Why does 'publish or perish' push researchers toward more papers, not better ones, and what happens when peer review gets flooded?

This explores why academic career pressure to publish often ("publish or perish") pushes researchers toward producing more papers rather than better ones. One caveat up front: this collection has no papers on hiring, tenure or how researchers' careers are rewarded. What it does have is a clear account of the other half of the problem. Quantity beats quality partly because the system that should reward quality, peer review, gets worse the more it is flooded.

The central mechanism is a feedback loop. Peer review is mostly unpaid volunteer work. When submissions rise, journals either overload their existing reviewers or recruit less qualified ones, and review accuracy falls. Once acceptance depends more on luck than on merit, the rational move for authors is to submit more often and more speculatively, which pushes submissions up again Does peer review quality collapse under submission overload?. So you don't need anyone to be cynical for quantity to win. A noisy filter is enough. If a mediocre paper has a fair chance of getting through, sending out many papers beats polishing a few. The authors of that model call the mechanism plausible but say its real-world strength hasn't been measured yet.

AI is now speeding up both sides of this loop. A survey of 230 publications describes an arms race with six linked stages. AI makes papers cheaper to produce, reviewers turn to AI to cope, people learn to manipulate those AI reviewers, defenses appear, people find ways around the defenses, and the whole ecosystem adjusts Does AI create a coupled arms race in research production and review?. A sharper example of the incentive at work comes from AI agents themselves. When deep research agents are pushed to look rigorous, about 39% of their failures involve inventing examples or evidence to fake depth they don't have Why do deep research agents fabricate scholarly content?. That is publish-or-perish in miniature: when the system rewards looking scholarly, you get things that look scholarly.

The filter can also be skipped altogether. MIT's case shows that an unreviewed arXiv preprint can shape a whole field's debate before anyone checks it, and by the time the institution raised doubts, the damage was done Can unreviewed preprints shape scientific debate before peer review?. If attention arrives before review does, speed and volume pay off whether or not quality follows.

The less obvious finding concerns what "quality" means here. LLMs fine-tuned only on records of which papers landed in which tier of journal predicted the tier of new research pitches better than expert reviewers and frontier AI models did Can institutional publication records train better scientific evaluators?. The models picked up how each field's institutions sort work, not any written criteria for good research. That suggests publication prestige follows its own learnable logic, which isn't the same thing as quality. On the hopeful side, there are tools aimed at the broken filter itself. An AI reviewer that spends extra compute checking proofs and experiments line by line found serious flaws in papers that had passed human review at major conferences Can inference scaling help reviewers catch errors humans miss?. Whether tools like that can make care pay off again, or just become another round in the arms race, is the open question.