Peer review is drowning in manuscripts. But some of our most useful contributions could skip that process entirely.

In 2025, the psychology journal Cyberpsychology received 780 submissions, 55 percent more than the year before. Its editors desk-rejected 87 percent of them, and about one in eleven was turned away for hallucinated references, so citations of papers that do not exist. In the first half of 2026, the desk rejection rate rose to 91 percent.

Similarly, journals on the ScholarOne manuscript platform received 33 percent more submissions in the first quarter of 2026 than a year earlier. The editors of Organization Science say that submissions are up 42 percent since ChatGPT arrived and they attribute nearly all of the increase to AI-generated writing.

More paper, same reviewers

But the growth did not start with ChatGPT. Between 2016 and 2022, the number of articles indexed in Scopus and Web of Science rose by about 47 percent, faster than the number of practicing scientists. In OpenAlex, the number of psychology journal articles with a DOI more than doubled between 2016 and 2025, while all fields together grew by 83 percent (left panel). Both curves climbed before 2022 and then got steeper again in 2024 and 2025.

Generative AI is accelarating a trend that was there before. Kobak and colleagues estimate that at least 13.5 percent of 2024 PubMed abstracts were processed with a large language model, up to 40 percent in some subgroups. They used marker words such as “delves” (see right panel). In 2022, 3.8 of every 10,000 PubMed abstracts used “delve” in some form. In 2024, it was 66.

Then the word got famous for “AI slop”, and people started deleting it. On arXiv, “delve” dropped soon after that was pointed out while other favorites of the models kept rising and in PubMed its rate halved in 2025. THat said, a new preprint from Kobak’s group estimates that at the end of 2025, 89 percent of the biomedical papers it analyzed showed excess LLM-associated vocabulary. So the models are still being used, just better hidden.

Even the fast lanes are clogging. PsyArXiv started checking every preprint before posting it in August 2025 after an influx of low-quality submissions and tightened its rules again in March 2026, citing a rapidly increasing submission volume. arXiv’s computer science section went further and stopped taking review and position papers that have not passed peer review, because language models made them easy to churn out.

Who is supposed to read all this?

Every one of these manuscripts lands on somebody’s desk to edit and to review. Peer review runs on donated time: already in 2020 reviewers worldwide spent more than 100 million hours on reviewing, roughly 15,000 years in total. Journals paid for almost none of it; universities most likely covered the time through salaries. Many editors work for free as well; even at major clinical journals, more than a third of those surveyed received nothing for their editorial work.

Some of the reviewing is now done by the same kind of machine that wrote the manuscripts. For ICLR 2026, a large AI conference, a company that sells AI detection software concluded that around 21 percent of the 75,800 reviews were fully AI-generated. The organizers of ICML 2026 got sneakier: they hid invisible instructions in the submitted PDFs asking any language model to put two specific phrases into its review. The phrases turned up in 795 reviews from reviewers bound by a no-LLM policy. So we arrived at a point where we completely cannot trust each other (both the publishers sneaking secret instructions in, as well as authors hiding LLM use against policy)..

But not using LLMs to review is also no solution given the scope of the problem. Alberto Brandolini observed that refuting nonsense takes an order of magnitude more energy than producing it. Language models have made the production side nearly free. Checking a manuscript still takes more effort – or a LLM as well. So when a journal precludes reviewers from using LLMs to review, they better not let pass AI slop past the editor’s desk.

Why everything becomes a paper

And it’s clear why everyone does it. “Papers and not software are the currency of academic research”. And when institutions reward such paper output, methods get worse over generations of labs.

In a survey of 1,453 members of the German Psychological Society about hiring for professorships, respondents said that quantitative indicators of research output should count less than they currently do. We would like to count less, but then we keep counting regardless.

So, I am a postdoc, and as such, we cannot fix the system. If hiring and funding reward papers, producing papers is THE rational response. It’s the hiring commitees who need to reward other stuff. But what is “other stuff” exactly?

