AI grant writing: what funders allow and how to use it well
By Mubarak MutesasiraUpdated 10 min read
Short answer:
AI grant writing is allowed by the major funders, with limits: you must stay the author of the ideas, every claim must be accurate, and reviewers may not use AI on your proposal. NIH will not treat applications substantially developed by AI as original. Use AI to draft from your own notes, build the background, map the call's criteria and review, not to invent the project.
This guide is for researchers and research managers preparing a funding application. It assumes you know the parts of a proposal; if not, start with how to write a research proposal, and for studies with human participants, how to write an IRB application.
What do funders say about AI in grant applications?
Funders allow AI-assisted writing but put the responsibility on you, and all four below restrict reviewers. These positions were checked on each funder's page in November 2026.
A funder AI policy is a funding agency's rule on how applicants may use AI to prepare applications and how reviewers may use it to assess them.
| Funder | What applicants may do | Disclosure | Reviewers |
|---|---|---|---|
| NIH (US) | Use AI, but applications "substantially developed by AI" are not treated as original [1] | Not specified in the notice | Generative AI prohibited [2] |
| NSF (US) | Use AI; responsible for accuracy and authenticity [3] | Encouraged in the project description [3] | May not upload proposal content to non-approved AI tools [3] |
| ERC (EU) | Use AI; keep "full and sole authorship responsibilities" [4] | Acknowledge help; no fixed format stated | May not use AI to summarize or assess proposals [5] |
| UKRI (UK) | Use AI with caution; no falsified, fabricated or plagiarized content [6] | Expected, except minimal use such as improving English [6] | Must not use generative AI except for language refinement [6] |
NIH: NOT-OD-25-132 and NOT-OD-23-149
NIH notice NOT-OD-25-132, "Supporting Fairness and Originality in NIH Research Applications," was released on 17 July 2025 and applies from the 25 September 2025 receipt date. It says 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," and that NIH will use current technology to detect AI-generated content. It also says NIH will accept only "six new, renewal, resubmission, or revision applications from an individual Principal Investigator/Program Director or Multiple Principal Investigator for all council rounds in a calendar year" [1].
NIH does not define "substantially." The safe reading: the aims, hypotheses and approach must be yours, and AI help should be limited to drafting, editing and search you can account for.
Since June 2023, NOT-OD-23-149 has prohibited NIH reviewers from using generative AI to analyze or write critiques of applications, citing confidentiality [2].
NSF
NSF's notice of 14 December 2023 says proposers "are encouraged to indicate in the project description the extent to which" they used generative AI and how, and that they "are responsible for the accuracy and authenticity of their proposal submission." NSF reviewers "are prohibited from uploading any content from proposals" to non-approved generative AI tools [3].
ERC
The ERC Scientific Council's position (1 December 2023) says that using outside help, including AI tools, to prepare a proposal "does not relieve the author from taking full and sole authorship responsibilities," including for acknowledgments and plagiarism [4]. In March 2026 the ERC added that evaluators may not use AI tools to summarize proposals, assess scientific merit or draft evaluations, or upload proposals to external AI systems [5].
UKRI
UKRI's policy (23 September 2024) says applicants "must apply caution when entering information into generative AI tools" and when using their outputs, must ensure the application contains nothing falsified, fabricated or plagiarized, and are "expected to be transparent" about AI use, while minimal use for English or formatting need not be disclosed. Applicants must not use generative AI to produce an entire application without human involvement. Upheld misconduct can lead to rejection and bars on future applications. Assessors must not use generative AI except for language refinement [6].
For journal and university rules, see AI in academia: university and journal policies.
Where does AI grant writing help most?
