Peter Degen noticed something weird last summer. A paper from 2017 that had earned a respectable few dozen citations over the years was suddenly being referenced every few days, racking up hundreds of new citations and becoming one of the most cited papers in his supervisor's career.
Most researchers would celebrate that kind of attention. Instead, his supervisor asked him to investigate what was going on.
What Degen found points to a growing problem in scientific publishing. The surge in citations wasn't coming from human researchers building on the work. It was coming from AI-generated papers that were referencing the study, often without properly understanding what it actually said.
This matters because citations are how science builds on itself. When you're researching a topic or using AI tools to help with literature reviews, you're trusting that highly cited papers are genuinely influential. If AI systems are inflating citation counts by mass-producing papers that reference work they don't understand, it breaks that trust.
The original paper assessed the accuracy of a specific type of statistical analysis on epidemiological data. Nothing flashy, just solid methodological work. The kind of paper that should accumulate citations steadily as researchers apply those methods, not explode overnight.
For anyone using AI to help with research or staying current in their field, this is a reminder to look beyond citation counts. The metrics we've relied on to identify important work are getting gamed, not by humans trying to boost their careers, but by AI systems churning out papers at scale.
