November 2026 – MERRI Newsletter
Editorial: The impact of AI on the reproducibility crisis and meta-research
We all hear about it ad nauseam, but this editorial will address the digital elephant in the room: how generative artificial intelligence (GenAI) is reshaping academic research, and more specifically meta-research. GenAI is increasingly being used across the research process, from research question generation and data analysis to manuscript and grant writing. These tools can save time, but whether greater efficiency translates into more reliable research is a separate question.
Interestingly, AI can both support and undermine reproducibility. GenAI can offer advantages such as producing more sophisticated and consistent code, workflows, and documentation, while being known to hallucinate and fabricate data and citations1 with limited traceability due to opaque models2. Overreliance on AI tools also erodes protections against low quality research by encouraging less critical review or verification of generated content3, and can facilitate integrity breaches such as plagiarism, fraud, and academic dishonesty1. In my opinion, the crux of the issue is that the publish-or-perish research culture already encourages quantity over quality, and AI can amplify these pressures by making it easier to produce large volumes of scientific text. It is also very likely not a flood of high-quality research if its raison d’être is to create publications and not scientific progress. Could it simply be the same variable-quality work in a more elegant package, which just makes it more difficult to detect by the researchers who are already overwhelmed by the volume of it?
Meta-research as a field which may particularly be able to benefit from tools like GenAI due to the resource- and time-consuming nature of its methods, but is also where introductions of errors could be particularly costly, due to its recognition as particularly a rigorous and trustworthy form of research and the trust that is placed on it. AI tools have so far proven to be very effective in tasks such as screening and data extraction for evidence syntheses, while also causing concerns to the reproducibility of their application, namely their sensitivity to very minor changes in instructions or parameters provided4. Human oversight however still seems to be imperative to ensure rigour and quality of outputs, which also raises doubts over AI tools designed specifically to accelerate or automate systematic review tasks5. An additional potential application for AI in meta-research is the verification, validation, and quality control at a scale that would previously have been impossible and offers new promise for evaluating scientific rigour. I also wonder whether in the face of all this, as meta-researchers, we have a particular responsibility not to compromise the rigour of our work for efficiency’s sake, even while we face the very same pressures as researchers from other fields. For one, a lot of us spend large parts of our time evaluating the quality of others’ work and it would be somewhat ironic for us to lose control over the quality of our own. More importantly, as frequent participants in discourse in factors impacting reproducibility, we may also be particularly well-placed to understand how AI may be used responsibly and reproducibly in research, and we should serve as examples for good research behaviour.
Ultimately, we may not be able to do as much as we’d like to limit the misuse of AI in research, particularly when its use is embedded in a research culture whose problems we were struggling to address well before widespread use of AI - but we can still make a difference. We can improve our understanding and communication of risks associated with AI use, and continue to develop guidelines, tools, and safeguards to support reproducible research. For AI-assisted research, this includes documenting the tools and model versions used, prompts, settings, and human checks, alongside the relevant data and code. Such transparency is an essential starting point. Meta-research provides an important link between primary research and understanding the conditions under which it is conducted. If we focus our efforts on continuing to investigate the factors that shape research quality in the age of generative AI, we will still be able to identify ways to make science more reproducible.
Let us know if you have any additional thoughts on this topic! And as always, if you have any content of any kind that you would want to share with the community, email us at fiona.ramage@univ-rennes.fr or varga.monika@uni-mate.hu
Fiona & Monika
Recent highlights from the field
OSIRIS final symposium highlights practical steps towards reproducible research
The OSIRIS final symposium, held in Prague on 5–6 October 2026, brought together researchers, funders, publishers and institutional representatives to discuss the project’s findings and their practical application. Discussions explored how reproducibility checklists, training and policy guidance could be integrated into everyday research workflows across disciplines. A central message was that lasting progress requires shared responsibility, adequate resources and institutional support to sustain the project’s legacy. Read more about it here.
Recordings from the FORRT AI in Metascience online conference are available online
Recordings from the FORRT conference on AI in metascience are split into 3 playlists, one for each day of the conference, and feature 32 talks hosted on their YouTube channel (Monday: AI in the Research Pipeline; Tuesday: Replications and Reproductions; Wednesday: Workshops and Analyst Variability).
DORA introductory course on Responsible Research Assessment (free online)
The introductory responsible research assessment (RRA) course aims to empower anyone to challenge existing assumptions around research quality and impact by introducing them to the concept of responsible research assessment, its foundational principles, and how RRA aims to improve evaluation practices”. This course is free, self-paced, and designed for you to pick and choose which lessons are most relevant and useful. Each lesson includes a glossary of key terms and an annotated bibliography to help guide you if you wish to dive more deeply into different topics touched on during the lesson.” The course is available to anyone regardless of career stage or discipline, and is available free and online. It is entirely self-paced.
DORA free online training module on papermills
DORA have created an online course designed to inform researchers and others on how papermills operate and why exactly they can be problematic. The tool aims to increase awareness of how papermills operate. As we know, this issue is particularly concerning to evidence synthesis which can heavily skew results and therefore affect resulting guidelines, policies, clinical decisions, as well as future work.