AI helps most where the work is reading and drafting, and least where the content is your own data, methods and costs.
| Section | Good AI use | Keep for yourself |
|---|---|---|
| Specific aims or summary | Turn your notes into a one-page draft; tighten wording | The problem, gap, hypothesis and aims |
| Background and significance | Find and summarize recent literature; draft with citations | Choosing which evidence matters |
| Innovation | List what existing approaches do, to sharpen your contrast | The claim of what is new |
| Approach | Check that each aim has methods, outcomes, pitfalls | Design, power, preliminary data |
| Budget justification | Draft justification text from your line items | Amounts, rates, institutional costs |
| Call compliance | Map each review criterion to a paragraph | Final judgment on fit |
Specific aims with AI
A specific aims page is a one-page summary, used in NIH applications, that states the problem, the gap, your long-term goal and hypothesis, two or three aims, and the expected impact. Other funders have an equivalent: a summary or objectives section.
Write the core yourself in bullet points, then let AI turn it into prose:
- Problem: one sentence with a number (example: "Flash floods cause most flood deaths in informal settlements in [region]").
- Gap: what is not known or not possible yet.
- Long-term goal, objective of this grant and central hypothesis.
- Aims: two or three, each starting with a verb and testing part of the hypothesis.
- Expected outcomes and impact.
Ask the AI to keep your facts and numbers unchanged and flag any claim that needs a citation. Then cut it to one page yourself.
Literature background with AI
Use AI search to find recent and highly cited work on your problem, then read the papers your argument depends on. Each sentence in the background should either cite evidence or state your inference. Our citation snowballing guide helps you find what a search misses.
Budget justification with AI
Give the AI your line items and amounts and ask for a justification paragraph per item: why it is needed, how the amount was calculated, and which aim it serves. Never let it invent amounts. Personnel effort, fringe rates and indirect costs come from your grants office.
Example justification (invented): "Field technician, 50% effort for 24 months ($[amount]): installs and maintains the 40 low-cost water-level sensors in Aim 1 and runs the monthly calibration checks in Aim 2."
How to keep an AI-assisted grant specific and checkable
Reviewers mark down proposals that are generic, and the main risk in AI grant writing is that drafts drift toward generic language. Check for it on purpose.
A checkable claim is a statement a reviewer can verify from a citation, your preliminary data or a stated method. Rewrite every sentence that is not one.
| Generic (typical AI draft) | Specific and checkable |
|---|---|
| "This innovative project will transform flood resilience." | "Aim 1 tests whether sensors costing under $[amount] each detect rising water 15 minutes before current gauges." |
| "Recent studies show growing interest in low-cost sensors." | "Three field trials since 2021 report detection rates of [x] to [y]% [citations]." |
| "Our experienced team is well placed to deliver." | "Dr. [Name] led the 2023 pilot that installed [n] sensors in [place]." |
How to review an AI-assisted grant before submission
Run this check before you send the draft to colleagues and again before submission.
- Ideas: Can you explain each aim and hypothesis in your own words, without the draft?
- Facts: Is every number from your data or a cited source?
- Citations: Does every reference exist, and does it say what the sentence claims?
- Call fit: Does each review criterion in the call have a paragraph that answers it?
- Limits: Is every section within the page or character limit and in the required template?
- Disclosure: Does the funder expect an AI statement, and is it accurate?
- Confidentiality: Did you avoid pasting collaborators' unpublished data into tools your institution does not approve?
- Second reader: Has a colleague outside your subfield read the summary?
How to draft a grant in AI for Research
AI for Research has a grant proposal tool in Proposal Studio. You choose the format (generic, NIH Specific Aims and Research Strategy, NSF Project Description, ERC Extended Synopsis or Wellcome), describe the project, and paste or upload the funding call so the draft addresses its criteria.
- Whole proposal (15 credits) or one part (3 credits): summary or specific aims, significance and innovation, approach and work plan, impact and dissemination, or budget justification.
- The significance draws on a literature search and cites real papers. Amounts and team details come back as [placeholders] for you to fill in; the tool does not invent them.
- Find Papers (free) and My Library for background reading; ask a PDF questions with page-cited answers (1 credit each).
