Using AI in grant writing responsibly: policies and workflows

AI writing tools have become part of many grant offices' daily routines. Used well, they can shorten tedious tasks: turning notes into an outline, tightening a paragraph to fit a character limit, summarizing a 90-page funding notice, or checking a draft against the funder's criteria. Used carelessly, they can produce confident fabrications, generic prose that weakens fit, and confidentiality breaches you cannot undo.

The principle that should guide every decision is simple: you are accountable for everything you submit. Funders do not accept "the tool made it up" as an explanation, and some treat AI-generated content as an originality or integrity issue. This guide covers the verified funder policies, the main risks, and practical workflows that let you benefit from AI tools while staying in control.

What funders actually say (as of October 2026)

Funder policies on AI are evolving quickly, and most private foundations have not published one. Below are policies verified from the funders' own sites. Always check the current guidelines and the specific funding notice before you apply.

NIH: originality and application limits

On July 17, 2025, NIH issued NOT-OD-25-132, "Supporting Fairness and Originality in NIH Research Applications." It does two things:

• AI and originality. NIH "will not consider applications that are either substantially developed by AI, or contain sections substantially developed by AI, to be original ideas of applicants." If AI use is detected after an award, NIH may refer the matter to the Office of Research Integrity while taking enforcement actions, which can include disallowing costs, withholding future awards, suspending the grant, and possible termination. NIH also said it will use the latest technology to detect AI-generated content.
• Application limits. NIH will accept only six new, renewal, resubmission, or revision applications per individual principal investigator (including multiple-PI applications) across all council rounds in a calendar year. The limit applies to all activity codes except T activity codes and R13 conference grants.

The policy took effect for applications submitted to the September 25, 2025 receipt date and beyond. NIH's notice acknowledges that AI "may be a helpful tool in reducing the burden of preparing applications" and "may be appropriate to assist in application preparation for limited aspects or in specific circumstances," while warning that AI use "may result in plagiarism, fabricated citations, or other kinds of research misconduct."

NIH: reviewers may not use generative AI

Since June 2023 (NOT-OD-23-149), NIH has prohibited its peer reviewers from using natural language processors, large language models, or other generative AI technologies to analyze or formulate critiques of grant applications. NIH states that uploading or sharing content or original concepts from an application or critique to online generative AI tools violates peer review confidentiality. Reviewers must certify they understand this prohibition.

NSF: reviewer prohibition and encouraged disclosure

NSF's December 14, 2023 notice to the research community sets guidelines for both sides of merit review:

• Reviewers are prohibited from uploading any content from proposals, review information, and related records to non-approved generative AI tools. NSF treats information uploaded to AI tools outside its firewall as entering the public domain.
• Proposers are encouraged to indicate in the project description the extent to which, if any, generative AI was used and how. NSF states that proposers "are responsible for the accuracy and authenticity of their proposal submission," including content developed with AI assistance, and notes that fabrication, falsification, or plagiarism can be research misconduct.

Private foundations

Some private funders have published policies. For example, the Wenner-Gren Foundation, which funds anthropological research, adopted a generative AI policy in March 2025. It strongly encourages applicants to disclose whether and how they used generative AI, holds applicants responsible for ensuring proposals contain no plagiarized writing or fabricated information (detection results in removal from consideration), and prohibits its reviewers from uploading applications or review materials to generative AI tools.

Most community, family, and corporate foundations have no written AI policy. That is not permission to stop thinking. Their program officers are still evaluating whether your proposal sounds like your organization and whether its facts are true.

The real risks

Fabricated facts and citations

Generative AI tools produce fluent text that can include statistics, studies, quotes, and citations that do not exist, or that exist but do not say what the tool claims. In a grant proposal, one invented statistic in your need statement can destroy a reviewer's trust in everything else. NIH specifically names "fabricated citations" as a risk.

Rule: Never put a number, citation, quote, or factual claim in a proposal unless a person on your team has checked it against the original source.

Outdated or wrong information about rules and funders

AI tools often have outdated or incorrect information about funders' priorities, deadlines, grant sizes, and federal rules. The federal grants landscape changed substantially in 2025 and 2026; a tool trained earlier may give you the old de minimis rate, an expired NOFO, or a policy that has been rescinded.

Rule: Get funder and regulatory information from the funder's own materials and primary sources, such as the NOFO, eCFR, and the funder's website.

Generic text that weakens fit

AI drafts tend toward smooth, generic language. Reviewers read hundreds of proposals; generic text reads as low effort and poor fit, regardless of who wrote it. The details that win grants (your community, your data, your staff, your results) come from you, not a tool.

