Yes. Most funders that have published a position accept applications written with AI help and judge them on content, not tools. What gets applications rejected is what AI produces when used carelessly: generic need statements, inflated outcomes and figures nobody checked. A few research funders go further and refuse applications "substantially developed by AI", so read each funder's guidance first.
A useful test: would the application still read as true, and as yours, if the assessor visited the next day? AI can organise your evidence, fit it to the questions and cut it to the word limit. It cannot supply the evidence. When it tries to, assessors notice, because they read hundreds of applications that share the same smooth phrases.
What funders have actually said about AI in applications
Positions fall into roughly three groups. These examples are from funders' own published guidance.
Accept it, judge the content
The Paul Hamlyn Foundation states: "Using AI tools alone will not disadvantage your application. But be careful of how it is used." It asks applicants to watch for AI's tendency to produce generic content, to make sure the application conveys what is distinctive about their work, and reminds them they are responsible for its accuracy and honesty. One of the largest community funders says similarly that it will not reject an application just because AI was used, while warning that AI-supported applications often "do not tell the unique story of your community" and that free tools may store what you type.
Accept it, but tell us
The MacArthur Foundation's AI policy expects agreements with grantees (other than general operating support grants) to require disclosure when AI is a primary contributor to work provided to the foundation, and it lists using AI for grant-making decisions or evaluation as not an acceptable use internally. Paul Hamlyn asks grant holders to acknowledge generative AI outputs in their work or reporting.
Restrict it
One of the world's largest public funders of medical research announced in July 2025 that, for applications submitted from 25 September 2025, it would not consider applications substantially developed by AI, or containing sections substantially developed by AI, to be the applicant's original ideas. It also capped applications at six per lead researcher per calendar year. Research funders with rules like this are the exception for small charities, but if you apply to one, their rule overrides general advice.
The research organisation IVAR noted in November 2025 that AI is already adding to the volume of applications and, used badly, is not improving their quality. Expect more funders to publish guidance, and to redesign questions so that generic answers score poorly.
Why AI-drafted applications get marked down
Assessors rarely need a detector. They mark down the symptoms:
- The same phrases as every other bid. "Person-centred support", "a safe space", "helping young people reach their full potential", "unprecedented challenges".
- Need described in general terms. National-sounding problems instead of what you see in the people you serve.
- Outcomes that don't match the budget. 200 young people "transformed" on a $15,000 grant with one part-time worker.
- Figures with no source, or invented ones. AI will write "studies show 70% of..." with nothing behind it.
- Answering a different question from the one the funder asked, because the draft came from a generic prompt.
Here is a before-and-after for a need statement (illustrative). A raw AI draft for a youth mentoring charity:
Young people today face unprecedented challenges, from rising mental health concerns to limited opportunities. Our person-centred mentoring programme helps young people overcome barriers and reach their full potential in a safe and supportive environment.
After the coordinator rewrote it with her own evidence:
Last year 31 young people were referred to us by two secondary schools; we could take 18. Of the 13 we turned away, schools told us nine were at risk of exclusion. This grant pays for a second mentor for 12 months, so we can take all referrals from both schools.
The second version has fewer adjectives and far more persuasive power, because every sentence can be checked.
Where AI helps in an application, and where it hurts
| Section | AI can help with | Keep in your own hands |
|---|---|---|
| Funder guidance | Summarising criteria, priorities, exclusions and word limits into a checklist | Deciding whether you truly fit |
| Need statement | Structuring your evidence, cutting to length | The evidence itself, and any statistic |
| Activities and outcomes | Turning your notes into clear, numbered activities | Realistic targets you can report against |
| Budget narrative | Explaining lines in plain words | Every figure, checked against the budget sheet |
| Case studies | Tightening a story you wrote, with consent obtained | The story and the permission |
| Declarations | Nothing | All of it |
Cutting to the word limit without losing the evidence
Trimming is one of the most useful jobs to hand to AI, and one of the easiest to get wrong. Asked simply to "cut this to 250 words", a model removes the specifics first, because numbers and names look like detail. An illustrative instruction that avoids it:
Cut this answer from 380 to 250 words. Keep every number, every named
school or partner, and the quote from the young person. Remove
adjectives, repetition and general statements about young people.
List what you removed underneath so I can check it.In the youth mentoring example, the "removed" list showed the model had dropped the sentence explaining that nine of the turned-away young people were at risk of exclusion, calling it "background". That sentence was the strongest evidence in the answer. The coordinator put it back and cut a line about the charity's history instead. Asking for the list of removals is what made the mistake visible.
