Academia Counted Citations for Decades. AI Figured Out the Denomination.

There's a postdoctoral researcher named Peter Degen who spent part of last summer investigating why his supervisor's 2017 paper was suddenly being cited every few days — not because the field had caught up to it, not because some landmark study had sent readers back to the original, but because something else entirely was happening. Published years earlier, the paper had accumulated a respectable few dozen citations over its lifetime. Then the pace changed. Hundreds of citations, arriving in a cluster, elevating it to among the most-cited work of a senior researcher's career. A writer at The Verge covered what Degen found.

I'm less interested in the finding than in the fact that this is now a detective story academics have to run.

The Currency Got Counterfeited

Citations were always a proxy. A shorthand for influence, rigor, relevance — the assumption being that if many researchers in many papers pointed back to a source, that source must have mattered. Peer review trusted this. Hiring committees trusted this. Grant bodies trusted this. The entire architecture of academic credibility was, in part, a citation count with better furniture.

What AI-generated research papers are apparently doing now is citing fluently. Not necessarily accurately, not necessarily meaningfully — but fluently, in the way that a system trained on the patterns of academic writing learns that papers cite things, that certain kinds of claims require certain kinds of references, and that a 2017 paper on statistical analysis of epidemiological data is exactly the sort of thing you'd gesture toward when making a methodological point. Whether the paper actually supports the claim is, to the model, a secondary concern. The citation is structural. It's punctuation.

The citation system never really measured what it claimed to measure. It measured legibility within a particular genre. And the genre just got a very fast, very tireless new contributor.

What the Flood Reveals

The Verge piece frames this as a problem for scientists, which it is. But it's also a stress test that the system was never designed to pass, because it was never designed with this kind of volume in mind. Peer review was built for a world where bad actors were slow. Fabricating a credible paper took time, expertise, and social risk. The friction was the filter.

Remove the friction and what's left? Reviewers who are already overwhelmed. Editors who can't read everything. A citation economy where the signal has been inflated past the point of reliable interpretation. Degen's supervisor didn't celebrate those hundreds of citations. He sent someone to investigate. That instinct — that something was wrong precisely because the number was too good — is exactly the kind of judgment the system itself doesn't have.

There's something almost clarifying about it. We built a credibility machine that ran on a metric, and now the metric is being gamed at scale, and the response is going to be to build more metrics to catch the gaming of the first metrics. Somewhere in that loop is the actual question: what does it mean for research to be trustworthy, and did we ever have a rigorous answer, or did we just have a number that was hard enough to fake that we stopped asking?

AI didn't break academia's citation system. It just ran the exploit fast enough that we couldn't pretend the vulnerability wasn't there.