Post #4192518

@lakens@mastodon.social The safest bet might be sticking quite closely to what you think is wrong with the manuscript, rather than whether the text was generated. You can then mention your suspicions in the confidential communication to the editor (if the journal has this). You don’t have to go through each individual issue, a few egregious examples might suffice, but if even that would be too much work, you might want to decline the review and refer to your suspicion in comm to the editor. I twice had manuscripts that on closer inspection had glaring issues suggesting they were pasted together rather than written (e.g., verbatim but out-of-context use of titles of papers, odd phrasings). In that case the suspicion was more paper mill than LLM, but there was a similar uncertainty - so I highlighted a few issues in the review but also asked the editor to consider whether this manuscript would meet their journal guidelines (in my case both journals had a paragraph on paper mills in their policy).