Academic journals are fielding far more manuscripts since ChatGPT arrived, and the editors of one leading management title say AI accounts for nearly all of its growth, according to a Forbes column by University of Georgia professor John Drake. The influx is straining peer review, the checkpoint that decides which studies earn credibility, while a new dispute at Dartmouth College shows how murky the boundaries of AI authorship remain.
How Big Is the Surge?
The editorial team at Organization Science fed 6,957 manuscripts and 10,389 referee reports, all received since January 2021, into the Pangram detection tool, Drake reports. Since ChatGPT's late-2022 debut, submissions have risen 42%, about twice the increase the journal saw in the pandemic era. Papers judged fully human-written have actually dwindled. By early 2026, most manuscripts carried some AI signal, and the quickest-growing bracket was those scoring 70% or more.
The wider picture points the same way. Josh Dahl, who leads the ScholarOne manuscript platform at Silverchair, wrote in The Scholarly Kitchen that journals on the system took in 33% more papers in the first quarter of 2026 than in the year-earlier quarter, after 17% growth in 2025. The two figures cover different scopes, one journal across several years and one platform across twelve months, so they cannot be compared directly.
Rewards for Volume Drive the Trend
The study, produced by the journal's AI task force led by Duke University's Sharique Hasan, ties the pattern to incentives. Business schools that compete on the UTD ranking, which credits faculty for articles in 24 designated journals, increased their submissions more after ChatGPT than other schools did, and the added papers leaned toward AI-written text, according to Drake's account. Claudine Gartenberg, a Wharton professor, senior editor and coauthor, summed it up: "It's not AI on its own. It's AI plus publish-or-perish incentives."
Small Journals Take the Hardest Hit
Dahl's data show the growth is uneven. Titles that logged under 15 manuscripts a quarter in 2025 expanded 81% in early 2026, while the busiest ones, above 1,500, rose 20%. He stresses that this fits ordinary growth in expanding fields as well as low-effort AI submissions steered toward thinly staffed journals, and the numbers cannot tell which is doing more of the work.
Desk rejections offer another hint. From 2022 to 2025 they increased 72%, versus 43% for total decisions, and editors now turn away 2.49 papers at the desk for every one accepted, up from 1.69, Dahl's figures show.
Reviewers Are Turning to AI, Too
AI has reached the refereeing side as well. Over 30% of reports at Organization Science show detectable AI involvement, up from almost none before ChatGPT, and unlike human-written reports they did not line up with editors' final calls, Drake writes.
Editors are still screening effectively. Just 3.2% of submissions with AI scores of 70% or above received a revise-and-resubmit verdict, compared with 11.9% for low-AI manuscripts. The burden lands on volunteers: the journal expanded its deputy-editor bench from six to eleven and roughly doubled its senior editors, to about 60.
Dartmouth Exposes the Disclosure Gap
The question of where acceptable AI help ends has surfaced at Dartmouth. On Monday, Sept. 21, the student newspaper The Dartmouth reported that Provost Santiago Schnell's writing this year drew a median 96% "AI-written" rating from Pangram across nine works he authored alone, including a Washington Post op-ed. Five of his papers from before ChatGPT's release all read as human.
The Chronicle of Higher Education added that the four outlets carrying his scholarship this year all ask authors to disclose generative AI use. Of the four papers its reporters reviewed, Schnell acknowledged AI in two; the other two carried no note, and Pangram rated them substantially AI-generated.
Schnell says he relies on AI for language polish and copyediting, reviews the final text and answers for whatever appears under his name. He has promised to disclose AI use in future work and has pointed out that detectors can wrongly flag writing by non-native speakers. The student paper noted that Pangram rated his earlier work as human and that a University of Chicago audit found error rates near zero, yet scores remain estimates, not proof. It also reported that he declined to share his AI logs, and that a Pangram test rated his own emailed statement as fully AI-written; he said an AI tool helped him polish it.
Jeff Sharlet, an English professor at Dartmouth, told The Chronicle that students see AI use as an arms race where abstaining risks falling behind, and that an administrator promoting AI while using it undisclosed sends a discouraging signal.
What Could Relieve the Pressure?
Dahl argues that detection software addresses symptoms. In his view the deeper work is rebuilding accountability norms suited to how manuscripts are now prepared, keeping reviewers engaged and backing the journals with the fewest resources.
Drake proposes that journals could use AI to screen readability, jargon and sentence complexity before humans read a paper, while warning the technology is not yet reliable enough for wide deployment. He also frames the study as a snapshot, noting that much of the AI text it caught came from earlier models.
Taken together, the data suggest the contest is less about machines than about a reward system that counts papers faster than people can read them.
