AI for grant writing
Where it helps across the arc of a proposal, and the two places to keep it out.
Grant writing is the first place most nonprofits try AI, and it's a reasonable instinct. The work is writing-heavy, deadline-driven, and repetitive in ways that look automatable from the outside.
It is partly automatable, but the parts that are and the parts that aren't don't split the way people expect. The narrative — the part that feels like the hard writing — is where AI helps least, because that's where your specific programs and your specific evidence live.
Where it helps most
Reading the RFP against your program.
Hand over the funder's guidelines and your program description, and ask which eligibility requirements you don't meet. This takes about two minutes and catches the disqualifier that would otherwise surface on submission day. It's the single highest-value use in the whole process.
First drafts of boilerplate.
Organizational history, governance structure, DEI statements, capacity narratives. You've written these fifty times, the funder needs them phrased their way, and reshaping existing text is exactly what these tools do well.
Reformatting between applications.
You have a 2,000-word narrative and the next funder wants 800 words in three sections with different headers. That's translation work, and it's fast.
Budget narratives from budget numbers.
Give it the spreadsheet figures and the program description, and it will write the prose that explains why line 14 is what it is. Check every number that comes back.
Compliance review before you submit.
Ask it to read your draft against the guidelines and list what's missing. It catches attachment requirements and word limits that human eyes skim past on the fourth read.
Where to keep it out
Your outcomes data.
These tools will produce plausible-looking numbers if you leave a gap, and a fabricated figure in a funded proposal is a serious problem. Every number in a grant should come from your records and be checked by a person who knows what it means.
The story of a specific participant.
Beyond the data problem, there's a consent question. A composite or invented client story in a proposal misrepresents your work even when the underlying pattern is real.
The workflow that works
Start with the RFP analysis before you write anything. Draft the boilerplate sections next, since they're the fastest wins and they warm up the context for everything after. Write the narrative yourself, or draft it yourself and ask for a critique rather than a rewrite. Finish with a compliance pass against the guidelines.
The order matters because context accumulates. By the time you're doing the compliance check, the tool knows the funder, your program, and your draft.
Add a real example here — a specific grant where this workflow saved measurable time, ideally with the funder type and rough hours.
The Proposer is the coach scoped to this work. It knows what an LOI is, how a logic model connects to an outcomes section, and what funders mean when they ask about capacity — so the first thing you type can be the actual question rather than a definition.