https://www.sciencedirect.com/science/article/pii/S2666389923001307" target="_blank" rel="noopener noreferrer">disproportionately flagged</a> because detectors often conflate simpler writing styles with AI-generated text. This has led to documented cases of false accusations, unnecessary investigations, and strained editor–author interactions.A researcher recently shared his experience as he received reviewer reports <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC11974187" target="_blank" rel="noopener noreferrer">alleging AI misuse, solely based on detector output, later proven false</a>.False positives are creating a secondary crisis: misplaced suspicion, strained editor–author communication, and delayed decision times. This represents a new dimension of the AI-integrity crisis, where researchers are also at risk of being wrongly accused of misusing it. As part of our RUAI Initiative, we specifically warn against “automated gatekeeping,” encouraging publishers to treat detectors only as supplementary tools—not evidence.<h2>Is AI Affecting Data Integrity Beyond Text, Especially in Images and Figures?</h2>Visual-data manipulation has become one of the most rapidly evolving concerns in academic integrity. A <a href="https://link.springer.com/article/10.1007/s10676-025-09835-4" target="_blank" rel="noopener noreferrer">2025 article</a> in Ethics and Information Technology noted increasing concerns about AI-generated scientific figures, warning that AI tools can create images that look “too perfect,” complicating detection efforts and increasing the risk of unintentional manipulation. Platforms like PubPeer have also flagged a growing number of altered biomedical images, showing that enhanced contrast, noise removal, or smoothing can unintentionally distort scientific meaning, even when authors intended only cosmetic adjustments.&nbsp;Even well-meaning authors who use AI for “clean-up” or “clarity” may inadvertently violate journal policies or data-integrity norms. As a result, publishers and editors are increasingly calling for transparency statements for figure preparation, similar to transparency expectations for text.<h2>How do we Distinguish Responsible use of AI from Misuse?</h2>When integrated responsibly, AI can improve clarity, accelerate workflows, and help researchers overcome language barriers. But the key difference lies in transparency and human oversight.<h3>Best Practices while using AI for Research Publication:</h3><ul><li>Grammar support or language editing (with disclosure). For instance, using AI to refine grammar when preparing a multi-author manuscript from non-native speakers.</li><li>Brainstorming ideas or summarizing literature (verified by the researcher). This would look like asking AI for a structured outline of existing literature, then cross-checking each point manually.</li><li>Analytic assistance where models are validated and outputs are cross-checked. Here, an AI-based statistical explanation tool can be to understand a method, not to generate results.</li><li>Assisting with non-interpretive tasks such as formatting or translation.</li></ul><h3>Misuse Occurs When:</h3><ul><li>AI text output is copied verbatim without disclosure.</li><li>AI-generated findings, data, or figures are used to compensate for lack of domain knowledge without disclosure.</li><li>Researchers skip critical thinking, and rely on AI as a surrogate for expertise.</li></ul>Responsible AI use means treating AI as a tool, not a co-author or a shortcut past scientific rigor.<h2>What are the Consequences of Irresponsible AI use for Researchers?