New journal for replication studies and null results moves towards launch
On 29 October 2026, KnowledgeWorks Global announced that it had been selected to develop the publishing platform for the NIH-supported Journal of Replication and Reproducibility. The planned journal will provide a dedicated outlet for replication studies and null results, without charging authors or readers. Its platform will connect publications with open peer review, supporting data and code, and facilities for rerunning selected analyses.
Read more about it here: https://www.kwglobal.com/news/nih-selects-kgl/
NSF and UKRI announce plans for metascience collaboration
In a joint statement released on 8 October 2026, NSF and UKRI outlined plans to collaborate on studying and improving how research is funded, conducted and evaluated. Proposed areas include research productivity, review and funding processes, and the use of new research tools. The collaboration also supports the development of an international network of metascience units to exchange experience and explore coordinated experiments. Read more about it here.
Opportunities
The Metascience Conference 2027 (14th-16th June, Montréal) is open for submissions
The metascience 2027 conference in Montréal, Canada (14-16th June 2027) is now accepting submissions! Submissions can consist of 10-12-minute talks, posters, and 60-minute panel sessions. Submit your abstract here. The themes for this edition of the conference can be found here.
Join the ReproducibiliTea global steering committee (closes 15th November)
ReproducibiliTea global is accepting applications to join their steering committee until the 15th November 2026. From Linkedin: “As a steering committee member, you’ll be part of a welcoming and supportive team of fellow early career researchers from different countries and disciplines. Together, you’ll work on projects or develop your own ideas, support local journal clubs within our global network, connect with other initiatives, and shape the future of ReproducibiliTea.”. Apply here.
JMIR publications have a new section called “Misinformation in Science Publishing” (as of August 2026) and invites meta-research submissions
JMIR Meta Research and Research Integrity which focuses on metascience has opened a section on misinformation in science publishing, “exploring misinformation, disinformation, and malinformation as generated or perpetuated through scholarly and scientific publishing”.
Events
Metascience Conference 2027 updates
Registration for Metascience 2027 is not yet open, but the dates and location have been announced: mark June 14-16th in your calendars, and the event will take place in Montréal, Canada. For updates, join the mailing list and follow them on LinkedIn and Bluesky. We hope to see many of you there!
Data for Metascience sandpit 2027 (part in-person, part online)
This is somewhere in the middle ground between an event and a funding opportunity. UKRI are inviting applications to join a metascience sandpit “develop novel projects that can address multi-faceted metascience questions related to research and development (R&D) funding and inform R&D policymaking”. It will take place over 5 days on the 1, 2, 3, 9 and 11 March 2027. Up to £2,000,000 will be allocated to support research projects arising from the sandpit. The first three days will be in-person and the final two days will be online. Applications close on November 24th 2026.
UKRIO expert webinars on Research Integrity (November 2026 to March 2027) – online
The UK Research Integrity Office are hosting a number of webinars over the next 5 months on various themes around research integrity. Framework for positive action in research funding (19th November); RESTORE: Trust in Research (10th December); Sustaining research integrity in the context of innovation and policymaking: The Athens Statement (27th January); Generative AI in Qualitative data analysis (11th March).
European Conference on Research Integrity (ENRIO) – Bucharest, 2027
“The ENRIO Congress is a key event for the European research integrity community, bringing together research integrity officers, trainers, policymakers, researchers, funders, and other stakeholders. It provides a unique platform for exchange of practices, discussion of emerging challenges, and the development of shared approaches to strengthening research integrity across Europe and beyond.” More information can be found on the official conference website.
Fun fact
Research has shown that using GenAI allows researchers to publish up to 60% more papers than before. However, some have had particularly remarkable bursts of productivity or inspiration. One multi-disciplinary researcher posted 258 papers on SSRN in 8 months in 2026, and up to 6 in a single day. That’s nearly a paper (typically 4,000+ words) per day on average, and this doesn’t account for research or submission time!
References
- Bin-Nashwan S.A, Sadallah M., Bouteraa M. (2023). Use of ChatGPT in academia: Academic integrity hangs in the balance. Technology in Society, 75 (2023), Article 102370
- Wolfrath N., Conroy M., Kosten T., Bell D., Neupane B., Kindel J., Banerjee A., Deshpande P., Taylor B., Kothari A.N. (2023). Model Retirement Creates Reproducibility Risk in Biomedical AI Publications. https://doi.org/10.48550/arXiv.2609.04699
- Fui-Hoon Nah, F., Zheng, R., Cai, J., Siau, K., & Chen, L. (2023). Generative AI and ChatGPT: Applications, challenges, and AI-human collaboration. Journal of Information Technology Case and Application Research, 25(3), 277–304. https://doi.org/10.1080/15228053.2023.2233814
- Lai H., Sousa-Pinto B., Cao C., Moher D., Estill J., Liu J., Zhao W., Wang Y., Ye Z., Tong B., Yang Z., Luo X., Wang B., Li Y., Pan B., Zhang L., Tian J., Chen Y., Shi N., Ge L. (2026). Large language models for systematic reviews were reported to perform well but rarely with verifiable safeguards: a cross-sectional study. Journal of Clinical Epidemiology 197: 112383, https://doi.org/10.1016/j.jclinepi.2026.112383
- Clark J, Barton B, Albarqouni L, et al. Generative artificial intelligence use in evidence synthesis: A systematic review. Research Synthesis Methods. 2025;16(4):601-619. https://doi.org/10.1017/rsm.2025.16