Those questions are looming large for a scientific community that has spent years struggling with an overburdened peer-review system and now faces immense new challenges created by generative AI. The number of scientific publications has doubled in 5 years, to more than 8 million in 2025 (see graphic, below), and some publishers report an even steeper rise in submissions—fueled in part by the wider availability of large language models (LLMs), which can churn out some drafts in minutes. Editors are increasingly struggling to find willing reviewers, contributing to lag times between submission and publication that can drag on for months to more than 1 year. To get just one completed review, , according to a report this year by the Silverchair content-hosting company.

Some researchers and publishers think AI will have to be part of the solution. Automating some of the peer-review process could help speed publication and save researchers’ time, advocates say. And, in fact, a growing number of reviewers are employing these tools: In a global survey of 1645 researchers across multiple scientific fields by the publisher Frontiers, just over half of respondents said they used AI for review tasks in 2025, including to help draft comments and in some cases to review methods and statistics. That’s up from 29% the previous year—despite many journals prohibiting those uses.

But others are skeptical that machines can give reliable feedback, especially for the most important, core aspects of peer review, including assessing novelty and significance, while avoiding risks such as bias and gaming. “The peer-review crisis is real, and AI looks promising [to help], but irresponsibly rushing towards automating a lot of the judgment might backfire,” says Joachim Baumann, a postdoctoral researcher at Stanford University who studies the societal impact of AI. “We need to make sure the tools we deploy are fit for the task.”