---
title: "The Explosion of Scientific Spam: How the AI Revolution is Breaking the Peer Review System and Devaluing Knowledge"
description: "The scientific community is facing a crisis: after the emergence of ChatGPT, the flow of articles increased by 42%, but quality dropped. The journal Organization Science analyzed thousands of works and found that AI texts are often rejected, while algorithm reviews are superficial. Scientists warn: without human control, science risks drowning in \"junk\" publications. 📉🤖🧠"
date: 2026-07-24T14:12:51.000Z
lang: en
url: https://xab.info/en/posts/explosion-of-scientific-spam-how-ai-breaks-peer-review-system
tags: [organization-science, chatgpt, james-evans, qwen3-14b, scientific-publishing]
publisher: "XAB.info"
---

# The Explosion of Scientific Spam: How the AI Revolution is Breaking the Peer Review System and Devaluing Knowledge

![Stressed scientist at paper-cluttered desk surrounded by AI holograms and scientific papers — visualizing the crisis of peer review due to AI-driven scientific spam explosion](https://xab.info/media/2026/07/24/vzryv-nauchnogo-spama-kak-ii-lomaet-sistemu-recentzirovaniya/vzryv-nauchnogo-spama-kak-ii-lomaet-sistemu-recentzirovaniya-1.webp)

The scientific community is facing an unprecedented challenge. The proliferation of generative artificial intelligence has led to a sharp surge in the number of scientific papers, creating a colossal burden on editors and reviewers. However, behind the growth figures lies a worrying trend: the quality of materials is declining, and the system for evaluating knowledge risks being overwhelmed by "junk" content.

The editorial board of the prestigious journal Organization Science conducted a large-scale analysis covering the period from January 2021 to February 2026. The sample included 6,957 submissions and over 10,000 reviews. The results of the study paint a clear picture: the emergence of ChatGPT at the end of 2022 became a point of no return.

### Figures that are alarming

After generative models became available to the general public, the number of submissions to the journal increased by 42%. By February 2026, researchers estimate that AI was used to some extent in the writing of the majority of incoming articles. To verify the data, the Pangram 3.1 algorithm was used, which assesses the probability of machine-generated text.

However, quantitative growth turned into qualitative decline. According to the Flesch readability index, which often correlates with the depth and clarity of presentation, the average quality of papers by January 2026 had decreased by 1.28 standard deviations compared to the beginning of the analysis period. This is not just a statistical error; it is a fundamental change in the nature of scientific communication.

### Editors vs. Algorithms

The scientific peer review system demonstrated high sensitivity to the "artificial" origin of texts. Papers in which the AI participation indicator exceeded 70% practically did not pass the editorial selection—in 70% of cases, they were rejected by editors before being sent to independent reviewers. For comparison, texts with minimal signs of neural network usage were rejected in only 43.7% of cases.

Furthermore, the probability of receiving a request for revision for papers with clear traces of AI dropped from 11.9% to 3.2%. This indicates that editors prefer to filter out such materials immediately, rather than spending resources on improving them. The statistical model confirmed: even if the text is stylistically impeccable and "smooth," the probability of rejection remains high. This suggests that it is the scientific content that suffers, proving to be below criticism.

### The Illusion of Peer Review

The problem also affected the reviews themselves. Texts prepared by AI turned out to be less diverse and deep. They discussed theoretical frameworks more often, but paid critically little attention to real data, methods, and experimental results. This creates a risk of superficial evaluation of research, where form begins to prevail over facts.

Nevertheless, researchers are not inclined to demonize technology. In a large-scale experiment, a group led by James Evans asked 6,749 scientists to evaluate ideas generated by language models. More than 25,000 assessments of the novelty and plausibility of hypotheses were conducted. It turned out that popular models often offer banal ideas, while more advanced algorithms are capable of surprising with novelty.

Tests were also conducted in the field of peer review. The Qwen3-14B model, specially trained on reviews prepared by scientists, showed a 27% superiority over universal models in the role of an expert. However, AI reviewers often disagreed with professional human experts.

### The Future of Science: Control or Chaos?

The authors of the study draw an important conclusion: AI has already become an indispensable assistant for data search, code writing, and draft preparation. But without strict human verification and reform of the scientific work evaluation system, technology could lead to poorly controlled growth in the number of publications. The main threat lies not in the lack of new knowledge, but in the inflation of information volume without a real increase in scientific understanding.