The Cheapskate Guide to Applying for Grants Using AI
By Sarah Pita
What follows is a complete guide to using AI tools to apply for grants, from start to finish. The approach described is for an individual practitioner, using only basic tools costing, in total, about $40/month. I use Claude Pro, ChatGPT Plus, and the free tiers of Copilot, Gemini, and Perplexity. I generally work in Word and Excel (rather than the Google suite). I work through all the phases:
1. Create a Project
2. Assign an AI “project manager”
3. Finding aligned funders and open RFPs
4. Reviewing RFPs
5. Project planning and coordination
6. Drafting narratives
7. Application submission
Note that for those with enterprise-level subscriptions to tools like Copilot that work within your email, calendar, and files, a whole bunch of additional efficiencies become available. I don’t cover those here because this is the cheapskate guide, but, to put it simply, personal tools should not have access to your work systems for security reasons.
Create a Project
Create a project in Claude or ChatGPT. You may call it “[My organization] Grant Writer” or even “[My project] Grant Writer.” Describe its function and provide it with some current knowledge about your organization, such as the annual report, program one-pagers, or even a recent grant application. This is its basic knowledge about your organization, which it will reference every time you work in the project.
Tip: Keep it up to date, or you’ll find yourself repeatedly correcting last year’s numbers.
Assign an AI Project Manager
A grant application is a complex project. Before you get started, assign a project manager.
This is a job for your paid AI tools. Whatever you use, you want it to keep your work, and you don’t want to run out of tokens halfway through.
Within your Grant Writer project, start a new chat. As a first step, explain exactly what you’re trying to do, providing all the context it could possibly need. This is the only time you’re going to explain the project, so do a good job.
From now on, you will be asking Claude to write prompts for you that you will be pasting into other chats and other AI models, so you want it to have all the information.
Tip: If you are using Claude as your project manager, there are several reasons it’s a good idea to use a different tool for research. It’s not because Claude is bad at research. Rather, outsourcing this task to one of your other tools introduces some good redundancy, leveraging the capabilities of different models. It also conserves tokens in your primary application, so that you are less likely to run out halfway through the project.
Tip: If you get a notification that you have used 90% of the available memory in a project (this is usually a Claude situation), your next prompt should be a request for a handoff prompt. You can hand off to a fresh chat, or even to another tool.
Finding Aligned Funders and Open RFPs
There are great subscription tools out there to help nonprofits find potential grant funders. We’re going to assume you do not have access to them.
That doesn’t mean you can’t find new grant opportunities, though.
Use AI research tools to identify other nonprofits doing similar work: serving the same or similar populations, working in the same geographical area, of similar size. Then, take a look at their funders in public-facing documents like annual reports.
I ask my project manager to draft a prompt, or perhaps several prompts that I take to one of the other tools I use—ChatGPT web or deep research, Gemini, or Perplexity. For a recent search, I had my Claude project manager draft four prompts for different angles I was researching, and had four ChatGPT windows open at once running the searches.
The result was four long lists with redundancies and errors. But instead of going through them myself at this stage, I put them all into a big, messy list and took them to Cowork, using another prompt drafted by my project manager to explain what I needed: duplicates removed, funder websites checked for compatibility, list of potential funders created and sorted by fit (go get a snack while it works).
That was when I went through the list and checked it against the funders’ websites. Reader, I found errors. I found cases where what the AI interpreted as a “peer organization” or a “good fit” was not convincing. I found clear cases where reaching out for a meeting was the next step, and things that precluded the nonprofit from applying that the AI missed.
But I was editing a list, and I found some good leads to work on—at least as many as I have found through a database search at the public library. And that’s a great place to start.
Reviewing RFPs
If you deal with government grants a fair bit, you can use AI to take a first pass at reading the RFPs—not to replace you, but to help you decide if you need to read all 85 pages yourself. I have explained a couple of ways to do thishere.
You can upload the RFP to your Grant Writer project and just ask questions about it, like “when is this due?” and “is this a good fit for my organization?”
