Hallucinated citations are polluting the scientific literature. What can be done?

The problem

It has been reported that a Nature analysis suggests tens of thousands of publications from 2025 might include invalid references generated by AI. In plain English: some papers may cite sources that don’t exist — made-up journal titles, bogus page numbers, phantom DOIs. The phenomenon, often called “hallucinated citations,” allegedly stems from researchers using large language models to draft literature reviews and reference lists without rigorous verification. Ouch. Trust takes a hit when citations turn out to be fiction.

Why this matters — and fast

Bad references aren’t just embarrassing footnotes. They can seed the scientific record with garbage, skew literature searches, corrupt citation metrics, and mislead reviewers and policy-makers. Imagine a network effect: one fake citation begets another. Before you know it, databases and indexing services are cataloging nonsense alongside real science. That’s a worst-case spiral for reproducibility and for the scholars who depend on clean trails of provenance. Who wants their grant proposal citing a ghost paper? Nobody.

Fixes — practical and painful

There are fixes, some technical, some cultural. Publishers can start mandatory automated reference checks against CrossRef and other registries; peer reviewers and editors should verify a sample of citations; journals could require authors to attest that reference lists were machine-checked; and toolmakers can build hallucination-detection modules into authoring platforms. Better training for researchers — and clearer policies from funders and institutions — will help, too. Yes, it adds friction. But when the alternative is trusting a paper that points to thin air, friction is preferable.

What’s next?

The key moment here is accountability. Will the community treat this as a nuisance or as a structural problem that needs standards, verification layers, and swift corrections? Tech can help, but humans must own the output. After all, automation created the problem; human systems will have to clean it up. Sound familiar? It’s like fixing a leaky roof while it’s still raining — messy, urgent, and very necessary.

Sources: reddit