How to use AI for grant writing

Where AI struggles by default

Before using AI, it’s important to take note of where it struggles most:

• It doesn’t know context. By default, it won’t know your organization’s history with a funder, what you’re applying for, your pitch, how you align with the funder’s mission, key grant outcomes, and more.
• It makes things up (“hallucinates”). It can sometimes invent facts, statistics, and even sources that sound completely real.
• It doesn’t know your organization. By default, it only knows the content posted on your website. And it will not understand how to frame this content for a grant.
• It reflects bias. The most common or popular views in its training data tend to dominate.
• Its information can be stale. It may not know recent funders, events, research, or rule changes.
• It lacks real strategy and emotional depth. It can mimic a tone, but it can’t set your vision or feel your mission.

How do you prevent these common AI pitfalls? Through Projects, CustomGPTs, and Gems.

Projects, CustomGPTs & Gems

Same idea, three different names:

• Claude → Projects
• ChatGPT → CustomGPTs
• Gemini → Gems

The setup is nearly identical with each. For the purposes of this guide, we’ll refer to them all as “Projects.”

Why should you use a project? A plain “new chat” forgets your context the moment you close it, so you re-explain your mission, programs, and voice every time. Plus, it fills the gaps with generic guesses from the open internet. A Project fixes both problems at once. You set it up once, and from then on every conversation starts with the AI already knowing who you are, grounded in your real information.

The result: more accurate, more on-brand, more useful drafts, with far less repetition. It’s also your single best defense against hallucination, bias, and stale information.

How many projects should I build for grant writing?

Create separate Projects per program model your nonprofit operates. If your organization has one program model (e.g. after school youth center), create just one Project focused on that. If your organization has multiple program models (e.g. housing, youth services, behavioral health services), create separate Project files for each.

The reason: you want each Project to be as focused as possible in order to help AI pull content relevant towards the work you are writing to. If you load everything you do into one Project and then ask the Project a question, it increases the chance that it will write responses that include irrelevant information.

How to build a Project

Find “Create new Project/CustomGPT/Gem” in your AI platform. When you create a new project, it will ask you for the following content:

1. Instructions
2. Knowledge Files

Step 1 — Write the instructions

Instructions are used to deliver the results you want in AI responses. Think of them as the instructions or training you would provide a new grant writer joining your team. Strong instructions should include the following details:

• Role — who the AI is: “You are a grant-writing expert for [org], a nonprofit focused on ___.”
• Audience — who it’s writing for: “Write for foundation program officers.”
• Voice & tone — “Warm, direct, and hopeful. No jargon like ‘synergy’ or ‘leverage.’”
• Boundaries — “Never invent statistics, outcomes, or sources. If you’re unsure, flag it. Only help with grant-related tasks.”
• Format defaults — “Keep drafts concise unless I ask for more, and show me what you changed.”
• Preferences — “Don’t use em dashes,” “Use ‘formerly incarcerated’ instead of ‘convict’” when referring to our population, and so on.

Step 2 — Load the knowledge files

As the name suggests, knowledge files are the library of content your AI Project will use to generate your written content. A curated, specific, and comprehensive library will result in stronger AI generated content. The best methodology I’ve found for uploading knowledge files is to upload two types of knowledge files: Internal content, focused on your organization’s information, and External content, focused on outside material such as research, assessments, and so on.

1. Internal content. Name the internal files: “[Internal] Organization Info.”

2. External content. Name the external files: “[External] LGBTQ Youth Research.”

What kind of files should I upload?

Upload PDFs, Word, or text files. Label each clearly as [Internal] or [External], keep files focused (break up giant documents), and keep them current (remove old versions). Treat your knowledge files as a living library and update them as your organization evolves.

The best file for AI is a Markdown file (.md). Second is a .txt file. They are best as they are easiest for the system to read through quickly and thoroughly. Therefore, it increases the likelihood that AI will return better content. AI can help convert and format any of your existing knowledge files into Markdown. Just ask it to do so for you.

Now that you have a completed Project with set instructions and knowledge files, it is time to start writing your grant. Click into your Project, and you should see a chat window to begin.

Context is key in your prompting

Even with a great Project, it’s important to provide context at the start of your chat. The single biggest predictor of a good draft is how much context you provide before asking for a word of narrative. Remember, context is key, and it’s the one thing AI will never have by default.

The grant context worksheet. To help provide context, fill out this grant context worksheet and upload it at the start of every new project chat, before writing your grant. Feel free to use bullet points. The content within this worksheet does not need to be well written prose. In fact, brevity is most helpful and it focuses on the point rather than unnecessary prose.

