Skip to main content
Guide9 min read·Updated July 24, 2026
🤖

How to Write Grant Proposals With AI Tools (2026)

B

A. Frans

Published July 24, 2026

Grant WritingNonprofitAI Writing ToolsFundraisingProductivity

A development coordinator I know keeps a whiteboard with six deadlines on it, and by the end of a quarter three of them have red circles because she ran out of hours. That's the actual job for most people writing grants at small nonprofits. The federal RFP lands at 47 pages of eligibility rules, formatting specs, and scoring rubrics. The foundation application wants a need statement, a logic model, three years of budget, and a sustainability plan, all by Friday. And the payoff on a cold foundation proposal is thin: success rates are often cited somewhere around 10 to 20 percent, which means most of the writing you do goes nowhere.

You can't fix the hit rate by writing faster. You fix it by spending your limited hours on the parts that actually move a reviewer, and handing the mechanical parts to a machine. AI tools are good at the mechanical parts. They're bad at the parts that win money. Knowing which is which is the whole skill.

Here's how the work maps to tools before we get into the steps.

Grant taskBest AI toolWhy
Finding funders whose priorities match your programPerplexity AICites live sources so you can verify a funder's actual giving history, not a guess
Turning a 40-page RFP into a checklistClaudeHandles long documents and pulls structured requirements without losing the small print
Drafting narrative sections from your notesChatGPT or ClaudeFast first drafts you rewrite; good at reshaping bullet points into prose
Budget math and budget narrative wordingGoogle GeminiHandles the arithmetic and explains each line in plain language
Grammar, tone, and readability polishGrammarlyCatches the errors you stop seeing after the fifth read
Storing and reusing boilerplate across proposalsNotion AIKeeps your org description, past outcomes, and stock language searchable
One honest caveat before you start. None of these tools writes the grant for you. They don't know your executive director quit last month, they don't know the program officer prefers concrete numbers over mission language, and they can't invent your outcomes data. The funder relationship and the specifics of your work come from you. The tool drafts, researches, and checks. You supply the truth and the judgment.

Step 1: Research funders and fit

Applying to a funder whose priorities don't match your program is the most common way grant writers waste a week. A foundation that funds early-childhood literacy in three counties will not fund your after-school robotics club two states over, no matter how good the proposal reads.

Start with Perplexity or Google Gemini to build a shortlist. Ask for funders active in your issue area and geography, then verify every name against the funder's own site and their most recent 990 filing. The AI will occasionally surface a foundation that stopped giving in 2021 or confuse two organizations with similar names, so treat its list as leads to check, not facts to trust.

A prompt that works:

> "I run a nonprofit that provides free after-school tutoring to middle schoolers in Oakland, California. Annual budget is about $300,000. List 15 foundations that have funded similar education programs in the San Francisco Bay Area in the last three years. For each, give the typical grant size and their stated funding priorities, and include a source link I can verify."

Take the output and do the human part. Read the funder's recently funded grantees. If your program looks like something already in their portfolio, you have a fit. If it doesn't, cross them off before you write a word. For deeper background on tools built for this kind of source-checking, see AI tools for researchers.

Step 2: Decode the RFP into a checklist

Government RFPs and detailed foundation applications hide the disqualifiers in the fine print. Wrong margins, a missing attachment, a budget category the funder doesn't allow, a page count you blew past. Reviewers toss non-compliant applications before scoring them, so this step protects the hours you're about to spend.

Paste the full RFP into Claude, which handles long documents better than most, and ask it to extract every requirement into a checklist. You want three buckets: eligibility rules you must meet to apply at all, submission mechanics like format and deadlines, and narrative questions the proposal has to answer with their point values.

> "Below is a federal grant RFP. Extract every requirement into a checklist grouped into three sections: (1) eligibility criteria, (2) formatting and submission rules including page limits, font, margins, and attachments, (3) each narrative question with its scoring weight. Flag anything that could disqualify an application if missed. [paste RFP]"

Read the checklist against the source once, because the AI can drop a requirement or misread a point allocation. Then build your outline directly from the scoring rubric. If the reviewer gives 30 points to the need statement and 10 to organizational capacity, your effort follows those numbers. This is the single highest-value thing AI does in grant work, and it takes ten minutes instead of an afternoon of highlighting.

Step 3: Draft the narrative sections

The narrative is where writers freeze, and it's where AI earns its place, as long as you feed it real material. A blank prompt gives you generic mush. Your bullet points, your data, and your program details give you a draft worth editing.

Take the need statement. You already know the facts: the problem, who it affects, the local numbers, the gap your program fills. Dump them as rough notes and let ChatGPT or Claude shape them into prose.

> "Turn these bullet points into a 250-word need statement for a foundation grant. Keep every statistic exactly as written, cite them inline as I've noted, and write in a plain, direct tone without buzzwords. Notes: 62% of students at our three partner schools read below grade level (district data 2025); only one school has a full-time reading specialist; families in the neighborhood have a median income of $34,000 so private tutoring isn't an option; our program served 140 students last year with a 71% improvement in reading scores."

