Study Finds AI-Assisted Grant Proposals Win More NIH Funding

A new study from Northwestern’s Kellogg School of Management found that research proposals showing stronger signs of AI-assisted writing were four percentage points more likely to receive funding from the National Institutes of Health (NIH). The study, led by researchers Dashun Wang and Yifan Qian, also uncovered a tradeoff: those same AI-assisted proposals tended to resemble ideas the agency had already funded, raising questions about whether AI could gradually steer scientific funding toward safer, more conventional research.

“Science advances by exploring ideas that don’t yet look obvious,” Wang said. “If AI increasingly learns from yesterday’s successful proposals, one of the questions we should ask is whether tomorrow’s scientific portfolio becomes less adventurous.” Continuous policy changes at NIH and the National Science Foundation (NSF) have made the funding landscape more challenging for scientists. Qian added that the patterns documented in the study “indicate that large language model (LLM) use is already reshaping how scientific ideas are articulated and evaluated in the federal funding system, with implications for research diversity, transparency and public trust in the stewardship of taxpayer-supported science.” 

The study, published in the Proceedings of the National Academy of Sciences, is among the first to examine how LLMs influence federal research funding at the point where scientists compete for resources, rather than after discoveries are made.

The researchers found that LLM use rose sharply after ChatGPT’s public release in late 2022, splitting grant writing quickly into two groups: one showing little sign of AI assistance and another relying on it extensively. Proposals with greater LLM involvement consistently appeared less semantically distinctive compared with proposals written with less AI assistance, more closely aligned with previously funded ideas, a pattern that appeared in both confidential proposals and funded awards.

At NIH, proposals with stronger AI involvement were more likely to be funded, and funded projects produced more follow-on publications, though not more highly cited “hit” papers, suggesting AI may boost productivity without increasing breakthrough discoveries. At NSF, the researchers found no significant relationship between AI use and either funding success or follow-on publications.

Qian said the findings raise a broader question than whether researchers should use generative AI to write proposals. “A central concern in science policy is maintaining a diverse and exploratory research portfolio,” he said, noting that higher LLM involvement was “consistently associated with lower semantic distinctiveness” across both agencies and both proposal and award stages, which “matters for public funders explicitly tasked with sustaining high-variance discovery.”

As for why AI-assisted proposals fared better at NIH, Qian said the team can only speculate, since both agencies rely on human peer review: “review norms may more strongly reward incremental, executable projects that yield multiple publications, and LLM-assisted drafting may help proposals conform to those established templates.” He added that federal funding “is the primary mechanism through which the United States converts public resources into scientific knowledge,” making it essential to understand what shapes that process.