AI for Grant and Proposal Writing: What Funders Allow

TL;DR:

• Yes, you can use AI on most grant proposals if a person stays in charge and every claim is true. Check each funder's own rule first, because they differ.
• Several funders ask you to declare substantive AI use. Light language help usually needs no declaration.
• The clear red lines are generating a whole application with no human involvement, inventing data or citations, and entering confidential or personal data.
• NIH in the US is stricter than most: it does not treat applications substantially developed by AI as the applicant's original ideas.
• Build a fact bank first, keep the budget in a spreadsheet, keep an AI use log, and check every claim before you submit.

Who this guide is for

This guide is for grants officers, programme managers, founders and consultants who write funding proposals for NGOs, community organisations and social enterprises. It applies whether you are writing a two-page concept note or a full application.

We focus on funder rules you can check yourself. Where we could not confirm a rule, we say so.

What funders actually say about AI (November 2026)

Funders do not have one shared rule, but most published rules allow some AI use with conditions. These are the ones we opened and read.

Three notes on this table.

First, most of these funders are in research or UK community giving. They are useful because they show how funders think, and because many of them publish their rules openly.

Second, we looked for public AI rules from the UK's FCDO and the EU's international partnerships department and found none for grant applicants. Absence from our search does not mean no rule exists. Check the call document and contract, and ask the donor's contact person.

Third, these policies change. Open the funder's current page before every application.

The four patterns behind the rules

Four patterns repeat across the rules above, and they are a good default when a funder says nothing.

You stay responsible. Every funder that addresses it puts accuracy on the applicant. "The AI wrote it" is no defence.

Light help and real authorship are treated differently. Wellcome and UKRI separate language polish from substantive use such as generating ideas. Spencer separates grammar help from AI-written content.

Confidentiality matters. Wellcome tells applicants to keep confidential and personal data out. The Heritage Fund warns that free tools may store what you enter.

Reviewers have limits too. UKRI, Wellcome and the European Commission all restrict how assessors may use AI on proposals. They treat the content of your application as confidential.

Is AI detection something to worry about?

Plan to be able to defend every sentence, whether or not anyone runs a detector. NIH's approach, as summarised by Emory University's research office, includes referral to the US Office of Research Integrity if AI use is detected after an award, with possible consequences such as suspension or termination of a grant.

We could not open NIH's own notice when we checked, so read notice NOT-OD-25-132 yourself if NIH funding applies to you.

Do not spend time trying to make AI text "undetectable". If a funder requires disclosure, hiding AI use is the real risk. If your facts are true and your thinking is your own, disclosure costs little: Wellcome, UKRI and the European Commission all say so in their different ways.

What is safe, what needs care and what to avoid

Use AI for structure, clarity and challenge. Keep facts, promises, budgets and strategy with people.

A seven-step workflow

This workflow keeps the thinking with your team and gives AI the clerical and challenging work.

1. Decide if you should apply. Your leadership decides, based on fit, capacity and honest odds. AI can list the eligibility rules, but it cannot judge your readiness.
2. Read the funder's AI rule. Copy it into your proposal folder with the date.
3. Build a fact bank. Gather proven results, numbers with sources, past grants you may name, partner confirmations, and staff roles.
4. Build a compliance matrix from the call document.
5. Write the core yourself. Write the problem, the change you expect and the reason your organisation is the right one, even if only as rough bullet points. This is what makes the proposal yours.
6. Use AI on structure, drafts and challenge, keeping a log as you go.
7. Run a claim check, then have someone who did not write the proposal read it against the fact bank and the matrix.

A worked example you can copy

Everything in this example is invented for training, including the fund and the organisation. Replace it with your own material.

The call: The "Community Climate Resilience Fund" invites concept notes of up to 1,500 words from registered East African NGOs, for 18-month projects.

The fact bank:

Organisation: Shamba Pamoja (invented), registered NGO, works with smallholder farmer groups in one county. Proven: Trained 340 farmers in water harvesting, 2024 to 2025. Source: attendance register. Proven: Follow-up survey of 120 of those farmers in 2025; 74 reported using at least one technique. Source: survey report. Not proven: any change in yield or income. We did not measure it. Partner: County agriculture office has confirmed support in writing, letter dated [date]. Staff: Programme lead with agronomy background. Two field officers. External statistics: none yet. [Team to find and read a published source.]

