Early this morning, we detected a peculiar situation in one of our journals, which typically shows moderate activity.
In a span of just two hours, we received 10 distinct articles with clear indications of having been generated using AI. All share very similar patterns: identical layout, repetitive structure, LLM-style or inconsistent writing, and content of highly questionable quality.
The submissions were made by users registered in OJS with apparently valid ORCID identifiers (though hidden in all cases). This means they passed both the ORCID and PKP verification checks.
Furthermore, all articles are co-authored. The first author is the same individual (whose ORCID is public), although this does not match the user who performed the registration and submission on the platform.
We are particularly concerned that such submissions appear to be becoming increasingly frequent and consume a considerable amount of editorial time before they can be identified and rejected.
We would like to know how the community is tackling this issue (“spam-articles” – has this term been used before?).
Specific questions:
-
What is the objective of the person or persons making these types of submissions? No such article would pass the peer-review process, so it seems like a waste of time for both them and the journal.
-
What procedures or best practices are you implementing to protect yourselves against this new phenomenon of “spam-articles”?
-
What specific tools are you using to detect and block fake users or suspicious articles? Is there already a plugin or development available to detect anomalous behaviours, such as mass user registrations within short timeframes or suspicious submissions?
Any experience or recommendation will be warmly welcomed.
1 Like
Hello Marc!
I don’t have a quick solution, and think that the issue of ‘spam-articles’ and AI generated content will just continue to increase. I’d recommend looking at some guidance from COPE, for example:
Topics on AI - specifically this one: “When editors suspect AI”
There is also a good case “Handling articles produced by AI” from 2025 that could give some good info.
Kat (Head of Community and Communications, DOAJ)
1 Like
Hi Marc!
Sorry for not responding earlier. I don’t know the answer and I doubt there is a definitive one, but here are a few comments.
Spam articles were very common even before the rise of generative AI. There are different kinds of problematic submissions. I believe that massive submissions can be dealt with on the server side. As for individual submissions that seek to get published, the initial editorial screening involving competent humans is crucial. Similarly to manuscripts created by paper mills, those generated by AI tools typically show some anomalies, most commonly in references, in the language used (like tortured phrases), or there is a mismatch between author names and the identity suggested by the structure of the email addresses. It’s very important to reject a submission as soon as something suspicious is spotted, without passing it to reviewers and wasting everyone’s time.
I’d always start with checking emails and references. References can easily be checked using Crossref’s Simple Text Query. The tool matches references against metadata deposited in the Crossref database and returns a list of references linked to their corresponding DOIs. Unmatched references should be checked manually, as they may originate from print-only publications or sources that do not assign Crossref DOIs, but they may also be fabricated. The aim is to reduce the number of references that require manual verification.
Sharing information about suspicious submissions and fraud attempts in scholarly publishing is important, because it helps the community recognize patterns. Platforms like PubPeer and databases such as PPS – Problematic Paper Screener can be useful for documenting and discussing these cases.
As for the motivation behind illegitimate actions, I believe there is a wide range of possible objectives, and these activities are not necessarily a waste of time for those involved. They may be training automated systems, doing cybersecurity exercises, attempting to discredit individuals or institutions, or pursuing other objectives outside the scope of scholarly communication.
3 Likes