Discussion

AI-assisted grant proposals track funded research more closely, study finds

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#14050

Scientists whose grant proposals showed more signs of AI-assisted writing also proposed research that was less distinct from recently funded projects, a new study reports. The association was modest, but appeared across both US funding agencies examined, raising a useful question about whether AI makes grant writing more efficient while nudging research towards familiar ground.

Watch Desk analysis

What happened

The study, published in PNAS and described by PsyPost, analysed confidential applications from two large US universities alongside nationwide records of funded grants. The researchers examined 1,600 National Science Foundation proposals and 4,100 National Institutes of Health proposals, as well as 57,000 NSF and 74,000 NIH awards from 2021 to 2025.

They estimated AI involvement in abstracts using patterns in word choice, then compared each researcher’s proposals with their own other applications. Higher estimated AI involvement was associated with lower distinctiveness: comparing the 75th and 25th percentiles of AI involvement corresponded to about a five-percentile-point drop for NSF awards and four points for NIH awards.

The associations with funding and later publications differed by agency. At NIH, greater AI involvement was associated with a roughly four-percentage-point increase in funding likelihood and about five per cent more publications in the first one to two years. The funding association was not statistically significant at NSF; neither were NSF’s results for publication counts. At NIH, the publication association weakened and was no longer statistically significant for highly cited papers.

Why it matters

Grant proposals help decide which research gets time and public money. If AI-assisted writing is associated with applications closer to recently funded work, the concern is not simply that proposals sound alike. It is that the ideas put forward may cluster around what funders already support.

There is a sizeable caution in the findings: this is observational research, not proof that using AI causes scientists to choose safer topics or improves their funding chances. The team estimated AI use from abstracts, which cannot capture every way researchers might use AI, and confidential application data came from only two universities.

Our read

This is a reason to look closely at how AI fits into grant writing, not a verdict that AI is flattening science. The reported differences are modest, and the study cannot establish cause and effect. Still, its comparison of proposals with recent funding patterns gives the debate something more concrete than a hunch about increasingly polished, increasingly similar prose.

What to watch

  • Whether future studies find similar patterns across more universities and funding agencies.
  • Whether research teams can distinguish AI-assisted writing from AI-assisted idea generation.
  • Whether funders examine novelty alongside feasibility and publication expectations.

Discussion spark: Should grant panels reward proposals that build on recently funded work, or give more weight to ideas that break from it, even when they carry greater uncertainty?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.