The budget narrative: where AI numbers drift
Budget narratives are where invented or mismatched figures do the most damage, because assessors cross-check them against the budget table. Before, from an AI draft (illustrative): "The grant will fund a full-time mentor ($12,000), materials ($2,000) and volunteer expenses ($1,500), supporting over 30 young people." The budget sheet said a part-time mentor at 20 hours a week, materials of $1,800 and volunteer expenses of $1,200, which with the mentor's $12,000 totals $15,000; the draft's figures added up to $15,500 and described a different job. After the coordinator rewrote it from the sheet: "$12,000 pays for a mentor for 20 hours a week for 12 months. $1,800 covers session materials and $1,200 reimburses volunteer travel. Together, $15,000 lets us take all 31 referrals from our two partner schools." Always write the budget narrative after the budget, with the sheet pasted into the prompt, and check each figure against it.
How much AI to use, by type of funder
| Funder type | Typical position | Sensible approach |
|---|---|---|
| Small family trust with a short form | Usually no published view; reads every word | AI for structure and trimming; your own voice throughout |
| Large community funder | Often published: AI use alone isn't penalised; generic answers score badly | Follow its guidance; lead every answer with your own evidence |
| Corporate foundation or company scheme | Varies; often a structured online form | Check the terms and any declaration; keep claims verifiable |
| Public research funder | May restrict applications "substantially developed" by AI | Read the rules closely; limit AI to checking and editing, or avoid it |
One more practical test applies to all of them. Some funders visit, phone or interview shortlisted applicants. If the coordinator can't talk comfortably about every claim in the application ("you mention a waiting list of 13; how do you decide who gets a place?"), the application has drifted away from the charity's real work, whatever tool wrote it. Read the final version aloud to the person who'll take that call.
Detectors, declarations and what to disclose
AI-writing detectors are unreliable, which is a good reason for funders not to rely on them. OpenAI withdrew its own AI text classifier in July 2023, citing its low rate of accuracy. That cuts both ways: a human-written application can be wrongly flagged, and a heavily AI-generated one can pass. Don't try to "humanise" text to beat a detector; make it specific and true, which is what assessors reward anyway.
If an application form asks whether you used AI, answer honestly. A short, factual statement is enough:
We used an AI writing assistant to help structure this application
against your criteria and to edit for length. All information, figures
and case studies come from our own records and were written or checked
by our coordinator and a trustee.A worked example: a $15,000 application in five hours
Picture a small youth mentoring charity applying to a trust for $15,000 towards a second mentor. The coordinator has written similar bids by hand in about 12 hours. With AI, the process might look like this (illustrative timings):
- Criteria first (30 minutes). She uploads the funder's guidance and FAQs to a project in her chat assistant and asks for a checklist of criteria, priorities, exclusions and word limits per question.
- Talk, don't type (30 minutes). She records a 20-minute voice note walking through the project, the referrals, the waiting list and what the second mentor would do, and has it transcribed.
- Structure (30 minutes). AI sorts the transcript into the funder's questions, flags gaps ("no evidence of demand from the second school") and marks every figure with [CHECK].
- Rewrite in her voice (2 hours). She rewrites the need and outcomes answers herself using the draft as scaffolding, and fills gaps with real figures from her referral log.
- Critique (30 minutes). She asks AI to score the draft against the criteria as a sceptical assessor would, and to list any sentence that could appear in any charity's bid.
- Human read (1 hour). A trustee reads it cold, checks every [CHECK] figure against the records and the budget, and signs off.
Five hours instead of 12, and the most important two hours are still hers. The time saved went on the part that wins grants: evidence and specifics.
Prompts that keep the application yours
1. Criteria checklist
From the funder guidance attached, list: eligibility rules, stated
priorities (in their words), what they won't fund, each question with
its word limit, and anything they say assessors look for. Quote the
guidance; don't paraphrase priorities.
2. Critique as an assessor
You are an experienced assessor for this funder. Score this draft
against each criterion from 1-5 with one line of reasoning. Then list
every sentence that is generic (could appear in any charity's bid),
every claim without evidence, and every figure that doesn't match the
budget below. Do not rewrite anything.The second prompt is the more valuable of the two. Asking AI to criticise rather than write keeps the words yours and uses the tool for what it does well: spotting patterns you have stopped seeing.
When a funder's guidance says nothing about AI
Most small trusts and family foundations haven't published anything on AI. Silence usually means they judge applications as they always have, but if AI played a large part in your bid, or the funder is a research body, a short email removes the doubt. Programme officers generally prefer a question before submission to a problem after it.