</h2>The most visible consequence is <a href="https://www.enago.com/responsible-ai-movement/resources/lessons-from-ai-retractions" target="_blank" rel="noopener noreferrer">rejection or retraction</a>, often occurring months or even years after publication. But the impact runs deeper:<ul><li>Damage to Credibility: Retractions can follow researchers throughout their careers, leading to a loss of trust with journals and collaborators as retractions or corrections become attached to a researcher’s profile in databases like PubMed and ORCID.</li><li>Institutional Repercussions: Misconduct investigations, lost funding, or damaged departmental reputation.</li><li>Harm to the Scientific Record: Flawed findings may mislead future research before they’re corrected.</li></ul>Perhaps the biggest cost is a cultural one, each case of AI misuse erodes trust in digital scholarship, making publishers more cautious and slowing scientific communication overall.<h2>What can Researchers, Editors, and Institutions do to Ensure Responsible AI Use?</h2>This is where frameworks like our <a href="https://www.enago.com/responsible-ai-movement/author-guide" target="_blank" rel="noopener noreferrer">RUAI Initiative</a> provide much-needed structure to promote transparency, disclosure, and human oversight across the research cycle. Here are a few steps researchers, editors, publishers, and institutions can take to support responsible, transparent, and integrity-driven AI use in research:<h3>For Researchers</h3><ul><li>Always <a href="https://library.smu.edu.sg/topics-insights/disclosure-use-ai-research-manuscripts-how-are-researchers-doing-it" target="_blank" rel="noopener noreferrer">declare</a> AI use in the manuscript or cover letter.</li><li>Treat AI content as preliminary and verify every fact, citation, and inference.</li><li>Use AI selectively for tasks where errors are unlikely to distort scientific meaning and can be rectified.</li><li>Avoid AI in data generation or image manipulation.</li></ul><h3>For Journals and Publishers</h3><ul><li>Implement clear, visible <a href="https://www.enago.com/academy/publishers-responsible-ai-coalition/" target="_blank" rel="noopener noreferrer">AI-use policies</a>, including acceptable and prohibited practices.</li><li><a href="https://www.sagepub.com/explore-our-content/blogs/posts/sage-perspectives/2025/06/11/ai-detection-for-peer-reviewers-look-out-for-red-flags" target="_blank" rel="noopener noreferrer">Train reviewers</a> and editors to recognize AI-related red flags.</li><li>Avoid over-relying on AI detectors</li><li>Adopt and strengthen <a href="https://www.enago.com/academy/publishers-responsible-ai-coalition/" target="_blank" rel="noopener noreferrer">post-publication monitoring</a> systems.</li></ul><h3>For Research Institutions</h3><ul><li>Integrate AI literacy and ethics into researcher training.</li><li>Encourage open conversations about AI in research groups.</li><li>Encourage responsible and transparent AI use by supporting researchers facing false-positive accusations.</li><li>Build internal guidelines that align with international integrity frameworks.</li></ul>Responsible use must become part of the research culture. Outright bans risks driving AI use underground, while unchecked adoption leads to integrity failures. A balanced position can ensures that AI enhances scholarship accompanied ethical judgment.&nbsp;Academic integrity in the AI era is now about keeping humans in the loop and ensuring that expertise, critical thinking, and accountability remain at the heart of research.