• Funder name and the specific grant or program
• Funder mission and giving priorities (in their own words — paste from their site or guidelines)
• What they fund and don’t fund; typical award size; deadline
• The project or program you’re seeking funding for
• Amount you’re requesting and how it will be used
• Why you’re a strong fit — connect their priorities directly to your work
• How to frame the language for this funder — what to emphasize, and what to downplay or avoid
• What language to utilize and how to style the content (inspiring, warm, research-focused, etc.)
• Relationship history — new to them? Past grantee? Warm intro? Prior site visit or call?

Go through the proposal one question at a time

Upload the worksheet to the Project chat with this prompt:

The AI Project will respond in the affirmative and confirm that it will help you write one grant response at a time.

This is how I typically work through a chat:

AI is excellent at writing and unreliable at knowing. It predicts what an answer should look like, not what’s actually true. Your job is to be the fact-checker and editor it can never be.

Truth check — before anything AI-assisted goes to a funder:

• Every statistic and fact is verified against a real, current source
• Every citation or source actually exists and says what’s claimed
• Nothing is out of date — funder info, deadlines, regulations, program data
• The content represents your community accurately and without bias
• No private or sensitive information slipped in
• It sounds like your organization, not a generic template
• Every promised outcome and impact figure came from a human

With AI becoming more and more popular, funders have begun asking how AI was used in the development of your proposal. From my experience, this is a rare question at the moment, but I can foresee it becoming increasingly popular.

How was AI used in the development of your proposal?

My response: AI was utilized to assist with drafting and refining my grant narrative. Every response was crafted and honed with human input, strategy, and revision.

I try to keep it simple, but if a lengthier, more detailed response is needed, the response could be:

(Insert AI platform here) was utilized to draft and refine grant responses. The AI platform was provided with the context and knowledge files necessary to craft a relevant and thought-out narrative. All AI output was reviewed by a human and refined, revised, and edited. The AI platform was also guided by human strategy in order to ensure that the responses adhered to and sufficiently addressed each question. All content was reviewed by a human before proposal submission.

By utilizing the skills outlined above, you’ll be able to craft stronger grant proposals in a more efficient manner. That efficiency will then enable you to focus on relationship building, which is the most important long-term strategy.

Every grant proposal can be stored within Grantary and Grantary can help hold the context and outreach that occurs through your relationship building efforts. There’s no AI writer inside Grantary on purpose. Check out my other resource: which AI your nonprofit should use.

Find out more about Grantary. See how it works →

• How to implement AI for your nonprofit — implementation and considerations for your nonprofit’s AI journey.
• Which AI should your nonprofit use? — platforms, nonprofit pricing, and whether to buy AI bundled into other software.
• How to use AI for grant prospecting — the same setup, pointed at funder research.

Will funders know I used AI to write my proposal?

Not if you use it well. If you employ the tactics I advise against, then yes — most likely a funder will be able to tell that you used AI to write your proposal. If you’re: (a) setting up a Project by uploading instructions and knowledge files and (b) providing your chat with context before you begin writing, you’ll craft a much more responsive and better proposal using AI. As always it’s important to refine and add your human touch to every narrative. It’s also incredibly important that you’re conveying the true human impact of your work into your narrative and the expertise that comes along with that.

Is it safe to put our organization’s information into AI?

My rule of thumb is to never put any content into AI that I wouldn’t put onto my website. If the content exists on your website, it’s safe to put it into AI as, more than likely, AI has already scanned your website and pulled your content.

It’s very important to protect the data of the individuals you serve. I would not put real names of your participants into AI unless those names are already published on your website. I also would not put any PII into AI tied to any of your employees or your participants.

Within the settings of whatever AI platform you have, you can also inform it not to train the model on the content you’re entering into the system. If you’re on an unpaid plan, this may not be possible, so that’s why it’s better to be on a paid plan. On a paid plan you have more control over your data and can turn that setting off.

Which AI platform is best for grant writing?

Check out my Which AI should your nonprofit use? resource for a detailed response to this question. Overall, all AI platforms are capable of assisting you with grant writing.

Can AI just write the whole proposal?

It could, but from experience it is better to go question by question. As described above, the reason to do so is that you have much more control over each response as you go. It also gives you the opportunity to hone each question’s response as you go. If you use AI to generate the entire proposal at once, you increase the likelihood that you will get generic and vague responses.

Do I need a paid account to do this?

I have personally only used paid accounts to do this work and I recommend that you utilize a paid account to do so as well as you need a paid account in order to use the Project feature. In addition, by using a paid plan, you can adjust your privacy settings to prevent AI from being trained on your data.