AI in Grant Writing: What Works, What Reviewers Notice, and What Funders Allow

Where does AI help in grant writing?

AI tools are good at work that’s mostly reading, sorting, and tightening. They’re weak at anything that depends on facts only you have. Here’s roughly where the line falls:

The biggest time saver is reading long funding notices. Federal notices often run long, and an AI tool can pull out the deadline, eligibility rules, page limits, and review criteria in a few minutes. Treat that summary as a starting point, though, and read the review criteria yourself. Our guide to reading a NOFO the way a reviewer will shows what to look for.

The same goes for compliance checklists. Ask the tool to list every required form, attachment, and formatting rule, then check its list against the notice line by line. A missing attachment can sink an application before anyone reads it, and AI tools do skip things.

AI is also handy for boilerplate and editing. Organization history, staff bios, and board lists don’t change much from one proposal to the next, and a tool can adapt your past text to a new funder’s word limit. It’s a decent editor for grammar, clarity, and length as long as you reread what it changed.

Where does AI hurt a grant proposal?

Needs statements are the clearest example. Left to itself, AI writes about poverty, health, or education in general, when reviewers want to know about the people in your service area, with local data and sources. A needs statement that could describe any city doesn’t make the case for yours.

Data and citations are the riskiest. AI tools can produce statistics and sources that look real and aren’t. NIH’s 2025 policy notice warns that AI use may result in plagiarism, fabricated citations, or other kinds of research misconduct. Don’t use a number or a citation you haven’t traced back to the original source yourself.

And there’s your voice. Funders back organizations they believe in, and a proposal should sound like the people who will run the program. When every section has the same smooth, even tone, reviewers have a harder time hearing who you are.

What do grant reviewers notice?

From our own work writing and editing proposals, these are the signs that a proposal was put together without enough care, whether by AI or by a rushed person:

• Language that never mentions the funder’s priorities or the notice’s criteria
• Big claims with no numbers, or numbers with no sources
• The same phrases repeated across sections
• Activities in the narrative that don’t appear in the budget
• Objectives that can’t be measured

None of these proves AI was used, but every one of them costs points. The fix is the same either way: tie each claim to the criteria, your own data, and your budget.

What do funders allow? Current AI policies

NIH has the strictest published rule, in notice NOT-OD-25-132, released July 17, 2025 and also published in the Federal Register (90 FR 37531). NIH says it will keep using technology to detect AI-generated content. If AI use is found after an award, NIH may refer the matter to the Office of Research Integrity and take enforcement action, such as disallowing costs, withholding future awards, or suspending or ending the grant. NIH’s FAQs put it plainly: if you use an AI tool to help write an application, you do so at your own risk.

The same FAQs explain how the six-application limit works:

• It counts by the calendar year of the Advisory Council that reviews the application. The January 2026 Council was the first under the policy.
• Only applications where you are the PD/PI or one of multiple PD/PIs count. Being a co-investigator, mentor, or subaward PI doesn’t count.
• Institutional training (T), Loan Repayment Program, R25 research education, R13 conference, and X01 applications not reviewed by Council are exempt. So are administrative supplements and non-competing continuations.
• An application counts once it passes administrative review, even if you withdraw it after peer review.

NIH also bars its peer reviewers from using generative AI to analyze applications or write critiques (NOT-OD-23-149). As of October 25, 2026, the NIH Guide lists NOT-OD-25-132 as active, with no notice replacing it.

NSF allows AI help but encourages disclosure and holds proposers responsible for what they submit. Its notice says generative AI can raise the risk of fabrication, falsification, or plagiarism.

Foundations were mostly undecided in Candid’s 2024 survey. If a funder’s guidelines say nothing, ask the program officer. These policies are still changing, so check each funder’s current guidelines before you apply.

What are the best AI tools for grant writing?

We don’t rank tools. They change too fast, and the right one depends on your work. What we do suggest is asking a few questions of any tool before your staff use it:

• What happens to your data? Find out whether the tool stores your text or uses it to train its models, and whether you can turn that off.
• Can it handle a long document? You’ll want to load a full funding notice, not paste it in pieces.
• Does it show where it got things? A tool that links to its sources is much easier to check.
• Does your organization have an AI policy? If not, write a short one before people start using these tools on grant work.

Don’t paste in confidential material, such as client records, staff salaries, or a funder’s confidential documents, unless your organization has approved the tool for that use. Wellcome tells applicants to keep confidential information and personal data out of AI tools.

Is grant writing going to be replaced by AI?

We don’t think so, at least not the parts that win grants. AI can speed up reading, formatting, and editing. It can’t build a relationship with a program officer, design a program that fits a community, or know which local numbers matter. NIH now says applications substantially developed by AI aren’t considered the applicant’s original ideas, and most foundations in Candid’s 2024 survey were still undecided about AI-written proposals.

What’s changing is how the time gets spent. Grant writers who hand the routine parts to AI have more time for strategy, data, and funder relationships. If you’re weighing outside help, see what a grant writer costs.

Safe AI workflow for grant writing: a checklist

1. Check the funder’s policy first. Read the notice and guidelines for any AI rules, and if there are none, ask.
2. Keep confidential data out of any tool your organization hasn’t approved, including client data, personal data, and confidential funder material.
3. Use AI to read and organize. Summaries, checklists, and outlines are low risk once you’ve checked them.
4. Write the core yourself. The need, program design, objectives, and evaluation should come from your team.
5. Trace every fact. Each number and citation should point to a source you’ve opened and read.
6. Edit with AI, then reread to make sure it didn’t change your meaning or your numbers.
7. Check the budget against the narrative, so every activity has a cost and every cost has an activity.
8. Disclose when asked or expected. Follow the funder’s rule, such as NSF’s request to note AI use in the project description.
9. Keep a record of which tools you used and for what, in case the funder asks later.

For the structure of a full proposal, see our grant proposal template and sample.

Frequently asked questions

This article is general information. It is not legal, tax, or accounting advice.