AI Misuse Behind a Third of Retractions – 7 Important Perspectives! 

Nearly one-third of AI-related research retractions stem from misuse. Discover 7 critical insights on AI ethics, false detectors, data manipulation, and how researchers can ensure responsible AI use.

AI Misuse Behind 33 Percent of Retractions – 7 Research Integrity Insights

An in-depth look at how misuse of generative AI is driving a growing wave of research retractions, the hidden risks of detection tools, and the best practices scholars must follow to protect academic integrity.

thumbnail_Article_01_340x200.png

Article_01_340x200.png

small_Article_01 (2).jpg

thumbnail_Article_01 (2).jpg

Article_01 (2).jpg

As AI tools like ChatGPT rapidly become integral to academic research, recent cases serve as powerful reminders of the need for responsible and transparent use of these technologies. Missteps in AI utilization—from accidental prompts left in manuscripts to large-scale undisclosed AI assistance—underscore the complex ethical challenges researchers and institutions now face. These lessons from AI misuse highlight the importance of oversight, transparency, and shared accountability in the evolving landscape of academic publishing.&nbsp;&nbsp;Lessons from AI Misuse in Research1. Physica Scripta – The ChatGPT Prompt That Slipped ThroughA paper was retracted after a researcher inadvertently left the prompt “<a href="https://futurism.com/the-byte/paper-retracted-authors-used-chatgpt?">Regenerate Response</a>” (from ChatGPT) in the manuscript text. The incident was flagged by Guillaume Cabanac, who&nbsp;monitors AI-generated publications. The journal, produced by IOP Publishing, emphasized that undisclosed use of AI violated its ethical guidelines. Although, likely accidental, the inclusion exposed the undisclosed use of generative AI.Lesson:&nbsp;Even seemingly minor cues betray AI involvement. Without full disclosure and human review, such slips can compromise the credibility of an entire manuscript.2. Neurosurgical Review – A Deluge of AI-Generated SubmissionsIn one of the most dramatic incidents to date, Neurosurgical Review (Springer Nature)&nbsp;<a href="https://retractionwatch.com/2025/02/10/as-springer-nature-journal-clears-ai-papers-one-universitys-retractions-rise-drastically/">retracted at least 129 articles</a>—all generated or significantly aided by large language models without disclosure.&nbsp;&nbsp;A bulk of them were sent by researchers from Saveetha University in Chennai, with around 87 papers traced to this institution. The editor-in-chief determined these pieces breached policy since the use of large language models wasn't disclosed.Lesson: Undisclosed AI use at scale can flood journals and overwhelm peer review systems. The result: mass retractions and institutional scrutiny.3. AI in Complex Medical Topics – The Chiari and Glioma CasesTwo separate retraction notices in February 2025 addressed AI-assisted papers: one on <a href="https://link.springer.com/article/10.1007/s10143-025-03380-w/">Chiari malformation</a> and another on <a href="https://link.springer.com/article/10.1007/s10143-025-03349-9">rare gliomas</a>. Both originally published in September 2024. While the data may have been sound, the language and references had been manipulated by using AI without transparent authorship or vetting.Lesson:&nbsp;Even specialized, peer-reviewed articles are not immune. Subtle language shifts, misused terms, or hallucinated citations can undermine trust, especially in high-stakes medical research.&nbsp;Why Responsible AI Use Must Be a Shared CommitmentThe growing use of AI tools like ChatGPT in research brings both promise and pitfalls. To navigate this evolving landscape, the Responsible Use of AI (RUAI) movement calls for a coordinated response from all stakeholders—researchers, institutions, and publishers alike.While the risks are real, they can be mitigated with shared accountability, transparent practices, and active education. Here are three of the most pressing challenges, and why collective action matters:Fabricated references are common:&nbsp;Studies show that AI tools frequently invent or distort citations. In one study, &nbsp;<a href="https://link.springer.com/article/10.1007/s44217-024-00333-1">52%</a> of AI-generated references did not exist. This undermines the credibility of the research and misleads readers. Human reviewers play a critical role in verifying references before publication.Fluent but flawed content raises red flags: AI-generated manuscripts can sound polished but often lack scientific rigor or contain inaccurate data.&nbsp;<a href="https://time.com/6695917/chatgpt-ai-scientific-study/">Melissa Kacena</a>, Vice Chair of Orthopaedic Surgery at Indiana University School of Medicine, has adapted her review strategy accordingly: “If I pull up 10 random references and more than one is inaccurate, I reject the paper.” Her approach reflects a growing awareness among editors and reviewers to screen for AI-driven inaccuracies.Lack of disclosure erodes trust:&nbsp;AI use is often omitted in manuscripts—whether intentionally or due to lack of awareness. The&nbsp;<a href="https://arxiv.org/abs/2411.15218">Academ‑AI dataset</a> has flagged numerous instances of undisclosed AI use in reputable journals. Some authors have even cited AI tools as co-authors, a practice that violates most editorial policies. Strong disclosure requirements remain one of the most effective safeguards publishers can implement today.To address these issues and uphold research integrity, the RUAI vision promotes the following actions:Publishers and journals are beginning to implement clear AI disclosure guidelines, include AI ethics in author instructions, and explore the use of AI-detection tools as part of the editorial workflow.