- Two reviewers examined 22 paper submissions and found 68% contained fabricated citations, hallucinated authors, or were clearly LLM-generated.
- Studies show tens of thousands of 2025 publications contain invalid AI-generated references, with early-career researchers and small teams most likely to include them.
- AI-generated reviews and rebuttals are also widespread, with some conferences detecting over 20% of reviews as fully AI-generated.
- Common LLM writing tells include dense sentences, bold formatting, em-dashes, hyped results, and numerical inconsistencies between text and tables.
- Both reviewers use LLMs for their own writing but thoroughly verify output, edit extensively, and manually check references.
- The authors developed a free, MIT-licensed tool called bib-audit that automatically checks bibliographies against publisher records to catch fabricated citations.
- Responsibility for slop papers is shared among authors, advisors, conference organizers, and the current incentive structure in academic publishing.