AI Grant Writing: What It Can Do, Funder Rules, and the Risks

AI grant writing means using artificial intelligence tools, such as general chat assistants or purpose-built grant software, to help research funders, summarize requirements, outline and draft grant applications. These tools are genuinely useful for the routine, high-volume parts of the work. They are unreliable at the parts reviewers score: fit with the funder, strict compliance with the funding announcement, accurate evidence and a credible plan. Some funders, notably NIH and NSF, also have rules about how AI may be used, so the right question is not only whether a tool can draft a section but whether the funder allows it.

The kinds of tools available

• General chat assistants. Large language models that answer questions and draft text on request. Flexible, but they know nothing about your program unless you tell them.
• Purpose-built grant writing software. Tools that combine funder search with drafting, often storing your organization's past proposals so drafts reflect your own language.
• Grant databases with AI search. Prospecting platforms that use AI to match your mission with funders.
• Writing and editing assistants. Tools focused on grammar, clarity and length rather than content.

The category matters less than how the tool fits into a process where a person makes the decisions and checks the facts.

What funders say about AI

Policies are changing quickly, so always check the current guidance for the program you are applying to.

NIH. Under notice NOT-OD-25-132, NIH will not consider applications that are substantially developed by AI, or that contain sections substantially developed by AI, to be the original ideas of the applicant, and such applications may be found noncompliant. Misuse that leads to plagiarism or fabricated citations can be referred to the Office of Research Integrity. The same notice limits each principal investigator to six new, renewal, resubmission or revision applications per calendar year, with exceptions for certain training and conference grants; NIH cited a surge of applications, some possibly AI-generated, as part of the reason. Separately, NIH prohibits its peer reviewers from using generative AI to analyze applications or write critiques (NOT-OD-23-149), because uploading application content to these tools breaches review confidentiality. More on NIH applications is in the NIH grants guide.

NSF. NSF's December 2023 notice on generative AI encourages proposers to indicate in the project description whether and how they used generative AI to develop the proposal. It also bars reviewers from uploading proposal content to non-approved AI tools, on the basis that information entered into public tools can no longer be kept confidential. See the NSF grants overview for how NSF proposals are structured.

Other funders. Some foundations and agencies have begun publishing their own positions, ranging from disclosure requirements to outright restrictions. If the guidelines are silent and the stakes are high, ask the program officer.

Where AI genuinely helps

• Understanding a long announcement. A tool can summarize a lengthy federal funding notice, list dates and required attachments, and explain unfamiliar terms. Verify every requirement against the original document afterwards.
• Outlining. Given the required sections and scoring criteria, a tool produces a sensible skeleton to write into.
• Routine text. Organizational background and descriptions of standard activities are lower-risk places to start from a machine draft that a person then sharpens.
• Editing to length. AI is a capable line editor for cutting a paragraph to a character limit, removing jargon or fixing passive constructions.
• Plain-language explanations. Dense instructions can be rephrased so your whole team understands what the funder is asking.

Where AI falls short

• Invented facts and citations. Language models produce statistics, studies and references that look authoritative and do not exist. In a needs section built on cited data, a fabricated figure is a credibility problem, and for research funders it can become a misconduct problem.
• Compliance. Page limits, required forms, formatting rules and specific questions in the announcement are not reliably tracked, and a single noncompliance can end an application before review.
• Budgets. A tool does not know your salaries, fringe rate or negotiated indirect rate, or which costs a given funder allows. A defensible grant budget needs real figures and the funder's cost rules.
• Generic writing. Reviewers reading many applications notice text written for an average reader rather than for this funder's priorities.
• Strategy. Which opportunity to pursue, how to frame the project and what to leave out depend on context a tool does not have.

Protecting confidential ideas

For research applications especially, the content of a proposal is valuable intellectual property. Both NIH and NSF treat uploading application content to public AI tools as a confidentiality breach when reviewers do it, and the same logic applies to applicants: anything typed into a tool may be stored or used under that tool's terms. Before pasting unpublished aims, preliminary data or partner information into any AI product, read its data retention and training terms, and check whether your institution has an approved tool.

A safer way to use AI on an application

1. Decide the strategy first. Choose the opportunity and the angle before opening any tool.
2. Use AI for summaries and an outline, then check every requirement against the funding announcement yourself.
3. Write the high-stakes sections from your own evidence. The need, the project design and the specific aims or objectives should come from your team's knowledge and data.
4. Verify every number and citation against the primary source. Treat all machine-produced facts as unconfirmed until checked.
5. Run a compliance pass against the funder's own checklist, not against what a tool says the rules are. Our pre-submission checklist helps.
6. Disclose AI use where the funder asks for it, and keep a note of how tools were used.
7. Have an experienced person review the whole package for strategy, accuracy and compliance before submission.

The fundamentals that decide outcomes have not changed, which is why our guide to writing a grant proposal from scratch still applies however a first draft was produced. If you want an experienced reader to check your application against the funder's criteria before it goes in, a professional grant review is built for exactly that.