&nbsp;<a href="https://www.wired.com/story/use-of-ai-is-seeping-into-academic-journals-and-its-proving-difficult-to-detect/">Wired</a> reports that Elsevier, JAMA Network, Science family, PLOS ONE, and Nature now require authors to disclose any AI use, ban AI as an author, and/or permit AI only with permission.&nbsp;<a href="https://www.ft.com/content/44d530bc-a710-4705-be3a-9d11bc23b131">Wiley’s new guidelines</a> (April 2025), advocate using AI as a companion, require disclosure when AI “altered thinking on key arguments or conclusions,” and maintain human oversight.&nbsp;<a href="https://www.sagepub.com/about/policies/ai-author-guidelines'">Sage Publishing</a> author guidelines mandate authors to disclose AI‐generated content, verify accuracy and citations, and to avoid listing AI tools as co‑authors.Universities and research institutions are integrating AI ethics modules into research training, helping early-career researchers build a foundation in responsible AI use. A&nbsp;<a href="https://arxiv.org/abs/2504.02636">2025 study</a> outlines a framework advising cross‑national universities to institutionalize generative AI governance—including ethics, transparency, and peer review training—as part of research policy.&nbsp;Similarly,&nbsp;<a href="https://journals.sagepub.com/doi/full/10.1177/10783903241245423?">Sage’s editorial via COPE</a> emphasizes educating researchers on AI ethics, developing detection tools, and embedding transparency into academic workflows.Collaborative initiatives, like those led by COPE and STM, are shaping policy frameworks that promote transparency, traceability, and accountability across the research lifecycle. STM and COPE panels at the&nbsp;<a href="https://www.highwirepress.com/blog/publishers-approaches-to-ai-adoption/">STM Frankfurt Conference (Dec 2024)</a> recommend publisher‑led integration of AI checks within a human-reviewed framework, emphasizing transparency, legal/ethical oversight, and content provenance.&nbsp;By aligning on best practices and fostering a culture of transparency, the academic community can ensure that AI serves as an enabler—not a threat—to research excellence. The goal is not to curb innovation, but to guide it with integrity, accountability, and human oversight. As tools like ChatGPT become increasingly embedded in the research process, the&nbsp;<a href="https://www.enago.com/responsible-ai-movement">Responsible AI movement</a> offers a critical framework to ensure their adoption supports—not undermines—scholarly values. Upholding these standards is not just a matter of compliance; it is essential to preserving trust in science itself.&nbsp;&nbsp;Lessons for a Safer AI-Integrated Research FutureTo avoid these pitfalls, researchers, institutions, and publishers must work together to implement safeguards that&nbsp;<a href="https://www.enago.com/responsible-ai-movement">ensure responsible AI use</a>. The path forward involves both policy and practice:<ol><li>Researchers must clearly state if, when, and how AI was used in any stage of writing or data handling. Vague or omitted declarations are no longer acceptable.</li><li>Every AI-generated sentence, citation, and claim should be manually reviewed. Misplaced trust in automation leads to misrepresentation and retraction.</li><li>Set parameters for what AI can and cannot do. While grammar checks and formatting are appropriate uses, scientific interpretation, literature analysis, and reference generation must remain human-led.</li><li>Train editors and peer reviewers to detect signs of AI-generated content—unusual phrasing, shallow reasoning, or inconsistent terminology. AI-detection tools can support, but not replace, human scrutiny.</li><li>Rather than banning AI outright, the academic community should promote the development of tools that prioritize transparency, source traceability, and verifiable outputs.</li><li>Early-career researchers need AI literacy to avoid misuse. Short modules, case studies, and real-world retraction examples can build awareness of risks and promote ethical, informed use of AI in research.</li></ol>AI tools offer great potential in academic writing, from enhancing clarity to supporting data synthesis. However, without transparency, oversight, and clear boundaries, they can undermine research integrity. Incidents like the Neurosurgical Review retractions and ChatGPT-generated errors highlight that automation must never outpace accountability.Recent retractions in reputable journals demonstrate the risks of misused or undisclosed AI involvement. While AI-powered tools like ChatGPT, Claude, and Gemini are revolutionizing manuscript drafting, their linguistic fluency increases the risk of unintentional errors, with real-world consequences.AI can improve scholarly communication when used responsibly, but unchecked use risks serious harm. These retractions are not isolated incidents—they signal a need for academic norms to evolve with technology. Transparency, verification, and ethical clarity must guide this transition, with human oversight remaining at the heart of research integrity.&nbsp;

