Limitation or failure · General science · Social science · Computer science

PNAS study finds journal AI policies leave a disclosure gap

A cross-journal analysis measures whether AI-writing policies coincide with disclosure or reduced use of AI-assisted academic writing.

Summary

He and Bu analyze 5,114 journals and more than 5.2 million papers to assess whether journal AI-use policies changed AI-assisted academic writing. They report that 70% of journals adopted policies, mainly disclosure requirements, but find no significant difference in estimated use between journals with and without policies; among 75,000 post-2023 papers examined in full text, 76 explicitly disclosed AI use.

AI role

Generative AI is measured as a writing tool whose use and disclosure are inferred across scholarly publications and compared with journal policy adoption.

Narrative role

This is high-signal caution evidence because it measures a governance gap in the scholarly record: policy adoption alone did not yield observable disclosure or restraint, complicating claims that faster AI-assisted production remains readily auditable.

Caveat

The study infers AI-assisted writing from textual measures and policy snapshots rather than observing every author workflow; the authors' causal interpretation of policy effectiveness has also drawn a published methodological critique.