The draft comes back structured and readable. Now you do the rewriting. Cut the sentences that sound like every other grant. Add the detail only you know, the story of the one student, the specific reason your method works. Do the same for goals and objectives, keeping them measurable, and for the methods section, describing what you actually do week to week. Never let the AI invent a statistic or a program feature. If a number appears in the draft that you didn't supply, delete it and check. For more on drafting tools and how they differ, see AI tools for writers.

Step 4: Build the budget and budget narrative

The budget is math plus explanation, and both parts trip people up. Google Gemini handles the arithmetic reliably and writes clean line-item justifications, which is the part reviewers actually read to judge whether you know how to run a program.

Give it your cost assumptions and ask for a full budget table plus a narrative that justifies each line. Personnel with fringe rates, direct program costs, indirect costs at your negotiated rate. The tool catches the arithmetic slips that creep in when you build a budget in a spreadsheet at 11pm.

> "Build a one-year project budget for a tutoring program. Personnel: one program manager at $52,000 (0.5 FTE on this grant), two part-time tutors at $22/hour for 15 hours/week over 36 weeks each. Fringe is 24% of salaries. Supplies $4,000. Indirect at 10% of direct costs. Produce a line-item table with subtotals, then write a two-sentence justification for each line in plain language."

Check every figure against your own numbers. AI arithmetic is usually right but not always, and a budget that doesn't add up tells a reviewer you're careless with money. Confirm the total matches what you're requesting and that no category violates the funder's rules from your Step 2 checklist.

Step 5: Tighten, tailor, and proofread

A first draft assembled from AI sections reads competent and forgettable. This step makes it specific and clean, which is what separates a funded proposal from a scored-but-rejected one.

First, tailor. Go back to the funder research and rewrite your opening and framing to speak to what this particular funder cares about. If their materials emphasize equity, your language reflects that honestly, assuming your work actually does that. Generic proposals lose to specific ones every time, and specificity is the one thing AI can't manufacture because it doesn't know your relationship with the program officer or the site visit you hosted last spring.

Then tighten. Read for the phrases that sound like filler and cut them. Run the full draft through Grammarly for the mechanical pass, typos, tense slips, the comma errors you've read past a dozen times. Ask ChatGPT to flag any sentence that's vague or unsupported, then fix those yourself rather than letting it rewrite blind.

Read the whole thing aloud once. If a sentence sounds like it came from a template, it did, and a reviewer reading their fortieth proposal will feel the same thing.

Step 6: Track and reuse for the next one

Most of what you write gets reused. Your org description, mission language, past outcomes, standard staff bios, and boilerplate answers show up in nearly every application. Storing them well turns the next proposal from a blank page into an assembly job.

Notion AI works for this because it keeps everything searchable and lets you pull the right paragraph fast. Build a library: approved boilerplate, a log of which funders you applied to and the outcome, and the narrative sections that scored well so you can adapt them. When the next RFP lands, you start from your best previous language instead of from nothing. Over a year this is the difference that lets one person handle six deadlines instead of drowning in three.

FAQ

Can AI write a whole grant proposal? No, and you shouldn't want it to. AI drafts sections, extracts requirements, and checks math. It can't supply your outcomes data, your relationship with the funder, or the specific reasons your program works. A proposal written entirely by AI reads generic and loses to a specific one. Use it for the mechanical 60 percent and spend your saved hours on the 40 percent that wins.

Will funders reject AI-written grants? Funders reject weak, generic, non-compliant proposals. They don't run AI detectors and reject on that basis, and most wouldn't know either way. The risk isn't getting caught. The risk is that AI-drafted text is bland by default, and bland proposals score low. Rewrite everything in your own voice with your own specifics and the tool becomes an asset instead of a tell.

Is it safe to paste our org data into ChatGPT? Be careful with anything sensitive. Don't paste personal data about the people you serve, donor records, or unpublished financials into a consumer AI tool, since that content may be retained or used for training depending on the plan. Public information like your mission and published outcomes is fine. Check whether your tool offers a business tier that excludes your data from training, and if you handle protected data, confirm it meets your obligations before pasting anything.

Which tool is best for beginners? Start with ChatGPT or Claude for drafting and Perplexity for funder research. Those three cover most of the workflow, and the free tiers are enough to learn on. Add Grammarly for the proofreading pass once you're comfortable. Don't buy six subscriptions on day one. Learn one tool well before adding the next.

AI won't raise your success rate on its own. It gives you back the hours you were burning on formatting checklists, budget arithmetic, and first drafts, and those hours are the ones you were never spending on the parts that actually win money: the funder relationship, the specific story, the detail only you know. Treat these tools as a fast assistant who drafts and researches and never as an autopilot that submits. The writer who uses that saved time to tailor every proposal to the person reading it is the one who moves off a 15 percent hit rate. The one who lets the machine write the whole thing stays exactly where they started, just faster.

Share this article

📬

Get More AI Tool Guides

New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.