What Do Recent Cases Reveal About Responsible Use in Academic Writing?

AI tools in academic writing offer benefits but carry risks. Discover what recent retractions teach us about the need for transparency and accountability.

Responsible AI Use in Academic Writing: Lessons from Retractions

thumbnail_June_2025_Article_02.png

June_2025_Article_02.png

small_Article_01.jpg

thumbnail_Article_01.jpg

Article_01.jpg

In recent years, academic research has undergone a profound transformation, largely driven by the rapid advancement of generative artificial intelligence (AI). From drafting literature reviews to refining final manuscripts, AI-powered tools&nbsp;are now integrated into nearly every stage of the research workflow. Their ability to quickly generate, summarize, and edit text has made them indispensable for researchers striving to keep pace with the accelerating demands of scholarly communication.But alongside these benefits comes a subtle, growing risk: the temptation to accept AI-generated content without rigorous human review.Fluent Language, Hidden FlawsAI language models like ChatGPT and its successors excel at producing grammatically flawless and stylistically polished text. For many researchers—especially those working in a second language—these tools offer valuable assistance through the process of scientific writing. They reduce the burden of repetitive tasks and allow more focus on scientific discovery.However, that same fluency can be deceptive. Because AI-generated text often looks professional, it can create a false sense of reliability. When the writing is smooth and confident, researchers may overlook factual inaccuracies or subtle shifts in meaning.The risks aren’t always obvious.&nbsp;<a href="https://onlinelibrary.wiley.com/doi/full/10.1002/leap.1626?msockid=1a8f778c15786b423f586286148e6abb">AI mistakes tend to be nuanced</a>: a slightly inaccurate technical term, a modifier that changes a sentence’s meaning, or a confident claim that’s misleading. For instance, an AI might state that a drug “prevents” a disease, when the&nbsp;more accurate phrasing&nbsp;would be “reduces the risk of.” While this might seem minor, such distinctions carry real weight in scientific literature.Worse still, AI tools are trained on large datasets that may include outdated, biased, or incorrect information. This can lead to "hallucinations": confidently presented but entirely fabricated facts, misinterpreted concepts, or even non-existent references. Because these outputs are so polished, they often slip through unnoticed—even by experienced researchers.How AI Errors Undermine Good ScienceThe consequences of unverified AI-generated content are&nbsp;becoming increasingly visible. Journals and preprint platforms are beginning to&nbsp;<a href="https://www.nature.com/articles/d41586-025-01180-2">flag</a>, reject, or retract papers that contain errors introduced by AI tools. These aren’t cases of fraud or data manipulation—they’re the unintended results of misplaced trust.A recent peer-reviewed&nbsp;<a href="https://www.nature.com/articles/s41598-023-41032-5">study published in Scientific Reports</a> systematically evaluated the prevalence of fabricated bibliographic citations produced by ChatGPT-3.5 and ChatGPT-4. The researchers found that 55% of citations generated by GPT-3.5 and 18% of those from GPT-4 were fabricated—that is, they did not correspond to any real scholarly work. The study also identified a significant rate of substantive errors in the citations that did refer to real papers.Another well-documented example of how subtle&nbsp;shifts to technical terminology—introduced by AI tools—can undermine perceived author expertise is the case of the phrase “<a href="https://retractionwatch.com/2025/02/10/vegetative-electron-microscopy-fingerprint-paper-mill/">vegetative electron microscopy</a>.” This nonsensical term appeared in nearly two dozen published scientific papers after an AI mistranslation jumbled the established phrase “electron microscopy of vegetative structures.” The error, unnoticed by peer reviewers, raised serious concerns about the authors’ subject-matter knowledge and the credibility of the research, despite otherwise solid underlying work. This incident illustrates how even minor shifts in technical language can signal a lack of expertise and lead to questions or outright rejection in academic publishing.Such incidents are becoming more common. They damage not only individual reputations but also the credibility of scientific communication as a whole. Perhaps most troubling is that these errors are often made in good faith. Researchers may simply not realize that the AI has introduced inaccuracies or misrepresented their intended meaning.Understanding the Limits of AI: Context and Judgment MatterAt the heart of the issue is a fundamental limitation of current AI models: they don’t understand science—they only predict language. These systems operate by identifying patterns in their training data, not by comprehending the meaning, context, or implications of what they generate.This means they can’t reliably distinguish between a proven fact and a speculative claim unless explicitly told. They don’t grasp the rationale behind a methodological decision or the weight of a cautiously worded limitation. These are the subtleties that&nbsp;demand human judgment—something no AI can replicate.Domain expertise, critical thinking, and contextual understanding remain irreplaceable. Only a researcher familiar with the field can ensure that AI-generated language accurately reflects the science it’s meant to support.As AI becomes increasingly embedded in the academic writing process, it's tempting to view it as a complete solution. However, this example underscores a critical truth: while AI tools can enhance grammar and polish, they often fall short in detecting logical inconsistencies or contextual ambiguities—areas where human judgment remains irreplaceable.Consider Original text, “Therefore, it could be possible that the feasibility of a throwing estimation considering individual differences. However, since our evaluations were limited to a specific age group, we have studied the widening of age groups to obtain new insights. As one of our future works, we deal with these limitations.” To be copy edited by LLMs and a human editor.