Published on 26.08.2025

Grant writing with AI: Seven smart practices

Generative AI tools like ChatGPT, Claude or Gemini are popular among researchers for drafting and polishing texts, including grant proposals. Used wisely, they can make it easier for you to submit a convincing proposal. But they are no magic wands either. Here are some practical tips to help you get the most out of these tools, without falling into common traps.

1. Check your funding agency regulation

Before anything, make sure you know the rules of the funding agency where you are submitting your proposal. For example, the SNSF and the European Commission both accept AI as a helpful tool but warn that you are still fully responsible for everything you submit—so no blindly trusting the model or copying entire sections without checking. These institutions also highlight the importance of originality and confidentiality.

2. Protect sensitive information

Regarding this last point: be careful what you feed into AI tools, especially public or cloud-based ones. Don't paste in unpublished data or confidential partner details. This is particularly important for industry-funded projects, where confidentiality matters a lot. If you need to feed confidential information to a model, make sure you use tools that have been approved as safe by your institution; at Unifr, this is Copilot Chat (which is not the same as M365 Copilot).

3. Learn how to prompt

If you have used generative AI tools before, you know that a good prompt makes all the difference. One helpful approach is given by the “Tiny Crab Riding Enormous Iguanas” mnemonic:

• Task: Be specific about what you want the AI to do (e.g., draft a project summary).
• Context: Give background info (like your research goals or target audience) to help the AI understand what you need.
• References: Provide documentation to help the AI accomplish its task (e.g. the official guidelines that your proposal should follow).
• Evaluate: Decide whether you are happy or not with the response.
• Iterate: Revise the generated text yourself or ask the AI to do it by giving it new instructions.

LLMs are making their mark across academia—sometimes in questionable ways. This screenshot shows papers hiding prompts designed to influence the outcome of AI-assisted reviews.

4. Don’t expect AI to do all the work for you

“Write me a proposal about climate change adaptation in alpine agriculture that will get me an interview for the next ERC Starting Grant call.” While it certainly would be nice to have AI write a full proposal for us in a single prompt, such a naive approach simply won’t work with the current state of the technology. All you will get is a generic, unfocused text that doesn’t stand a chance. AI can help you with specific parts of your proposal—drafting a clear project summary, suggesting a way to format your roadmap, coming up with descriptive headings for your sections—but it will not write your proposal for you.

5. Fact-check everything

By now, you are probably aware that AI is surprisingly good at sounding confident while being completely wrong. While this problem can be mitigated by providing the model with the necessary knowledge to respond (which you can do by providing the relevant documents alongside your prompt or by using something like the “Deep Search” feature), you still carefully check every claim, figure and reference the model generates.

6. Rewrite the output

In most cases, you will want to rewrite what the model generates. AI tools are often wordy and vague. For example, they often introduce conjunctions where a single term would be enough (e.g. “novel and innovative”). Revising your text in this way will make it more impactful and easier to read.

7. AI won’t save you time, it will spare you effort

This one is somewhat counterintuitive. For many people, the motivation for using gen AI tools is to save time. While this may hold for some applications, it does not really work if your goal is to write a grant proposal. Writing a good prompt, checking all the claims of the model, revising the response, integrating it with your current text, and repeating these steps until you are satisfied with your proposal will take time – perhaps even more time than if you had written the whole thing yourself. However, while AI won’t necessarily save you time, it can spare you effort. It is often much easier to get a good text by starting from a mediocre one and removing its issues than by writing the whole thing from scratch. As such, AI tools are a great way to get you past writer’s block.

In summary, generative AI can be a valuable ally in crafting strong grant proposals. Just remember: like a word processor or a search engine, it is only a tool to enhance your workflow, not something you can trust to replace your work.