- Reviewer set to "proposal" scores each criterion out of 10 and quotes passages with a fix (10 credits per 10,000 words). Citation check confirms references exist (free).
- Proposal Studio also drafts a data management plan (3 credits) in NIH, NSF, Horizon Europe or UKRI format.
Try the grant proposal writer without an account, or the research proposal generator for a non-funded proposal. The free plan has 45 credits a month; see pricing. Limits: the tool does not know your institution's rates or the funder's latest template, and a draft it writes is not your original idea under NIH's policy until you have made the aims and approach your own. Use it for structure, background and review.
FAQ
Does NIH allow AI in grant applications?
NIH does not ban AI tools for applicants, but notice NOT-OD-25-132 (July 2025) says applications substantially developed by AI, or with sections substantially developed by AI, will not be considered the applicant's original ideas. NIH says it uses detection technology and may act after an award. The same notice limits each principal investigator to six applications per calendar year. Reviewers may not use generative AI at all.
Can AI write a grant proposal?
AI can draft sections, structure the proposal to the funder's headings and find literature for the background, but it cannot supply your preliminary data, your team's track record or a budget that matches your institution's rates. Funders hold you responsible for every claim, and NIH will not treat AI-developed applications as original. Use AI for drafts and checks, and write the ideas yourself.
Do I have to disclose AI use in a grant application?
It depends on the funder. UKRI expects applicants to be transparent about generative AI use, except minimal use such as improving English. NSF encourages proposers to say in the project description how AI was used. The ERC says AI use does not relieve you of full authorship responsibility. Read the current call and the funder's policy page, and disclose when in doubt.
Is it safe to paste my grant idea into an AI tool?
Be careful. UKRI tells applicants to apply caution when entering information into generative AI tools, and unpublished ideas, collaborators' data and personal information may be stored or reused depending on the tool's terms. Check your institution's approved tools, avoid pasting confidential material from others, and read the tool's data policy before uploading a draft.
What part of a grant is AI most useful for?
AI is most useful for the literature background, turning your notes into a first draft of the aims page, mapping the call's criteria against your draft, and spotting vague or unsupported claims. It is least useful for the approach and budget, which depend on your methods, preliminary data and institutional costs. Treat those sections as yours to write and AI as a reviewer.
Next steps
- Get the structure right first: how to write a research proposal.
- See a worked example: research proposal example.
- Prepare the ethics side: how to write an IRB application.
- Check the rules for papers too: AI in academia: university and journal policies.
Sources
- National Institutes of Health. NOT-OD-25-132: Supporting Fairness and Originality in NIH Research Applications. 17 July 2025. https://grants.nih.gov/grants/guide/notice-files/NOT-OD-25-132.html (checked November 2026)
- National Institutes of Health. NOT-OD-23-149: The Use of Generative Artificial Intelligence Technologies is Prohibited for the NIH Peer Review Process. 23 June 2023. https://grants.nih.gov/grants/guide/notice-files/NOT-OD-23-149.html (checked November 2026)
- National Science Foundation. Notice to Research Community: Use of Generative Artificial Intelligence Technology in the NSF Merit Review Process. 14 December 2023. https://www.nsf.gov/news/notice-to-the-research-community-on-ai (checked November 2026)
- European Research Council. Current position of the ERC Scientific Council on AI. 1 December 2023. https://erc.europa.eu/news-events/news/current-position-erc-scientific-council-ai (checked November 2026)
- European Research Council. ERC clarifies limits on AI use in grant evaluation. 24 March 2026. https://erc.europa.eu/news-events/news/erc-clarifies-limits-ai-use-grant-evaluation (checked November 2026)
- UK Research and Innovation. Use of generative artificial intelligence in application preparation and assessment. 23 September 2024. https://www.ukri.org/publications/generative-artificial-intelligence-in-application-and-assessment-policy/use-of-generative-artificial-intelligence-in-application-preparation-and-assessment/ (checked November 2026)
- AI for Research. Grant proposal writer. /tools/grant-proposal-writer