Does GenAI help your proposal get accepted?

Does using Generative AI help in getting your proposals accepted? 🧐 

While I teach courses on how to use GenAI efficiently for writing research proposals, this is a question that is very hard to accurately answer as it’s not so easy to get access to the necessary data. However, some first evidence is showing up, albeit in the context of US funding 🇺🇸. A recent paper published in PNAS shows how the use of LLMs in writing project proposals led to a higher likelihood of being funded at the National Institutes of Health (NIH), while there was no effect for the National Science Foundation (NSF).

The authors speculate that this might be because NIH rewards more incremental research, while NSF funds more high-risk projects. This also aligns with their finding that proposals in which LLMs are used more, are more closely aligned with recently funded work. However, they do caution that this does not necessarily implies a causal effect – correlation is not always causation – as investigators might be more inclined to use LLMs for proposals meant to include incremental research.

As with all research, the study has several limitations. It only considers the ‘traces’ of LLM use, but does not distinguish in how it was used during the writing process. It’s also limited to proposals from two US universities and it uses a within-investigator design, meaning it doesn’t compare proposals across investigators. Hopefully more research will come out so a clearer and more complete picture can be formed.

🇪🇺 So how does this translate to the EU, and to Horizon Europe funding?

Horizon Europe is a very broad program, that invites various types of research proposals. From bottom-up to top-down defined topics, and from incremental to breakthrough research. It would make sense to use LLMs for those programs inviting more incremental research within top-down defined topics, while for more breakthrough and bottom-up defined research it’s safer to stick to a more human-led approach. However, even in the latter, LLMs could be used to polish your writing or help with the less scientific and more administrative parts.

Whatever you think about it (and there’s a whole other discussion to be had about that beyond the scope of this post), it’s clear that using Generative AI and LLMs is starting to have an impact on how proposals are written and evaluated. This is already visible in the increasing numbers of submitted proposals, and the inevitably associated lower acceptance rates.

What these numbers and the paper discussed in this post do not say, is how GenAI can be used for writing better – not just more – proposals. And that’s a conversation we do need to have in my opinion.