Note the "not proven" line. Writing down what you cannot claim is the single most useful thing in the file.

Prompt 1: Turn the call into a compliance matrix

Below is a funding call. Make a table with four columns: Requirement, Exact words from the call, Where it goes in our proposal, Status (leave blank). Include eligibility, word limits, required sections, deadlines, attachments and scoring criteria. If something is unclear, list it as a question for the funder. Add nothing that is not in the call. Call: [paste public call text]

Prompt 2: Draft from the fact bank only

Clarity: Draft the "Track record" section of a concept note. Context: We are the organisation in the fact bank, applying to the fund in the call summary. Constraints: Use only the fact bank. Do not add numbers, partners, outcomes or sources. Do not suggest any change in yield or income, because we did not measure it. Where more evidence would help, write [EVIDENCE NEEDED: what]. Maximum 150 words. Avoid words such as "transformative", "empower" and "holistic". Customisation: The reader is a reviewer who scores many proposals and wants specifics. Write in short, direct sentences, in our voice. Fact bank: [paste]

The kind of result you want:

Between 2024 and 2025, Shamba Pamoja trained 340 smallholder farmers in water harvesting. In a 2025 follow-up survey of 120 of those farmers, 74 said they were using at least one technique they had learned. We have not yet measured changes in yield or income, so this project includes a simple way to do that. [EVIDENCE NEEDED: confirm this measurement is in the plan and budget.]

Admitting the measurement gap makes the proposal more believable, and it turns a weakness into a project activity. Your team must still confirm the promise is real and funded.

Prompt 3: Challenge the logic

Act as a sceptical reviewer. Here is our theory of change in bullet points. Do not rewrite it. List the five weakest links between our activities and the change we promise. For each one, say what evidence or assumption is missing. Do not invent evidence. Theory of change: [paste]

Some funders would count this as substantive use, so log it.

Prompt 4: Score it the way the funder will

Below are the funder's scoring criteria and our draft. For each criterion, give a score out of five and two sentences of reasons, quoting our text. List the three changes that would raise the score most. Do not suggest claims we cannot prove. Fact bank: [paste]. Criteria: [paste]. Draft: [paste]

Treat the score as a prompt for your own thinking. It tells you nothing about what a real reviewer will decide.

Prompt 5: Audit the claims

Below are our fact bank and final draft. List every number, name, partner, date, result, promise and cited source in the draft. For each, say whether it appears in the fact bank (Yes, Partly, No). Do not rewrite anything. Fact bank: [paste]. Draft: [paste]

Then do the same audit by hand, at least for numbers, partners and promises.

Keep the budget out of the chat window

Build your budget in a spreadsheet with your finance lead. The National Lottery Community Fund has been reported as telling applicants to check AI-suggested budgets because it assesses value for money, and the Heritage Fund advises against using AI to generate one.

AI can still help safely in one way. Take your finished budget, remove names and salaries, and ask:

Below are our activity list and budget headings, with no names or individual salaries. List any activity that has no matching budget line, and any budget line with no matching activity. Do not suggest amounts. Activities: [paste]. Budget headings: [paste]

For Excel help, ask for formulas, and keep real figures in your own file.

What goes wrong

The first row is the mistake that hurts most. A number can look right, feel right and still have no source. Under deadline pressure, people accept it. For official Kenyan figures, start at the source, such as the Kenya National Bureau of Statistics or the relevant ministry, and read the original table.

Keep an AI use log

A simple log lets you answer any funder's question in five minutes. Keep one for each application.

Writing a disclosure statement is easy once you have the log. Spencer's template asks for the tool, and where, how and why it was used, along with confirmation that the authors reviewed the output and accept responsibility. That is a good structure for any funder.

A sample statement, to adapt only if every part is true:

We used [tool name] to organise the funder's requirements, to produce a first draft of the [section] from our own verified records, and to edit for clarity. Our staff reviewed and verified all content. We take full responsibility for this application. No personal or confidential data was entered into the tool.

If the funder gives its own wording or form, use that.

What to do when the funder says nothing

When a funder is silent, write to the contact person. A short question costs nothing and shows care.

