For editors & reviewers: spotting AI fabrications

Fast signals worth a second look

• A reference that won't resolve. Pick a few citations and search the titles. If a recent, specific title returns nothing, treat the whole list with suspicion.
• DOIs that land on the wrong paper. Open a sample of DOIs and confirm the destination matches. A working link to a different work is a strong tell — see DOI hijacking.
• Real first author, unfamiliar co-authors. Invented author lists often keep one plausible lead name.
• Citations that are oddly perfect. Uniformly tidy formatting and suspiciously on-topic titles for an emerging niche can indicate generated content.
• Diffuse contamination.Fabrications are usually sprinkled — a few fake references among many real ones — rather than concentrated, so don't assume a mostly-solid list is clean.

A lightweight screening workflow

1. Run the full reference list through an automated existence-and-consistency check.
2. Triage only the flagged entries — verified references need no human time.
3. For each flag, open the source and decide: fabricated, hijacked identifier, wrong metadata, or a false alarm (a real but obscure or non-English work).
4. Return specific, checkable findings to the author rather than a blanket accusation.

Be fair about false positives

Not every unverifiable reference is fake. Legitimate books, non-English journals, standards, and very new preprints are sometimes poorly indexed. The point of screening is to focus attention, not to auto-reject — which is why a good tool reports a specific, checkable reason for every flag instead of a verdict.