Things that help without being papers

Open data is the first and strongest case. In 2005, Wicherts and colleagues requested the data behind 141 articles in APA journals, and 73 percent of the authors did not share. The SCORE project on the credibility of social and behavioral science found shared data for only 24 percent of 600 papers from 2009 to 2018. That matters: when data and code were both available, 91 percent of the papers could be reproduced at least approximately; when the team first had to reconstruct the data, 38 percent. (Disclosure: I am one of nearly 300 co-authors on the replication paper from that project)

Sharing has limits and costs. Participant data need consent and proper anonymization, and journals such as Psychological Science accept justified constraints on public sharing. Usability is another matter: when Hardwicke and colleagues re-ran the analyses of 25 Psychological Science articles with an open data badge, only 9 reproduced without help from the original authors. Good documentation says whether reaction times are in seconds or milliseconds and how missing values are coded. It lists the exclusions that happened before the file was saved. That is real work, and it deserves some of the credit we hand out for manuscript prose. What’s nice is that in four psychology journals, articles with open data received about 20 percent more citations.

Good datasets become infrastructure. The Confidence Database started with 145 datasets, more than 8,700 participants and almost 4 million trials, and has since supported analyses no single lab could run, such as a meta-analysis across the entire database.

Generative AI strengthens the case for data. The FAIR principles asked a decade ago that data be reusable by machines as well as by people, and the machines have arrived: reproducing a study from its code and data is now a benchmark task for AI agents. In a randomized study with 103 teams, those working with AI assistance reproduced published results about as often as human-only teams (91 versus 94 percent) but caught fewer major coding errors, and teams that let the AI lead managed only 37 percent. A human still has to be in charge. The data still have to be there.

But there is an uncomfortable flip side. Open data plus AI is also a formula for mass-producing papers. Single-factor studies based on the US NHANES health survey went from about four per year until 2021 to 190 in 2024, counting only through early October. I read this as support for my argument: the damage comes from the reflex to turn every query into a manuscript.

Software is the second case. Much of experimental psychology runs on free tools such as PsychoPy, OpenSesame, jsPsych and JATOS, and plenty of our statistics on R packages such as lavaan. Each of these tools came with a paper, because a paper is what people know how to cite. The software paper is a genre that exists so that code can put on a tie and apply for citations. JOSS was built around that problem, and the software citation principles ask us to cite software itself.

Blogs are the third. The preregistration platform AsPredicted was launched in a Data Colada post. Daniël Lakens pairs his methods blog with a free online course that more than 79,000 people have enrolled in. My guess is that this course changed more statistical practice than most papers of its time, and it has no impact factor.

Short videos for peers are the fourth, and the most underused. The audience I have in mind is colleagues: five minutes explaining your paradigm or your current project, like a poster presentation that fits into a coffee break. Show an attentional capture task at its actual pace and link the materials. The evidence on video abstracts is observational: at the New England Journal of Medicine, reports with one got about 35 percent more views and no clear citation advantage. So make the video for your colleagues. Your h-index can look after itself.

What it costs and what is changing

Currently, the cost is recognition. A dataset with a DOI does not impress a committee that counts first-author papers, and a blog post impresses it even less.

But this is slowly changing. More than 27,000 people and organizations have signed DORA, which asks funders and institutions to “consider the value and impact of all research outputs (including datasets and software)“. The 847 member organizations of CoARA have committed to recognizing outputs beyond journal publications. The DFG now lets applicants list up to ten publications plus up to ten other outputs, explicitly including datasets, software packages, preprints and blog posts. And a working group of the German Psychological Society has proposed that hiring in psychology credit data sets and research software as well.