You can also create a GPT in ChatGPT. It will have information about your organization in its memory, and comes preloaded with your instructions about what you would like in your synopsis. Set it up right, and all you have to do is upload it and prompt “please review.” Ask for page numbers to verify key pieces of information.
Tip: If you have Cowork you can put all your RFPs in a folder, and have Cowork review them all at once and make an assessment—perhaps even a ranked list.
Project Planning
You already knew that grants aren’t just about the narratives, right? Once you find a promising opportunity, you will need to coordinate with other people to put together the application. Use your project manager tool to come up with a list of things you need, who you’ll get them from, and how long they are likely to take.
This is where the shoestring approach holds you back a bit. If you have Copilot, Gemini for Google Workspace, or Enterprise Claude Cowork, you can automate some of these tasks. On a shoestring, you get a checklist to work from.
Tip: Be sure you’re familiar with the application yourself before you start this step. AI has been known to miss things. So have humans. This way, there is redundancy!
Once you know what you need, you’ll probably be sending some emails to people, with requests and instructions, including the all-important message to finance asking for the grant budget. I will typically let my project manager draft the instructions, but I’ll check it and write some human language around it. These are human beings we’re writing to, after all, and it can be depressing to get a message that was clearly, entirely drafted by AI. So let the AI make sure the request is crystal clear, and you get to be the human and help the other person feel good about the request.
Tip: If you’re generally neurotic about your email, you can have AI read your drafts before you send them to make sure you’re saying what you think you’re saying. Just tell it NOT TO REWRITE. (I have an articlehere about how to adjust your settings so that it never rewrites by default.)
Drafting Narratives
I wrote an article about drafting grant narratives a few months ago;you can find it here.
One thing that didn’t make it into that article: how to research and create a good, credible statement of need. Statements of need typically require information that originates outside of your organization, describing the problem your project is trying to solve and the context it exists in. If you don’t have all the information you need on hand, you can use the research capabilities of tools like ChatGPT, Gemini, or Perplexity to round it up.
To save yourself some time, have your project manager write the prompt you’re going to take to one of the other tools. By now, it knows about your project because you have already explained it, and it knows about your organization because it has good, useful information in its memory already. It has the exact wording of the Statement of Need question. So just ask it to write the prompt you’ll need to get the answers for the questions. Tell it that you need credible, easily verifiable sources.
While the research is running (okay to run it in two different tools if you’re feeling frisky) go get yourself another cup of coffee, or maybe go outside for a few minutes. You’ll want to be fresh for the next step.
Bring your report or reports back to your project manager chat. Ask it which facts you need to independently verify, and ask for the actual links. And then go through the list yourself. This is really important, because there will always be at least one mistake and you don’t want your application, which will be read by people who are experts on this topic, to be based on hallucinated information. You’ll want to lay eyes on each number and confirm that it is what the AI says it is. You’ll also want to confirm that, if articles are cited, they’re actually about what the AI says it is.
When you’re done with that, go back to your project manager and let it know what you’ve found, and instruct it to draft your statement of need.
And move on to the next question.
Submitting Your Application
In my opinion, this is the most stressful, tedious part of the whole process. And, unfortunately, if you are operating on a shoestring, you have to do this yourself. You can go back to the checklist you created in section 5 if you’d like.
Virtually no grants require a paper submission any more; it’s all online portals or occasionally a PDF of the entire application, emailed as a file attachment. Funders typically want an array of organizational information—budgets and financial statements, your 501c3 determination letter, perhaps your 990 or most recent audit, perhaps some signed disclosures.
If you’re operating on a shoestring, and don’t have Copilot or Claude Cowork or Gemini working within your systems and contained in your organization’s cloud, there isn’t a great way to automate this. You’ll just go through and paste in your narratives, and upload your documents, and when you’re sure you’re done, you’ll hit submit.
As you’re checking your work and clicking on attachments to make sure you got the right ones, you’ll probably think about how nice it would be to be able to reliably automate this step. But then you’ll hit submit, wait for the acknowledgement email, and go for a walk outside or go talk to one of the people in your life for a while. Because while AI-assisted processes can make the work faster, make sure you use some of that time you just got back for something that feeds you, and that AI can’t do.