AI fuels surge in manuscript submissions, increasing strain on peer review

Analysis of nearly 7,000 manuscripts reveals a 42% increase in submissions and links intensive use of artificial intelligence to a declining quality of scientific writing

Rodrigo de Oliveira Andrade

An open laptop in a dark environment, with an AI chatbot interface on the screen displaying the question "What can I help with?" and options to search, reason, or create an image. The laptop, with a backlit keyboard, is on a wooden desk illuminated by streaks of blue light.

Large language models (LLMs) driven by artificial intelligence (AI), combined with academia’s intense pressure to publish, are leading to the publication of more scientific articles, but not necessarily better ones.

As a result, the editorial operations of scientific journals are becoming increasingly burdened by a growing number of low-quality submissions produced with the help of AI, affecting both editors and peer reviewers, who are usually volunteers.

This is the conclusion of a paper published in Organization Science in April. In the study, researchers from various teaching and scientific institutions across the United States analyzed how submission patterns and the journal’s own peer review process have changed since the 2022 launch of ChatGPT, the first commercial LLM chatbot to achieve widespread popularity.

The authors analyzed 6,957 manuscript submissions and 10,389 peer reviews entered directly into the journal’s editorial system between January 2021 and February 2026, covering periods both before and after ChatGPT’s release.

The first two years of the data series served as a control period, allowing the team to compare writing patterns before the widespread adoption of generative AI tools.

For the manuscripts, the analysis focused primarily on abstracts, although the researchers also examined the full text of some submissions to determine whether AI use detected in abstracts was reflected throughout the papers.

AI assistance score

The researchers analyzed the texts using Pangram, a widely used and validated tool for detecting AI-generated content.

Pangram uses a statistical mechanism that significantly reduces the risk of false positives (when text written by humans is mistakenly classified as AI-generated).

It then produces an “AI assistance score,” which ranges from 0 to 1 and determines the extent to which AI was used in the analyzed text.

Scores close to 0 indicate content predominantly written by humans, while scores close to 1 suggest texts largely produced with the aid of AI.

Worrying results

According to the researchers, the number of submissions to Organization Science has increased by 42% since November 2022, when ChatGPT was launched. At the same time, the writing quality of submitted manuscripts has declined.

Using the Flesch Reading Ease score, a standard measure of text readability, the authors found that the writing quality in manuscripts submitted in January 2026 was 1.28 standard deviations lower than that of those submitted in January 2021.

The quality of peer review reports also deteriorated sharply after ChatGPT was released, driven by an increase in reviews written largely with the assistance of generative AI.

In addition to exhibiting poorer writing quality, these reports have become narrower in scope, focusing more heavily on theoretical discussions and less on the assessment of data and empirical evidence.

In summary, more papers are being submitted to the journal, but they are increasingly composed of poor-quality AI-generated text.

“While many believe that AI improves academic writing quality, we show that AI has led to a deterioration of prose in submitted manuscripts and reviews,” the study authors wrote.

Editorial system under strain

The researchers point out that the patterns observed at Organization Science are likely widespread across other scientific disciplines.

“AI has entered academia with gusto,” the authors wrote. “This trend puts additional pressure on an editorial system that was already under stress even before ChatGPT’s introduction.”

According to the researchers, the continued increase in the number of submissions was already being fueled by highly competitive academic incentive structures, such as career advancement criteria based on publication output and financial rewards linked to scientific productivity.

“When institutional incentives push publication over rigorous research, AI will be used to drive volume rather than to fulfill its potential as a critical research technology,” wrote the authors.

“As editors who serve as the bulwark against inundating reviewers with poor-quality AI-driven submissions, we view the current system as unsustainable.”

* This article may be republished online under the CC-BY-NC-ND Creative Commons license.
The text must not be edited and the author(s) and source (Science Arena) must be credited.