AI-Written NIH Grants Win More Funding But Favor Safe Ideas

Scientists using artificial intelligence to write grant applications are securing more funding from the National Institutes of Health, but the practice may be steering research toward safer, more conventional projects at the expense of breakthrough discoveries.

A study published this month in the Proceedings of the National Academy of Sciences examined more than 125,000 grant applications submitted to the NIH and National Science Foundation between 2021 and 2025. Using word-distribution modeling to detect large language model involvement, researchers identified a sharp increase in AI-generated proposals following the late 2022 release of widely available generative AI tools.

Different agencies, different outcomes

The research revealed striking differences between how the two major federal science funders respond to AI-assisted applications. NIH submissions showing high LLM involvement corresponded to a four-percentage-point increase in funding probability compared to proposals with minimal AI use. The NSF showed no such correlation.

Successful NIH grants written with heavy AI assistance also generated 5 percent more publications than those with less AI involvement, according to the study. However, this higher publication volume didn't translate to greater scientific impact—among the most cited papers, projects from AI-heavy applications showed no notable advantage.

Yifan Qian, a research assistant professor at Northwestern University's Kellogg School of Management and co-author of the study, told Inside Higher Ed that the research was possible because the team obtained confidential submission data from two large research universities, supplementing publicly available information about awarded grants.

The conformity problem

Across both agencies, proposals with high LLM involvement more closely resembled previously funded projects. The researchers emphasized this convergence reflects substantive shifts in how proposals position their research, not merely surface-level language changes.

"A portfolio that is closer to recent funding patterns may reflect improved clarity, tighter alignment with reviewer expectations, or lower transaction costs in articulating a fundable project," the study noted. "But it also implies reduced exploration in the idea landscape, which matters for public funders explicitly tasked with sustaining high-variance discovery."

Why it matters

Federal research funding transforms public investment into scientific knowledge, making the forces that shape grant decisions critical for both scientific progress and public accountability. If AI tools systematically favor incremental research over novel exploration, the long-term pipeline of breakthrough discoveries could narrow even as short-term publication metrics improve. The findings suggest agencies may need to redesign review processes to counteract AI-induced conformity.

Policy responses

Both agencies have established AI use policies emphasizing research integrity. The NSF encourages disclosure while holding applicants responsible for accuracy. In September 2025, amid surging application volumes, the NIH implemented stricter rules considering applications substantially developed by AI a violation of expectations that proposals represent original ideas.

Qian suggested agencies need more detailed guidance on acceptable AI use—for instance, clarifying whether researchers should draft proposals themselves before using AI for grammar checking.

These findings were first reported by Inside Higher Ed.