By Austin Little
Grant writing is mostly unglamorous: reading guidelines closely, turning real work into clear sentences, and getting the numbers right. Free AI can genuinely help with the middle part. It can also quietly sink an application if you let it near the parts that have to be true.
AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events. We do not list specific funders' AI rules except where we read an official source, and even those can change — always read the current policy of the funder you are applying to.
Who this is for
This is for the person who writes grants without a grants department. The executive director of a small nonprofit who also does the books. The teacher applying for a classroom grant. The researcher at a small institution writing their first federal application. The artist applying for a fellowship. The volunteer who got handed "the grant thing" because they are "good with words."
You probably do not have a budget for AI subscriptions, and you do not need one. Everything on this page works with free tools: the free tiers of the big chat assistants, or a free local model running on your own computer.
What you do need is a clear line between the jobs AI is good at and the jobs that must stay human. In grant writing, that line is sharper than almost anywhere else, because a grant application is a document someone is going to rely on to give you money.
The one rule that covers most of it
Anything that must be true has to come from you, your records, or a source you checked. That includes your organization's history, the people you serve, your outcomes, your budget, your partners, your timeline, your citations, and every number.
AI can help you say true things more clearly. It cannot know them. When a model fills a gap with something plausible, that plausible thing goes into a document that a reviewer will treat as your claim. If it is wrong, it is your claim that is wrong.
Everything else on this page is a variation on that rule.
Funder AI policies: they vary, and you have to read them
Here is the honest situation: funders do not share one rule about AI. Some have published explicit policies. Many have said nothing. Some prohibit their reviewers from using AI. Some ask applicants to disclose use. Policies are new and many are being revised.
One example we did verify: NIH
As one concrete example of what a funder policy can look like, the U.S. National Institutes of Health published a notice, NOT-OD-25-132, "Supporting Fairness and Originality in NIH Research Applications," also published in the Federal Register in August 2025. Based on the official notice and NIH's own explainer, as read on October 1, 2026:
- NIH said it "will not consider applications that are either substantially developed by AI, or contain sections substantially developed by AI, to be original ideas of applicants."
- NIH said AI tools "may be appropriate to assist in application preparation for limited aspects or in specific circumstances," and warned that AI use "may result in plagiarism, fabricated citations, or other kinds of research misconduct."
- NIH said that if AI use is detected after an award, it may refer the matter to the Office of Research Integrity and take enforcement actions, which can include disallowing costs, withholding future awards, suspending, or terminating the grant.
- NIH's explainer also reminded readers that NIH peer reviewers are prohibited from using AI for their critiques.
- The policy was stated as effective for applications submitted to the September 25, 2025 receipt date and beyond.
That is one funder. It is a large public one, and its rule is fairly strict. Others may be looser, stricter, or silent. The lesson is not "everyone follows NIH." The lesson is: policies can carry real consequences, so find the one that applies to you.
How to find the policy that applies
- Read the funding opportunity itself. The call, the RFP, the guidelines PDF. Search it for "artificial intelligence," "AI," "generative," "ChatGPT," "originality," "disclosure," and "authorship."
- Read the funder's general application or grants policy page. Large funders often keep AI guidance there rather than in each call.
- Check the application form. Some forms include a question or checkbox about AI use.
- If you find nothing, ask. Email the program officer or the contact listed: "Does [funder] have guidance on applicants using AI tools to help prepare applications?" A short, honest question is normal and nobody will hold it against you.
- Save what you found. Copy the exact wording, the link, and the date into your project notes. If the policy changes before you submit, you will know what you relied on.
If there is no policy at all
Silence is not permission to let a model write your application. It just means you fall back on the universal expectations of any grant: the ideas are yours, the facts are true, the citations are real, and you can stand behind every word in a conversation with the program officer. Write in a way that would hold up if the funder published a strict policy tomorrow.
What free AI genuinely helps with
These are the jobs where a free chat tool or local model saves real time without putting your truthfulness at risk — as long as the funder's policy allows the kind of help you are using.
Understanding the guidelines
Paste the funding guidelines (or a section of them) and ask:
- "List every required section, attachment, and formatting rule in this document, with page references."
- "What eligibility requirements are stated here? Quote the exact sentence for each."
- "What questions should I ask the program officer based on what is unclear here?"
Then check the list against the original. Models skip things. But a first-pass checklist from the guidelines, verified by you, beats reading a forty-page PDF cold at eleven at night.
Building a compliance checklist
Turn the guidelines into a checklist file: every section, word or page limit, font rule, attachment, signature, and deadline. Ask the model to draft it; you verify each line against the source document. Missing a required attachment or going over a page limit can disqualify an application regardless of how good the writing is.
Outlining from your own material
Give the model your raw material — your notes, past reports, program descriptions you wrote — and ask it to organize them under the funder's required headings. It is rearranging your facts, not inventing new ones. Check that nothing was added.