Plagiarism and originality

AI tools can reproduce phrasing from existing text, and some funders treat substantially AI-generated content as not original. For research funders, this can rise to a research integrity issue.

Confidentiality and privacy

Anything you paste into a public AI tool may be stored, reviewed, or used in ways you cannot control, depending on the tool and its settings. That is why NIH and NSF prohibit reviewers from uploading application content. The same logic applies to your own sensitive information.

Never paste into a public or unapproved AI tool:

• Another organization's proposal, or any reviewer materials, if you serve as a reviewer
• Confidential funder communications or materials marked confidential
• Client or participant personally identifiable information, health information, or case notes
• Donor information
• Staff personnel information
• Unpublished data or partner information you do not have permission to share

If your organization uses AI tools, choose ones with data protections appropriate for your information, read the terms, and set a written policy for staff.

Disclosure: when and how

Disclose AI use when:

• The funder requires it.
• The funder encourages it (as NSF and Wenner-Gren do).
• The funder asks you directly.

When the funder is silent, use judgment. Using a grammar checker or asking a tool to shorten a sentence is not the same as having a tool draft whole sections. If AI played a substantive role in drafting, a brief, factual note may be appropriate where the funder allows, and you should be ready to answer honestly if asked.

Keep disclosures short, specific, and accurate. Do not overstate or understate what the tool did.

Workflows that keep you accountable

The safest pattern is: people own the substance, tools help with the mechanics, and people verify everything at the end.

Good uses

• Reading and summarizing funding notices. Use a tool to produce a first-pass list of requirements and criteria from a long NOFO, then verify the list against the original before building your compliance matrix.
• Outlining. Turn your notes and the funder's criteria into a section outline that mirrors their headings.
• Editing for length and clarity. Tighten a paragraph to fit a character limit, simplify jargon, or vary sentence structure, using your own draft as the input.
• Consistency checks. Ask a tool to list every number in the draft so you can check each against the budget and your source documents.
• Reviewer simulation. Ask a tool to critique your draft against the funder's published criteria, then decide yourself which suggestions are valid.
• Plain-language translation of technical content for a general reader, which you then check for accuracy.

Risky uses

• Asking a tool to "write a grant proposal" for a program from a short prompt
• Generating need statistics or citations
• Asking a tool for information about a funder's priorities or deadlines without checking the source
• Drafting partner letters of commitment for partners who will not review them
• Producing reports or outcome data descriptions without checking against your actual records

A practical step-by-step workflow

1. Check the funder's AI policy and the NOFO before using any tool.
2. Write the substance yourself: the need (with your sources), program design, outcomes, staffing, and budget logic. Gather your real data first.
3. Use tools for structure and polish, feeding them your own draft and non-confidential material only.
4. Verify every fact, number, and citation against primary sources and your records. Keep a source log.
5. Rewrite in your organization's voice. Add the specific details only you know.
6. Run a human review using the proposal review checklist, including a cold read by someone who knows the program.
7. Disclose as the funder requires or encourages.
8. Keep records of what tools you used and how, in case questions arise.

An organizational AI policy for grant work

If several people write grants for your organization, or you work with freelance grant writers, a one-page policy prevents confusion. It might cover:

• Approved tools and the data settings required
• Prohibited inputs (client data, donor data, confidential funder materials, others' proposals)
• Required verification of all facts, numbers, and citations before submission
• Disclosure practice for funders that require or encourage it
• Who signs off on final proposals
• Expectations for contractors, including in your grant writer contract

If you hire a freelance writer, ask how they use AI tools, how they protect your information, and how they verify facts. Our guide to hiring a grant writer covers contracts and ethics, and the grant writer contract template is a starting point for writing these expectations into an agreement.

If you serve as a reviewer

Many grant professionals also review for foundations, community foundations, or government panels. Treat every application and review document as confidential. Do not upload them to AI tools unless the funder has explicitly approved a specific tool for that purpose. Federal agencies such as NIH and NSF prohibit it, and violating reviewer confidentiality can have serious consequences.

The bottom line

AI tools can save time on mechanics, but the things that win grants have not changed: real fit with the funder, a specific and well-documented need, a credible plan, honest numbers, and an organization the funder can trust. Use tools to help you say those things more clearly, never to invent them. For the essentials of each proposal section, start with the anatomy of a grant proposal, and for building a need statement on verified data, see the statement of need guide.