It's the last week of the financial year, the appeal goes out on Friday, and you've been staring at "Dear friend of the cause" for forty minutes. AI prompts for fundraisers can get you past the blank page quickly. They can also, if you're careless, invent that your last campaign "helped 1,200 children," a sentence nobody at your organisation can back up. The first problem costs you time. The second costs you donor trust.
This post gives you prompts for donor appeals, thank-you notes, impact updates and lapsed-donor emails, with rules built around one idea: the model writes, you supply every fact. These are untested templates. I haven't run them on a live campaign, and the worked example is illustrative. For general small-organisation prompts see AI prompts for small business owners, and for the copy-paste pattern library check the prompt library.
What stays out of the prompt: donor data
Decide this before you open a chat window.
- Never paste donor names, emails, phone numbers, addresses, giving history, PAN or bank details, or notes about someone's health, bereavement or family.
- Write drafts with merge fields such as
[FIRST NAME],[LAST GIFT AMOUNT],[LAST GIFT DATE]. Your mailing tool or CRM fills them in later. - For segment analysis, paste counts ("212 donors gave once in the last 12 months"), not rows.
- Beneficiary stories need written consent before they appear anywhere, and children, patients and survivors of abuse need extra care. Change or remove identifying details, and ask your programme lead what your consent forms allow.
- Use only tools your organisation permits, and read the data terms for the exact plan you use. The privacy guide on what not to paste has the general checklist.
The fact sheet: the one thing that makes everything else safe
Every prompt below starts from the same block. Fill it once per campaign and reuse it.
FACT SHEET (the only facts you may use):
Organisation: [NAME], working on [CAUSE] in [PLACE].
What we do, in one plain sentence: [SENTENCE]
Numbers I have verified and can show records for:
- [e.g. 3 learning centres open, as of DATE]
- [e.g. Rs 600 per month covers X, per our programme budget]
Recent real update (permitted to share): [WHAT HAPPENED, WHEN]
Story (consent on file, details changed): [SHORT STORY]
Ask: [AMOUNT OR ACTION], deadline [DATE]
Tone: [warm, plain, specific]. Sender: [REAL PERSON, ROLE]
Then add this rule to every prompt: "Use only the fact sheet. If you need a number, name, date or outcome that is not on it, write [FACT NEEDED: what] and keep going. Never estimate."
That one sentence is what separates a usable draft from a liability. Without it, models fill gaps with plausible details, and "plausible" in fundraising means "an unsupported impact claim."
How do I write a donor appeal email with AI?
Ask for structure, not poetry. A good appeal is short: one person or problem, one thing that changed, one clear ask, one deadline.
Write an appeal email to [SEGMENT: e.g. past donors, no gift in 12 months].
[PASTE FACT SHEET]
Structure:
1. Subject line: give 5 options under 50 characters, none starting with
"Urgent" or using all caps.
2. First sentence: a concrete moment or fact from the fact sheet, not a
greeting or a statistic dump.
3. One paragraph on what the money does, using only fact-sheet figures.
4. The ask, with the amount options and the deadline.
5. One sentence of honest context for why now. Do not invent urgency.
6. Sign-off from [SENDER].
Rules: under 200 words. Plain English. Use merge fields as written. No
"you can make a difference", "every rupee counts", "thanks to generous donors
like you". Do not state any outcome that is not in the fact sheet.
The banned phrases aren't fussy. Donors have read those lines a hundred times, and an email full of them gets deleted.
Ask for the subject lines separately and pick one by hand. If your tool supports an A/B split, test two. I haven't got a tested open-rate figure to offer, and numbers for "best subject lines" floating around are usually from vendors measuring their own customers.
How do I draft thank-you notes that don't feel automated?
The thank-you is the email most organisations rush, and it is the first thing a new donor hears back from you. Make it specific and quick.
Draft a thank-you email for a donor who just gave [AMOUNT FIELD] to
[CAMPAIGN].
From the fact sheet use only: [ONE TRUE UPDATE or "what happens next"]
Include: thanks in the first sentence, what happens to the gift next (only
if stated in the fact sheet), how they'll hear from us and when, and a
reply-to address that a person reads.
Do not: ask for another gift, mention the amount in words, or say "your gift
fed/saved/educated" anyone unless the fact sheet states that exact outcome.
Under 120 words.
The "how they'll hear from us" line is the one most thank-yous skip. A concrete promise ("I'll write in January with what the learning centre has done") that you then keep is worth more than a warm paragraph. Receipts and tax wording belong in a separately approved block. Don't have the model write or paraphrase them.
How do I write an impact update without inventing impact?
This is the highest-risk prompt in the post, so be strict. The model should arrange results you already measured. It must not produce them.