Plain-language editing
Grant prose gets dense. Ask for help making a paragraph you wrote clearer:
- "Mark sentences over 30 words and suggest a split. Do not change meaning or add information."
- "Point out jargon a non-specialist reviewer might not know."
- "Which sentences in this paragraph are vague? List them; do not rewrite."
Asking for a list instead of a rewrite keeps the words yours.
Cutting to the limit
When you are 200 words over, a model can suggest cuts: "Suggest five places to cut from this section. Do not remove any numbers, names, or commitments." You choose which to take.
Reviewer's-eye questions
Ask the model to play a skeptical reviewer: "Based only on this section, what questions would a reviewer have? What claims lack support?" This is often the single most valuable use. It does not write anything for the application; it helps you find the holes.
Consistency checks
Paste your budget narrative and your project description: "List any place where the numbers, dates, staff roles, or activities in these two sections do not match." Mismatches between sections are a classic reason reviewers lose confidence. You then fix them from your actual records.
What gets you rejected (or worse)
Invented facts and numbers
A model asked to "write a needs statement about food insecurity in our county" may produce statistics that sound right and are not. It may produce a percentage, a year, and a source name that look real. Reviewers who know the field will notice. Reviewers who do not may still check.
Rule: every number in your application comes from a source you opened and read, recorded in your notes with the link and date. If you cannot find the source, the number does not go in.
Fabricated citations
This is the one NIH called out by name, and it is a well-known failure of language models generally: references that look correctly formatted but point to papers that do not exist, or exist but do not say what the sentence claims.
Rule: never put a citation into an application that you did not personally locate and read. Do not ask a model for "sources for this claim" and paste what it gives you. Use it, if at all, to suggest search terms — then search yourself in a library database, the funder's own resources, or the original publication.
Made-up outcomes and testimonials
"Last year, [some percentage] of participants reported improved confidence." If you did not measure it, you cannot say it. "As one participant told us…" followed by a quote nobody said is fabrication, full stop. Models will happily produce both if you ask for "a compelling impact section."
Rule: outcomes come from your data. Quotes come from real people who agreed to be quoted, in their words, with their permission.
Generic voice that reads like everyone else
Even when nothing is false, a fully AI-drafted narrative tends to sound like every other AI-drafted narrative: smooth, vague, full of "leverage," "holistic," and "transformative." Reviewers read stacks of applications. Specific, concrete, slightly rough human detail — a real program, a real place, a real problem you have seen — stands out. Generic polish does not.
Substantially AI-developed content where the funder forbids it
As the NIH example shows, at least one major funder has said it will not consider applications substantially developed by AI as the applicant's original ideas, with possible consequences after award. Other funders may have similar or different rules. If the funder you are applying to has such a rule, letting a model draft whole sections is not a shortcut. It is a risk to the application and potentially to your organization.
Pasting confidential material into a tool that trains on it
Grant drafts often include information about people you serve, unreleased research ideas, partner agreements, and financial details. Many free cloud tools may use your conversations to improve their models unless you opt out. Before you paste:
- Check the tool's training setting and turn it off if you can.
- Remove names and personal details about clients, patients, students, or participants.
- For truly sensitive material, use a free local model on your own computer, where the text does not leave your machine.
A safe workflow, start to finish
Here is a step-by-step workflow that uses free AI where it helps and keeps it out of where it hurts.
Step 1: Read the policy first
Before you open any AI tool, find and save the funder's AI policy (or confirm there is none and ask). Decide what kinds of help you will use. Write it in your notes: "Using AI for: checklist, plain-language edits, reviewer questions. Not using AI for: drafting narrative, citations, data."
Step 2: Gather your true material
Collect your facts: past reports, program data, budget, staff bios, letters from partners, verified sources. Put them in one folder. This is the only place facts come from.
Step 3: Make the checklist with AI, verify by hand
Have a model draft a requirements checklist from the guidelines. Check every line against the source.
Step 4: Write the first draft yourself
Write rough. It is fine if it is clumsy. The ideas, the structure of your argument, and the facts are yours. If you are stuck, ask a model for questions ("What would a reviewer want to know about our evaluation plan?") rather than for paragraphs.
Step 5: Use AI for editing, as lists
Paste one section at a time. Ask for lists of suggested changes, not rewrites. Accept or reject each one yourself.
Step 6: Run the reviewer and consistency checks
Ask for skeptical reviewer questions and cross-section mismatches. Fix them from your records.
Step 7: Fact and citation pass, no AI
Go through every number, date, name, and citation and match it to your source folder. This pass is human only. If something cannot be matched, remove or fix it.