&nbsp;Copy Edited text output using Deepseek"Therefore, developing throwing performance estimations that account for individual differences may be feasible. However, since our current evaluations were limited to a specific age group, future studies should expand the age range to gain new insights. Addressing these limitations will be a focus of our future research."Key Improvements:<ul><li>Restructured the first sentence for grammatical correctness and clarity.</li><li>Changed "could be possible" to "may be feasible" for more precise academic language.</li><li>Replaced "we have studied" with "future studies should" to better reflect research intentions.</li><li>Improved the flow between sentences with better transitions.</li><li>Strengthened the final statement about future work.</li><li>Maintained consistent academic tone throughout.</li></ul>This version keeps your original meaning while making the text more polished and suitable for academic publication. Copy Edited text output using ChatGPT"Therefore, it may be possible to estimate throwing feasibility while accounting for individual differences. However, since our evaluations were limited to a specific age group, we have explored expanding the age range to gain new insights. Addressing these limitations is one of our future research directions."How AI Helped—But Only Partly:Despite its grammatical issues, the intended meaning of the original text seemed to center on evaluating the possibility of a throwing estimation while accounting for individual differences. However, the phrasing “it could be possible that the feasibility…” was logically and stylistically unclear.Two different AI-generated revisions helped improve grammar and tone, but they also carried over some logical flaws. Phrases like “these limitations” were left ambiguous, and the relationship between the limited age group and the need to widen the age range remained underdeveloped. AI smoothed the surface—but didn’t fix the structure.Human Review:Remark 1: The intended meaning of “it could be possible that the feasibility of a throwing estimation” was slightly unclear. We infer that you are referring here to the throwing estimation being possible, and we have revised the text accordingly.&nbsp;Remark 2: The meaning of this text is unclear; in what way are the widening of age groups related to the limitation of specific age groups?"Therefore, throwing estimation considering individual differences may be feasible. However, since our evaluations are limited to a specific age group, we have studied the widening of age groups to obtain new insights. In a future work, we plan to address this limitation."Where Human Editing Added Real Value:The human editor stepped in to do what AI couldn’t:<ul><li>Clarified vague phrasing by rewriting “it could be possible that the feasibility…” as “throwing estimation considering individual differences may be feasible.”</li><li>Resolved a key logical flaw by recognizing that only one limitation (age group restriction) was mentioned and revised “these limitations” to “this limitation.”</li><li>Inferred contextual meaning by referencing ideas from elsewhere in the manuscript to strengthen argument flow and anticipate reviewer concerns.</li></ul>This version not only reads better but is also logically sound, contextually appropriate, and publication-ready—thanks to human editorial insight.Safeguarding Research Integrity: A Human-AI PartnershipThe solution isn’t to abandon AI, but to use it responsibly. AI should be treated as an assistant, not an authority. Every output—from individual sentences to full manuscripts—needs careful human review. That includes checking for shifts in meaning, confirming facts and citations, and ensuring that the text faithfully represents the researcher’s intent.Universities, academic journals, and funding organizations are starting to establish policies and guidelines to address the use of AI tools. Many now require authors to disclose AI use and to certify that all content has been&nbsp;<a href="https://www.wiley.com/en-in/publish/book/ai-guidelines?utm_source=newsroom&amp;utm_medium=pr&amp;utm_campaign=AIAuthorGuidelines25&amp;utm_content=presspost">reviewed by a human expert</a>. These policies aren’t red tape—they’re essential safeguards for research quality and credibility.Generative AI is changing the landscape of academic research, offering tools that boost productivity and streamline communication. But these benefits come with responsibilities. Relying on AI without human oversight invites subtle errors that can undermine even the best research—not through carelessness or dishonesty, but through unchecked automation.The future of science depends on a balanced partnership between AI and human expertise. Researchers must be trained in AI literacy, and institutions should implement AI review checkpoints to critically evaluate AI-assisted work before publication. To protect the integrity of research, human judgment must remain the final checkpoint—ensuring that every claim, every citation, and every conclusion is accurate, intentional, and trustworthy.&nbsp;

Getting AI Right Starts with Getting Human Oversight Right

AI tools can improve research writing, but human oversight is essential to catch errors, clarify meaning, and maintain academic integrity.

Why Human Oversight Matters in AI-Assisted Research

thumbnail_May_2025_Article_01.png

May_2025_Article_01.png

Explore the latest articles, podcasts, webinars, and lectures on ethical AI use in manuscript preparation to stay informed in this fast-evolving field.

Dr. Rebecca D'souza

Articles Published:

Dr. Rebecca D'souza

Why 65% of Publishers Still Have No Clear AI Policy and What It Means for Researchers? – 6 Important Points to Consider

Dr. Rebecca D'souza

AI Misuse Behind a Third of Retractions – 7 Important Perspectives!

Dr. Rebecca D'souza

What Do Recent Cases Reveal About Responsible Use in Academic Writing?

Dr. Rebecca D'souza

Getting AI Right Starts with Getting Human Oversight Right