Dear [Name], we plan to apply to [call]. Does [funder] have any guidance on using generative AI tools during proposal preparation, or any disclosure requirement? We would use such tools only with human review, and would not enter personal or confidential data. Thank you. [Name]

If you hear nothing, follow the common pattern: human authorship of the core thinking, no personal data, checked facts, and a brief honest note if AI did substantive work. This is our opinion and it goes beyond what a silent funder requires.

Your staff need the same rules each time. Our guide to writing an AI usage policy for Kenyan organisations can help. For how Kenyan data law treats personal data, including staff CVs, the data protection and AI guide was written for public bodies but its principles apply widely.

Final checklist before you submit

• We read this funder's current AI rule and saved a copy.
• Every requirement in the call is ticked off in the matrix.
• The problem, change and reasons for our organisation came from our team.
• Every number has a source that a person opened and read.
• Every citation exists and says what we claim.
• Every partner named has agreed in writing.
• Delivery staff confirmed that we can do what we promise.
• Finance built and checked the budget.
• No personal or confidential data went into an unapproved tool.
• The AI use log is complete.
• Any required declaration is filled in and true.
• A person who did not write it has read it against the fact bank.
• No [EVIDENCE NEEDED] markers remain.

Questions people ask

Can a funder reject my proposal for using AI?

Some can, in particular cases. Spencer Foundation reserves the right to reject proposals that show substantial AI use, especially if it was not disclosed. Others, including Wellcome and the Heritage Fund, say that AI use or a declaration of it does not by itself count against you.

Do I have to disclose that I used ChatGPT for grammar?

Usually not, under the rules we checked. Wellcome and UKRI treat language help as minimal use, and Spencer does not require disclosure for grammar, spelling and punctuation. Check the rules of your own funder.

Can AI write a whole grant proposal?

It can produce text, but several funders bar it from generating a whole application without human involvement. It also tends to produce weak proposals, since it does not know your results, partners or community.

Is it safe to put our proposal into a free AI tool?

It depends on what is in the proposal and on the tool's settings. Remove personal data, confidential partner information and unpublished results first. Use an account your organisation has approved.

Does AI help non-native English writers?

Yes, it can. Wellcome and UKRI treat translation and improving English as minimal use. The National Lottery Community Fund has presented its AI guidance as improving accessibility for organisations applying for funding. Your ideas stay yours.

How much time does AI save on proposals?

We have no reliable figure, and you should be careful with anyone who gives you one. Time your own next proposal in full, including checking.

Sources

• Wellcome, Use of generative AI funding policy (updated 8 April 2026), last checked 9 November 2026
• UK Research and Innovation, Use of generative AI in application preparation and assessment, last checked 9 November 2026
• European Commission, Horizon Europe Proposal Evaluation Standard Briefing, version 13.0 (hosted by Leuphana University), last checked 9 November 2026
• Netherlands Enterprise Agency (RVO), Horizon Europe and Eureka: how to (not) use AI in grant proposal writing, last checked 9 November 2026
• Emory University, Supporting Fairness and Originality in NIH Research Applications, last checked 9 November 2026
• Spencer Foundation, Policy on the Use of Generative AI, last checked 9 November 2026
• MacArthur Foundation, Artificial Intelligence policy, last checked 9 November 2026
• Civil Society, Major funder says charities can use AI tools when applying for grants (27 January 2025), last checked 9 November 2026
• The National Lottery Community Fund, AI can be a powerful force for good (24 January 2025), last checked 9 November 2026
• National Lottery Heritage Fund, Using artificial intelligence in your funding application, last checked 9 November 2026
• Art Fund, Using AI to apply for a grant or funding, last checked 9 November 2026

Where FuKazee fits in

FuKazee trains teams in practical AI for everyday work, and proposal writing is part of our NGO training. We assume nobody in the room knows AI yet, avoid jargon and say plainly that nobody becomes an expert in a few days. We show the free version of every feature, teach the four Cs before anything more advanced, and never promote one tool.

In a proposal session your team would practise building a fact bank, a compliance matrix and a claim check on invented or cleaned material. Training cannot give you a track record or a good project idea. It can help your team present real work clearly and honestly.

See our approach in how to train NGO staff on AI in Kenya, or visit our page on AI training for NGOs in Kenya.

To talk through training for your grants team, email lead@fukazee.com