Nick Morley from Grounded AI here (https://groundedai.company)

We collaborated with Nature here to study the extent of fake/frankenstein citations in scholarly literature (from top 5 publishers - Springer, Elsevier, Wiley, Sage, Taylor & Francis)

We're estimating hundreds of thousands of papers affected in 2025 with hallucinated citation issues

As part of the work we analysed 20k papers generated with ChatGPT API to figure out which citation errors are characteristic of gen AI use and use that classify the errors we saw in the wild.

The world's gone mad, publishing is in a nuts state, the training data is poisoned!

I genuinely don't understand how that's still an issue. Sure, when chatgpt just came out 3.5 years ago, I could see it. But even after so much press, and blogs and everyone writing on this subject, how come no-one is verifying the citations? It's a 10min job at most, even if you don't automate it...

If only there existed a free database of all publications…

Then this would be easy to guard against