I want to reflect on getting a grant, and what I think made a difference for a specific senior research grant I recently received, but also across the long stretch of my career as a grant writer.
Because it has been a long stretch. And a lot of it has involved rejection.
Three submissions before an interview
I submitted the idea behind a senior research grant twice before I was ever invited for an interview. Same research direction. Same principal investigator. Two rejections. Then, on the third attempt, suddenly I was in.
What changed?
What I believe mattered most was that I had focused the science considerably, partly based on reviewer feedback from the earlier rounds. The project felt simpler and more feasible. The message was easier. That clarity helped me communicate it, both in writing and in the interview.
The other thing I did differently: I thought more carefully about what mattered to the funder, not just to me or to the science.
This sounds obvious... and it is. But it's also easy to forget when you're deep in your own research ideas. For this particular funder, the central motivation was impact: for the patient, but also for society. The angle about unnecessary device implantations (the idea that a better test could prevent people from getting a device they don't need) was one I deliberately brought to the foreground. That's not just a scientific result. It's a healthcare and societal argument. And it landed.
The numbers, put plainly
Let me put my track record in context, because I think the honest version is more useful than a highlight reel.
I received some travel grants early on: small amounts, but hugely important. They gave me international lab experience and data that became the foundation for everything that followed. Then an early-career research grant. I also received a junior postdoctoral grant in the same year, but it overlapped exactly with the early-career grant, so I couldn't accept it. And now the senior research grant, on the third try.
What I didn't get: two different fellowships — both rejected. A major mid-career grant — rejected. The senior grant I eventually received — rejected twice before this acceptance. A technology programme: twice submitted, never granted. A few others: rejected.
I sat down recently and calculated my overall success rate of granted vs submitted proposals: 21%.
I expected it to be much lower, because all those rejections made it feel like I was getting nowhere. But this ratio is apparently slightly above average. It still means that for every grant I've received, I've written several that did not lead to funding. That's a lot of time that wasn't spent doing science.
Where the system breaks down
Perseverance matters. But I want to go a bit further than that.
I think the grant system has structural problems. Low success rates mean that a large fraction of scientific effort is going into writing proposals that don't get funded. When you're writing, you're not collecting data or developing the ideas you actually have. That's a real cost to science, not just to individual researchers.
There's also a compounding effect. Grants beget grants. If you've already secured one or two major grants, you have the time, the staff, and the track record to compete effectively for the next one. Early-career researchers (who often have the freshest ideas, and who need funding the most!) face the steepest climb, especially if they're writing in a second language or don't have the institutional support that comes with an established lab.
The system doesn't select for the best ideas. It selects for the best proposals, which is related but not the same thing.
What AI can (and can't) change here
I do think we're at a turning point, because AI has become a genuinely useful writing assistant. There's research suggesting that when you give people a writing task and access to generative AI, everyone improves, especially the weakest ones (and let's be honest - most of us are better scientists than grant writers...).
I've used chatbots myself for drafting and refining proposal sections. My practical recommendation: use Claude rather than ChatGPT for writing tasks. It handles nuance and scientific tone better, in my experience. But for longer grant workflows, chatbots lose their train of thought. You end up copy-pasting back and forth, re-explaining the context, and manually tracking which section you're working on. It's useful, but it's not structured for the task at hand.
That's why I built GrantorAI. Because I needed it, as a scientist myself.
What I actually wanted to build
The idea is simple: you focus on the science; the tool handles the form.
You upload two documents: the call for proposals, and the submission form (the empty template you actually need to fill in). GrantorAI extracts all the form elements: what each section should contain, the word limits, the funder's evaluation criteria. Then you add your raw ideas. Notes from meetings, emails with collaborators, bullet points, things written in multiple languages with typos intact. This is the ideas stage, not the writing stage.
From there, for each section of the form, you can generate a first draft that aligns your ideas with the specific requirements of that grant. If you've put in substance, it gives you something usable: not a finished proposal, but a starting point. You're no longer staring at a blank page.
The feedback function has been the most surprising part. You can ask for a section-by-section evaluation against the funder's stated criteria. In my own testing, it consistently surfaces gaps I'd missed: assumptions I hadn't made explicit, arguments I'd buried that deserved more space. That kind of structured critique is hard to get from colleagues at short notice.
The goal is less writing, not more
I want to be clear about what GrantorAI is not for.
It's not a way to submit more grant applications. If anything, I'd rather it helped researchers submit fewer, stronger ones: spending their time on the ideas and the science, not on reformatting the same proposal for three different funders.
The system already pushes researchers to write too many proposals. Adding a faster tool to the current setup without changing the incentives just produces more proposals, not better science. For now, this may be your hardest job when using AI tools: to write fewer, stronger proposals, not more.
If GrantorAI can help a non-native English speaker compete on the same terms as a senior PI at a well-resourced institution, I'll count that as a success. If it lets a PhD student get useful feedback on their first fellowship application at midnight before the deadline, that's also a success. Not because more grants will be funded, but because the playing field will be a little more level.
That's what I built it for.