Real retractions, re-reviewed.

What automated review can — and cannot — catch

PeerReviewAI evaluates what is in the manuscript: the structure and methods as described, the statistics as reported, whether references exist, and whether cited sources actually support the claims attached to them. It cannot see what is not in the manuscript. Fabricated or falsified data, manipulated images or values, misconduct in how research was conducted, and fraud in the submission process (such as peer-review manipulation) leave no trace in a manuscript’s text — no manuscript-analysis tool can catch those, ours included. Nor is it an AI-writing detector: it does not attempt to judge whether a manuscript’s text was written by a human or by AI — it evaluates what the manuscript claims against what its sources and methods can support. Every case study below reports both sides: what the review flagged, and what it could not have caught.

How we choose cases.

• We only cover published retractions whose public notices document problems visible in the manuscript itself — fabricated or unverifiable references, claims unsupported by their sources. We avoid retractions that are contested, litigated, or attributable to honest error.
• We never name authors in our case studies or posts; we publish only findings that are machine-reproducible against public records. The story is the system, not a person.
• Every case study reports what the review caught and what it could not catch. If a review misses something manuscript-visible, we publish that too.
• Checks are run on the published version, after retraction, and results are never silently updated. They are historical artifacts of the engine version noted on each page; dated revisits may be appended, never substituted.
• We never publish reviews of customer manuscripts. Customer work is processed under Zero Data Retention and never stored.