Step 8: Disclose if required
If the funder asks about AI use, answer accurately. Your notes from Step 1 and a simple log of what you used make this easy. Plain, honest language works: for example, "AI tools were used to check formatting requirements and suggest plain-language edits; all content, data, and citations were written and verified by the applicant."
Step 9: Read it out loud
Before submitting, read the whole narrative out loud. Anything that sounds like a brochure, or that you would be embarrassed to explain to the program officer in person, gets rewritten.
Prompts you can copy
Save these in a text file and paste as needed. They are written to keep the model away from inventing.
Guidelines checklist
"Below is a funding opportunity document. Make a checklist of every required section, attachment, page or word limit, formatting rule, eligibility requirement, and deadline. For each item, quote the exact sentence it comes from. If something is unclear, list it under 'Questions for program officer.' Do not add requirements that are not in the text."
Plain-language edit
"Below is a paragraph from a grant application that I wrote. Do not rewrite it. List any sentences over 30 words, any jargon, and any vague phrases, with a suggested fix for each. Do not add facts, numbers, names, or claims."
Reviewer questions
"Act as a careful, skeptical grant reviewer. Based only on the section below, list the questions you would have and any claims that lack evidence. Do not suggest new facts; only point out what is missing or unclear."
Consistency check
"Compare the two sections below. List every place where numbers, dates, staff roles, activities, or timelines do not match between them. Quote both versions. Do not resolve them; just list."
Cut to the limit
"This section is [N] words and the limit is [M]. Suggest specific cuts totaling at least [N–M] words. Do not cut any numbers, names, outcomes, or commitments. Show each cut as original text and proposed replacement."
Free tool options
You do not need a paid plan for any of this.
- Free tiers of the big chat assistants — for example, Claude Free and ChatGPT Free — can handle checklists, list-style edits, and reviewer questions, within their published limits. Our separate Apiary guide compares the two free tiers for writing. Check each tool's training setting before pasting anything sensitive.
- A free local model through an app like Ollama or LM Studio runs on your own computer. It is slower and less capable on modest hardware, but the text stays on your machine, which matters for confidential grant material. Our guide on using AI without paying walks through setup.
- Your word processor's built-in spelling and grammar check still catches plenty, costs nothing extra if you already have it, and involves no AI policy questions at all.
Whatever you use, the honesty rules are the same.
Special cases
Foundation grants with a relationship
Smaller family and community foundations often fund relationships as much as documents. A program officer who knows you will notice if your application suddenly sounds like someone else. Keep your voice. Use AI for the checklist and the cuts, not the story.
Government grants
Government applications tend to have the most detailed requirements and, at least in the NIH example, published AI rules with consequences. Read the current policy for the specific agency and program.
Fellowships and individual artist grants
These usually ask about you: your work, your vision, your history. That writing is almost by definition yours alone. AI can help with formatting and length, and with reviewer-style questions, but the personal statement should be written by the person it describes.
Reports to funders after the award
Progress and final reports carry the same truth rules, sometimes with more at stake, because they describe what happened with the money. Every number comes from your records. AI can help you organize and clarify; it does not supply outcomes.
If you are a reviewer
Some people reading this review grants as well as write them. If you are reviewing, check the funder's rules about reviewers using AI before you touch any tool. NIH, for example, says its peer reviewers are prohibited from using AI for their critiques. Other funders may have their own rules, and application materials are often confidential. When in doubt, do not paste applications into any AI tool, and ask the funder.
Quick answers
Can I use ChatGPT or Claude for free to write a grant?
You can use free tiers to help with checklists, editing, and reviewer-style questions. Whether you can use them to draft content depends on the funder's policy — and some funders restrict it. Read the policy first.
Will funders know I used AI?
Some funders say they use detection technology; NIH, for example, said it would continue to employ the latest technology in detection of AI-generated content. Detection is not the real issue, though. The real issues are accuracy, originality, and following the rules you agreed to when you applied.
Do I have to disclose AI use?
Only the funder's policy can answer that. If it asks, answer honestly. If it does not say, consider asking the program officer.
Is it okay to have AI write the needs statement?
Not with numbers it supplies. A needs statement is built on data, and data must come from sources you checked. You can ask AI to help structure a needs statement around facts you provide.
What is the safest free option for confidential material?
A local model on your own computer, so the text never leaves your machine — combined with removing personal details you do not need.
The short version
Free AI is a good grant-writing assistant and a terrible grant writer. Let it read guidelines with you, build checklists, suggest plain-language edits, play skeptical reviewer, and spot inconsistencies. Keep it away from your facts, your numbers, your citations, your outcomes, your quotes, and your story.
Read the funder's AI policy before you start, because policies vary and at least one large funder has attached real consequences to breaking its rule. Save what you found. Disclose honestly when asked. And before you hit submit, read it aloud and ask yourself whether you could defend every line to the program officer in person. If you could, you are fine — whatever tools you used.