Write a 250-word impact update for donors from the data below.
DATA (verified, from our records dated [DATE]):
[TABLE or bullet list of outputs and, if measured, outcomes]
Rules:
- Separate OUTPUTS (what we did, for example sessions held) from OUTCOMES
(what changed for people). Only report an outcome if it is in the data.
- If outcomes were not measured, say so honestly in one sentence and say
what we plan to measure.
- Do not round up, extrapolate or compare to other organisations.
- Include one thing that did not go as planned and what we changed.
- End with a list of every number you used, so I can tick it off.
Two things here are worth defending to a sceptical trustee. Asking for one thing that went wrong feels risky, but it makes the good news easier to believe (my view, not a measured result). And the "list every number you used" line gives you a checking list in thirty seconds. If a number appears in the email that isn't on your list, the model added it. Strike it.
How do I reach lapsed donors?
Lapsed-donor emails fail when they guilt-trip. Ask for a plain check-in.
Write a short email to donors who gave in [YEAR] but not since.
Do not mention the size or date of their last gift beyond the merge field
[LAST GIFT YEAR]. Tone: friendly, no guilt, no "we miss you".
Content: one real update from the fact sheet, one easy way to give again,
and one easy way to say "send me less" or "stop". Under 130 words.
The opt-out line isn't there as a compliance device; it's there because a donor who can tell you "twice a year is enough" is, in my view, more likely to stay on your list than one who feels trapped. Whether specific unsubscribe or consent wording is legally required in your situation is something to check with a professional, and I haven't stated any rule here.
Can AI help with grant narratives and board updates?
It can restructure what you know. It can't verify it. For a grant narrative, paste the funder's questions and your own programme facts, and ask for a draft that answers each question in order and flags every claim that needs a source or a number.
Funder's question: [PASTE]
Word limit: [N]
Our facts: [FACT SHEET + PROGRAMME DESCRIPTION]
Draft an answer using only these facts. After the draft, list (a) every
claim that a funder might ask us to evidence, and (b) any part of the
question the facts do not answer. Do not fill (b) with guesses.
List (b) is the valuable output, because it tells you what to find out before the deadline. Eligibility, registration and reporting requirements for any funder, especially foreign or CSR funding, are for your adviser and the funder's own guidelines. I haven't verified any government or regulatory page for this post, so I've made no claims about FCRA, 80G or similar rules. If a draft mentions one, remove it and replace it with wording your professional has approved.
An illustrative example
An invented organisation, to show the shape of the output rather than a real result.
Fact sheet says: three evening study centres open as of 1 October; a volunteer-teacher training ran in September; no learning-outcome data yet; ask is to fund books for the winter term, deadline 31 October.
A good draft opens like this:
Hi [FIRST NAME],
In September, our volunteer teachers spent a Saturday learning how to run
a study centre in a small room with no whiteboard. Three centres are open
now. [FACT NEEDED: how many children attend, if we are allowed to share it]
We haven't measured learning outcomes yet, and I'd rather tell you that than
guess. What we do know is that the winter term needs books...
The model has flagged the missing attendance figure and has stayed honest about what hasn't been measured. If your draft says "transforming lives," the prompt didn't hold.
Where this goes wrong
| Failure | What it looks like | What to do |
|---|---|---|
| Invented outcome | "Your gift fed 40 families" with no record | Fact sheet only; [FACT NEEDED] rule; number list |
| Invented donor detail | "Your support last year helped..." for someone who didn't give | Use merge fields only; never let the model assume history |
| Stock phrases | "Every rupee counts", "make a difference" | Ban list in the prompt; read aloud |
| Fake urgency | Deadlines or matching gifts that don't exist | Only state deadlines and matches on the fact sheet |
| Identifiable beneficiaries | A real child's story with details intact | Consent on file; change details; ask programme lead |
| Regulatory wording by guess | Tax or foreign-funding sentences written by the model | Professional-approved blocks only |
| Same email, every segment | Lapsed donors get the new-donor pitch | Separate prompts per segment |
What not to automate
- Facts, figures and outcomes. You supply them, you verify them, you answer for them.
- Regulatory and tax language. FCRA, tax-exemption certificates and receipts: use professionally approved wording and check the government sources with your adviser.
- Beneficiary stories without consent. Especially children and vulnerable people.
- Major-donor conversations. A person who can give a large gift deserves a call or a handwritten note from someone who knows them. A prompt can help you prepare questions. It shouldn't write the relationship.
- Sending without a read. Read each email as the donor, then ask who at your organisation could prove every sentence.
For why the "use only this" pattern works, see avoiding hallucinations, and for ways to check any AI claim before it goes out, use the fact-checking checklist.