Subject: Question about AI tools in applications
Dear [name],
We're preparing an application to [fund name]. We use an AI writing
assistant to help structure our answers and edit for length; all
content, figures and case studies come from our own records and are
checked by our team. Is there anything in your process we should be
aware of, or any declaration you'd like us to include?
Many thanks,
[name, role, charity]Keep the reply. If a funder later asks how the application was prepared, you have a record that you asked and followed their answer.
Build a fact bank once, draft fresh for each funder
The biggest time saving across a year comes from separating facts from writing. Keep one document, updated each quarter, with your verified figures (referrals, people supported, waiting times, costs per session), three or four consented case studies, your outcomes framework and short descriptions of each project. Every application then starts from verified material rather than from a blank prompt.
When you ask AI to draft an answer, paste in the relevant part of the fact bank and the funder's exact question, and tell it to use only those facts. That single instruction goes a long way to stopping invented statistics, because the model has real numbers to work with and has been told not to add others. It also means a second funder's application a month later reuses the same checked figures instead of fresh approximations, so your bids stay consistent if funders compare notes.
A realistic mistake shows why. A coordinator reused last year's AI-drafted answer for a new funder, and it still said "we support 120 families a year". The fact bank said 94, after a service closed. Drafting fresh from the fact bank would have caught it; copying the old answer did not.
Confidentiality and consent
Two rules protect you. First, don't paste anything confidential that belongs to the funder, such as assessor feedback marked confidential or pre-release guidance; Paul Hamlyn's guidance, for one, asks grant holders not to share its confidential information with AI tools. Second, case studies about real people need their consent for use in the application, and their details should not go into a free personal AI account. Change names or use initials, work in a business plan that doesn't train on your content, and never include health or safeguarding details in a prompt.
A final check before you submit
- You have read the funder's own guidance on AI, if it has any, and followed it.
- Every figure matches your records and the budget sheet.
- The budget narrative was written after the budget and matches it line by line.
- Anything AI removed while trimming to the word limit has been reviewed, and key evidence put back.
- No statistic appears without a source you have seen.
- Each answer responds to the exact question asked, within the word limit.
- At least one sentence per answer could only have been written by your charity.
- Case studies have consent and no identifying details beyond what was agreed.
- Any AI declaration on the form is answered truthfully.
- Someone who didn't write it has read it.
For the full drafting method, see how to use AI to write grant applications, step by step, and for tool choices, AI grant writing tools compared for small non-profits. Checking numbers is covered in more depth in catching made-up figures in AI-drafted proposals. If your monitoring reports feed your next bid, turning charity data into impact reports helps, and using AI with almost no budget covers the free tools a small charity can start with.
AI and funding applications: follow-up questions
Should I say I used AI if the form doesn't ask?
It isn't usually required, and most funders that have spoken about it care about accuracy rather than tools. A brief line in a covering note does no harm if AI played a large part. If a funder asks you to acknowledge AI use in reports once you hold a grant, as some do, keep a note of where you used it so you can answer honestly.
Will funders use AI to assess my application?
Some may use it for administration such as sorting or summarising, while others have said publicly that they don't use AI to assess applications or make grant decisions. Read the funder's AI or privacy statement. Either way, writing to the published criteria in plain, specific language helps with human and machine readers alike.
Can I reuse an AI-drafted application for several funders?
Reuse your facts, figures and stories, not whole answers. Each funder has different priorities and questions, and an answer written for one reads as generic to another. A good approach is a master document of verified facts, outcomes and case studies, from which you draft a fresh answer to each funder's actual question.
Further reads
- How to Write an AI Policy for a Small Charity — Put your grant-writing AI rules into the charity's policy.
- Charity AI Risks: Data Protection, Deepfakes, and Donor Trust — Consent and data risks when beneficiary stories meet AI tools.
- AI Hallucinations Explained for Business Owners: Causes and Fixes — Why AI invents statistics, and how to catch them.
- How to Write an AI Disclosure Statement for Your Website — Wording for a short statement on how you use AI.
- AI Proposal Mistakes That Cost You the Job — The same traps in commercial proposals.
- Charity Appeals and Social Posts With AI That Keep Your Voice — Keeping your voice in AI-assisted appeals too.
- How to Write a Business Plan With AI, and What to Check Yourself — Draft each section of a business plan with AI from your own facts, build the financials yourself, and check the claims lenders and partners will test.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Paul Hamlyn Foundation, Using AI in your work with us; MacArthur Foundation, Use of Artificial Intelligence policy; a large community funder's guidance on AI tools in funding applications; research funder notice on originality and AI (July 2025); IVAR article on AI and funding applications (November 2025); OpenAI announcement withdrawing its AI text classifier (July 2023).