None of this helps unless we use it. A few things anyone can do next week:

• Give your next dataset or analysis script a DOI. Zenodo can archive a GitHub repository and issue a new DOI for every release, and a CITATION.cff file adds a “Cite this repository” button on GitHub.
• If your materials live in OSF projects, plan the move now. OSF stops allowing new projects on November 16, 2026, and makes all projects read-only on February 19, 2027; public projects stay viewable.
• Cite the datasets and software you used, with version numbers, and list your own non-paper outputs on your CV.
• When you review or sit on a hiring committee, open the data and the code, and ask for them when they are missing.

Papers are not going anywhere, and they should not. The operative word in the title is “everything”. Some of our work needs to be a paper. Much more of it needs to be public, documented well enough that someone else, human or machine, can pick it up.

The word publish comes from the Latin publicare, to make something public. Reviewer 2 came later.

References

• Abele-Brehm, A. E., & Bühner, M. (2016). Wer soll die Professur bekommen? Eine Untersuchung zur Bewertung von Auswahlkriterien in Berufungsverfahren der Psychologie. Psychologische Rundschau, 67(4), 250-261. https://doi.org/10.1026/0033-3042/a000335
• Aczel, B., Szaszi, B., & Holcombe, A. O. (2021). A billion-dollar donation: Estimating the cost of researchers’ time spent on peer review. Research Integrity and Peer Review, 6, Article 14. https://doi.org/10.1186/s41073-021-00118-2
• Bonnevie, T., Repel, A., Gravier, F.-E., Ladner, J., Sibert, L., Muir, J.-F., Cuvelier, A., & Fischer, M.-O. (2023). Video abstracts are associated with an increase in research reports citations, views and social attention: A cross-sectional study. Scientometrics, 128(5), 3001-3015. https://doi.org/10.1007/s11192-023-04675-9
• Brodeur, A., Valenta, D., Marcoci, A., et al. (2026). AI-assisted teams outperform AI-led teams but not human-only teams in assessing research reproducibility in quantitative social science. Proceedings of the National Academy of Sciences, 123(22), Article e2524747123. https://doi.org/10.1073/pnas.2524747123
• de Leeuw, J. R. (2015). jsPsych: A JavaScript library for creating behavioral experiments in a Web browser. Behavior Research Methods, 47(1), 1-12. https://doi.org/10.3758/s13428-014-0458-y
• Gartenberg, C., Hasan, S., Murray, A., & Pierce, L. (2026). More versus better: Artificial intelligence, incentives, and the emerging crisis in peer review. Organization Science, 37(3), 795-812. https://doi.org/10.1287/orsc.2026.ed.v37.n3
• Gärtner, A., Leising, D., & Schönbrodt, F. D. (2024). Towards responsible research assessment: How to reward research quality. PLOS Biology, 22(2), Article e3002553. https://doi.org/10.1371/journal.pbio.3002553
• Geng, M., & Trotta, R. (2025). Human-LLM coevolution: Evidence from academic writing. In Findings of the Association for Computational Linguistics: ACL 2025 (pp. 12689-12696). https://doi.org/10.18653/v1/2025.findings-acl.657
• Hanson, M. A., Gómez Barreiro, P., Crosetto, P., & Brockington, D. (2024). The strain on scientific publishing. Quantitative Science Studies, 5(4), 823-843. https://doi.org/10.1162/qss_a_00327
• Hardwicke, T. E., Bohn, M., MacDonald, K., Hembacher, E., Nuijten, M. B., Peloquin, B. N., deMayo, B. E., Long, B., Yoon, E. J., & Frank, M. C. (2021). Analytic reproducibility in articles receiving open data badges at the journal Psychological Science: An observational study. Royal Society Open Science, 8(1), Article 201494. https://doi.org/10.1098/rsos.201494
• Holzwarth, L., González-Márquez, R., & Kobak, D. (2026). Most biomedical publications show signs of LLM-assisted writing [Preprint]. arXiv. https://arxiv.org/abs/2608.10715
