Scientists who use artificial intelligence to draft grant applications are more likely to win funding from the National Institutes of Health-but those proposals tend to resemble existing research patterns rather than explore new territory, according to a study published this month in the Proceedings of the National Academy of Sciences.
The researchers analyzed more than 125,000 grant applications submitted to the NIH and National Science Foundation between 2021 and 2025, including funded, unfunded and pending proposals. Using word-distribution modeling to flag submissions with high levels of large language model involvement, they found a surge in AI use between 2023 and 2025-coinciding with the rapid adoption of generative AI tools that became widely available in late 2022.
Federal funding is the primary mechanism the U.S. uses to convert public resources into scientific knowledge, so understanding what shapes that process matters, said Yifan Qian, a research assistant professor at Northwestern University's Kellogg School of Management and co-author of the paper. "Understanding the forces that influence the federal funding process is critical not only for science policy and scientific progress, but also for accountability of public investment in research," Qian said.
The study is among the first to examine AI's role in federally funded research, largely because proposal data is confidential. "Our team was able to get two full sets of confidential NIH and NSF submissions from two large research universities, which allowed us to study this question," Qian said.
Divergent outcomes across NIH and NSF
The two agencies responded differently to AI-assisted applications. The researchers found no association between high LLM involvement and accepted NSF proposals. But for NIH submissions, heavy AI use corresponded to a four-percentage-point jump in funding probability compared to low-LLM proposals.
The researchers then asked whether LLM-assisted proposals led to more published research-and the answer diverged again. Successful NIH applications with high LLM involvement produced 5 percent more follow-up publications than less AI-assisted equivalents. But that publication advantage didn't translate to research impact-there was no notable citation advantage among the top-cited papers for NIH-funded proposals showing heavy AI use.
The authors didn't draw firm conclusions about why the agencies differ, but offered possible explanations. "NIH funding and 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," the paper said.
AI-assisted proposals converge toward familiar ground
Across both agencies, proposals with high LLM involvement were more similar to projects that had already received federal funding. The researchers said that convergence "is not simply a byproduct of LLM-induced surface-level rewriting, but reflects shifts in the substantive positioning of proposals and awards."
"A portfolio that is closer to recent funding patterns may reflect improved clarity, or tighter alignment with reviewer expectations," they noted, adding that it also implies reduced exploration in the idea landscape-a trade-off for public funders charged with supporting risky, high-variance discovery.
Both agencies have policies addressing AI use. The NSF introduced a 2023 policy encouraging transparency about generative AI use. In September 2025, the NIH cracked down, treating applications "substantially developed by AI" as a violation of its expectation that proposals be original work.
Qian said the agencies will likely need product-specific guidance to prevent a sustained slowdown in scientific novelty. "Getting specific about whether they want people to write their own first drafts and use AI to help detect grammar errors, for example," he said. "That would be helpful."
Why this matters for Science and Research
Researchers using AI to draft grants may win more NIH funding and publish more total papers-but the evidence suggests their work gains less citation impact. When you apply for federal grants, consider how the agency's review mechanics affect your proposal's substance, not just its drafting. A proposal that reads more "standard" probably also resembles less-novel research territory. If you're applying for NIH money, a template approach may work-at least in the short term. But if the goal is to expand scientific knowledge, the paper's findings suggest the math changes. Whether you use an LLM to draft, revise, or polish, don't let the promise of a funding bump push your actual research toward where everyone else already went.
Your membership also unlocks:
