Paper mills and the integrity of the literature
Research integrity, plagiarism, research misconduct and retractions appear on 28 recent papers across the AI, finance and economics journals. Retractions have reached record numbers, and paper mills sell authorship and citations, with language models making fabrication cheaper. How much of the recent literature in these fields is affected, and can it be detected?
What prompted this
Since October 2025, 28 papers carry a research-integrity, plagiarism, research-misconduct, retraction, paper-mill, predatory or citation-manipulation keyword (13 in AI journals, 7 in finance, 4 in economics): "research integrity" on 10, "plagiarism" on 5, "research misconduct" on 3, "retraction" on 3. The tracked journals publish more than 40,000 papers a year between them, so the integrity literature is tiny relative to the output it concerns.
Questions
- What share of recent papers in the tracked journals show paper-mill signals (tortured phrases, impossible author affiliations, citation rings), by journal and country?
- Do special issues and guest-edited collections carry more such signals, and did journals that cut special issues see fewer?
- Can AI-fabricated results in empirical finance and economics be detected from their statistics, for example from the distribution of reported test statistics?
Data that could help
The full text and metadata of recent papers, retraction records (Retraction Watch), and the published detectors for tortured phrases and citation anomalies.
Next step
Run tortured-phrase and citation-anomaly detectors over the last two years of papers in the tracked journals and compare rates across journals and special issues.