Fear of stigma blamed as 0.1 per cent of papers declare AI use

Only one in 40 scientific papers suspected of deploying artificial intelligence (AI) writing tools admit using them, says a new study which suggests the stigma of admitting ChatGPT use might explain this exceptionally low figure.

Analysing more than 75,000 research papers published since 2023, researchers at Peking University found only 0.1 per cent, or 76 in total, explicitly disclosed the use of AI writing tools despite 70 per cent of journals having clear rules on this matter.

Nonetheless, that low disclosure rate had risen substantially from early 2023 – shortly after the launch of ChatGPT – when it was just 0.01, reaching 0.43 per cent by the first quarter of 2025, explains the paper in the journal PNAS.

“Despite this growth, transparency lags dramatically behind actual adoption,” explains the study’s authors Yongyuan He and Yi Bu, who found a far greater proportion of the papers had likely used AI writing tools for general tasks such as writing assistance, editing and readability improvements.

“For every 40 papers showing statistical evidence of AI usage, only one formally disclosed it,” concludes the paper.

Disclosure rates did not significantly differ depending on whether a journal had a policy requiring a statement on AI use or not, it adds, suggesting there is still “ambiguity regarding the extent to which usage requires formal disclosure”.

With a few exceptions, academic publishers have opted against outright bans on the use of AI with nearly all of the 5,114 journals surveyed allowing AI for writing and editing and 63 per cent permitting its use for language and grammar checking, explains the paper.

However, the failure to admit AI use might also be driven by concerns “about how such disclosures will be perceived by editors, reviewers and the wider scientific community,” the paper argues.

“[Authors] might worry that admitting to the use of AI could cast doubt on the originality of their intellectual contributions, potentially leading to stricter scrutiny, bias during peer review, or negative impact on their reputation,” it says.

“Academia has yet to establish a consensus on norms governing the use of AI, and relevant policies are still evolving, which contributes to authors’ cautious approach in deciding whether and how to disclose their use of AI.”

jack.grove@timeshighereducation.com