• Jin, S., Verhaeghen, P., & Rahnev, D. (2022). Across-subject correlation between confidence and accuracy: A meta-analysis of the Confidence Database. Psychonomic Bulletin & Review, 29(4), 1405-1413. https://doi.org/10.3758/s13423-022-02063-7
• Kobak, D., González-Márquez, R., Horvát, E.-Á., & Lause, J. (2025). Delving into LLM-assisted writing in biomedical publications through excess vocabulary. Science Advances, 11(27), Article eadt3813. https://doi.org/10.1126/sciadv.adt3813
• Lange, K., Kühn, S., & Filevich, E. (2015). “Just Another Tool for Online Studies” (JATOS): An easy solution for setup and management of web servers supporting online studies. PLOS ONE, 10(6), Article e0130834. https://doi.org/10.1371/journal.pone.0130834
• Lee, J. C. L., Watt, J., Kelsall, D., & Straus, S. E. (2021). Journal editors: How do their editing incomes compare? F1000Research, 9, Article 1027. https://doi.org/10.12688/f1000research.25620.3
• Mathôt, S., Schreij, D., & Theeuwes, J. (2012). OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44(2), 314-324. https://doi.org/10.3758/s13428-011-0168-7
• Miske, O., Abatayo, A. L., Daley, M., et al. (2026). Investigating the reproducibility of the social and behavioural sciences. Nature, 652(8108), 126-134. https://doi.org/10.1038/s41586-026-10203-5
• Naddaf, M. (2025). Major AI conference flooded with peer reviews written fully by AI. Nature, 648(8093), 256-257. https://doi.org/10.1038/d41586-025-03506-6
• Pan, X., Li, K., Kuang, Q., & Ma, Z. (2026). Does open data promote the impact of research articles in psychology? Journal of Information Science. Advance online publication. https://doi.org/10.1177/01655515261461729
• Peirce, J., Gray, J. R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., & Lindeløv, J. K. (2019). PsychoPy2: Experiments in behavior made easy. Behavior Research Methods, 51(1), 195-203. https://doi.org/10.3758/s13428-018-01193-y
• Rahnev, D., Desender, K., Lee, A. L. F., et al. (2020). The Confidence Database. Nature Human Behaviour, 4(3), 317-325. https://doi.org/10.1038/s41562-019-0813-1
• Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 1-36. https://doi.org/10.18637/jss.v048.i02
• Siegel, Z. S., Kapoor, S., Nadgir, N., Stroebl, B., & Narayanan, A. (2024). CORE-Bench: Fostering the credibility of published research through a computational reproducibility agent benchmark [Preprint]. arXiv. https://arxiv.org/abs/2409.11363
• Smaldino, P. E., & McElreath, R. (2016). The natural selection of bad science. Royal Society Open Science, 3(9), Article 160384. https://doi.org/10.1098/rsos.160384
• Smith, A. M., Katz, D. S., Niemeyer, K. E., & FORCE11 Software Citation Working Group. (2016). Software citation principles. PeerJ Computer Science, 2, Article e86. https://doi.org/10.7717/peerj-cs.86
• Suchak, T., Aliu, A. E., Harrison, C., Zwiggelaar, R., Geifman, N., & Spick, M. (2025). Explosion of formulaic research articles, including inappropriate study designs and false discoveries, based on the NHANES US national health database. PLOS Biology, 23(5), Article e3003152. https://doi.org/10.1371/journal.pbio.3003152
• Tyner, A. H., Abatayo, A. L., Daley, M., et al. (2026). Investigating the replicability of the social and behavioural sciences. Nature, 652(8108), 143-150. https://doi.org/10.1038/s41586-025-10078-y
• Wicherts, J. M., Borsboom, D., Kats, J., & Molenaar, D. (2006). The poor availability of psychological research data for reanalysis. American Psychologist, 61(7), 726-728. https://doi.org/10.1037/0003-066X.61.7.726
• Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., et al. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data, 3, Article 160018. https://doi.org/10.1038/sdata.2016.18