How to Use AI to Find Grant Funding for Your Nonprofit

Grant prospecting is one of the most tedious parts of fundraising. You need to find foundations whose giving aligns with your mission, verify they fund organizations like yours, figure out how much they typically give, and then repeat that process hundreds of times to build a real pipeline.

AI is starting to change this. Not in the “magic button that writes your grant application” way that some vendors are selling, but in a genuinely useful way: AI can search, filter, and pattern-match across millions of grant records in seconds, surfacing prospects that would take a human researcher weeks to find.

But there’s a catch. AI is only as good as the data you give it. And most nonprofit grant data is locked behind expensive subscriptions or buried in unstructured PDF filings. Here’s how to actually make this work.

What AI needs to be useful for grant prospecting

When people say “use AI for grants,” they usually mean one of two things:

1. AI-assisted research — using ChatGPT, Claude, or similar tools to help you find and evaluate potential funders
2. AI-powered search — tools that use AI to match your organization against a database of grants

Both approaches need the same thing: structured, searchable grant data. Without it, you’re asking an AI to guess based on whatever it learned during training, which is often outdated, incomplete, or just wrong.

The best source of structured grant data in the U.S. is IRS 990 filings. Every public charity that awards $5,000+ in grants reports the recipients, amounts, and purposes on Schedule I of their Form 990. Every private foundation reports all grants on Part XV of their 990-PF. Between these two sources, you get a remarkably complete picture of philanthropic grantmaking.

The challenge has always been access. These filings are public record, but until recently, making them searchable required either an enterprise subscription to Candid (starting at $3,499/year) or a lot of manual work.

A practical workflow: AI + grant data

Here’s a workflow that actually works today, using free or affordable tools:

Step 1: Find organizations similar to yours that are receiving grants

Start with organizations whose mission overlaps with yours. Search for them in a grant database like 501(see) and look at who’s funding them. If a foundation gave $500K to an organization doing similar work in a neighboring state, they’re a real prospect for you.

This is where AI can help. Instead of manually reading through grant lists, you can describe your organization’s mission to an AI tool and ask it to evaluate whether each funder in your results is a good fit.

Step 2: Look at funder patterns, not just individual grants

A single grant is a data point. A pattern is a strategy. When you find a promising funder, dig deeper:

• How much do they typically give? If their median grant is $10,000 and you need $500,000, it’s probably not a fit.
• Do they fund in your geography? Many foundations have geographic restrictions.
• What’s their NTEE focus? A foundation coded as T20 (Philanthropy/Voluntarism) gives very differently than one coded as E20 (Healthcare).
• Are they increasing or decreasing their giving? Compare across tax years.

All of this data is in 990 filings. Tools like 501(see) let you filter by amount, state, NTEE category, and tax year to surface these patterns without manual digging.

Step 3: Use AI to draft your approach

Once you have a shortlist of 10-20 qualified prospects, AI becomes genuinely useful for the writing part:

• Summarize the funder’s giving history and priorities
• Draft an initial letter of inquiry tailored to their focus areas
• Identify connections between your programs and their stated purposes

The key is that you’re feeding the AI real data about the funder, not asking it to make things up. “This foundation gave $2.3M across 47 grants last year, primarily to education nonprofits in the Southeast” is a much better prompt than “find me foundations that might fund education programs.”

What to watch out for

AI hallucinations are a real risk in grant prospecting. If you ask ChatGPT “what foundations fund youth mentoring programs in Georgia,” it will happily generate a list. Some of those foundations will be real. Some will be invented. And you won’t know which is which unless you verify against actual data.

This is why starting with structured data matters. When your AI analysis is grounded in real 990 filing data, you can trust the foundation names, amounts, and dates. The AI adds value by finding patterns and drafting language, not by making up facts.

Don’t over-automate the relationship. Grant funding is fundamentally a relationship business. AI can help you find the right doors to knock on and prepare you for the conversation, but the conversation itself still needs to be human. Program officers can spot a mass-produced AI letter from a mile away.

Verify before you apply. 990 data runs about 12-18 months behind (foundations file after their fiscal year ends, and processing takes time). A foundation’s most recent 990 tells you what they funded two years ago, not what they’re funding right now. Use it as a starting point, then check the foundation’s website for current priorities and open RFPs.

Tools that make this easier

For the data layer:

• 501(see) — search 21M+ foundation and federal grants by funder, recipient, amount, state, year, and source. Free tier available.
• IRS Tax Exempt Organization Search — free but limited to basic organization info, no grant-level data.
• Candid/Foundation Directory Online — comprehensive but starts at $3,499/year.

For the AI layer:

• Claude or ChatGPT — feed them grant data exports and ask for analysis, pattern matching, or draft language.
• AI tools with MCP (Model Context Protocol) support can connect directly to data APIs, which means the AI can search grant databases in real-time instead of relying on whatever’s in its training data.

The combination matters more than either piece alone. A grant database without AI means you’re still manually scanning results. AI without real data means you’re working with hallucinations. The sweet spot is structured grant data that AI can actually reason over.

Getting started

If you’re a grant writer or development director and this is new territory for you, start simple:

1. Search for 3-5 organizations similar to yours in 501(see)
2. Look at who’s funding them and how much
3. Copy the top 10 funders into a conversation with Claude or ChatGPT
4. Ask the AI to evaluate each one against your organization’s mission and needs

You’ll have a qualified prospect list in under an hour instead of under a week. That’s not magic. It’s just better tooling applied to public data that’s always been there.