Ten prompts cover most of a small nonprofit's writing: meeting minutes, an email-thread summary, a one-page policy draft, a funder update and an answer trimmed to a word limit, plus a first-gift thank-you, an appeal letter, a lapsed-supporter email, an impact update and a monthly-giving invitation. None needs a donor's name, address or gift history pasted in.
The method is merge fields. ChatGPT writes the letter with [first name], [gift amount] and [campaign] left as literal text, and your own database or spreadsheet fills them in when you send. That keeps donor records out of the chat, and it makes each prompt reusable, which is what turns ten prompts into a month of work.
What may go into a prompt, and what stays in your database
Settle this before anything else, because it is the one rule that matters more than any wording. The examples follow a made-up organisation: a community garden project with two part-time staff, 35 volunteers and about 600 supporters. All of its figures, stories and quotes are illustrative, and yours must be real. The drafts shown are illustrative too: they demonstrate the kind of slip to look for, and yours will differ.
| Paste as written | Replace with a tag first | Never paste |
|---|---|---|
| Your mission statement, published reports and programme descriptions | Donor first names as [first name]; gift amounts as [gift amount]; campaigns as [campaign]; dates as [date] | Donor lists, addresses, email addresses and phone numbers |
| Event details and costs per item that you already publish | Volunteers as [volunteer]; trustees by role, such as "the Chair" | Bank or card details, and giving history by named person |
| A story the person agreed in writing to share, with identifying details removed | Beneficiaries as "a participant", with no names and no ages if small numbers could identify them | Case-file notes, and health or safeguarding information about anyone |
That is the short form. What donor data must never be pasted into ChatGPT gives the longer one. In practice the habit is simple: if a prompt could be read aloud at a public meeting without embarrassing anyone, it is safe to paste.
The ten prompts at a glance:
| # | Job | You paste in | You check |
|---|---|---|---|
| 1 | Minutes and actions | Rough notes, roles not names | The decisions list against your notes |
| 2 | Email thread summary and reply | The thread, with letters for names | That each quoted line is in the thread |
| 3 | One-page policy draft | How you actually work | Every invented rate or rule |
| 4 | Funder update | Planned and delivered figures | Percentages and stated reasons |
| 5 | Answer cut to a word limit | Your full answer | Numbers kept, and the real word count |
| 6 | First-gift thank-you templates | Impact facts, merge fields | No tax or receipt promises |
| 7 | Appeal letter | One agreed story, one ask | No invented details |
| 8 | Lapsed-supporter email | One update, one small ask | No reference to their giving |
| 9 | Quarterly impact update | A few numbers and a quote | No unsourced claims |
| 10 | Monthly-giving invitation | Levels and what each covers | The sums, and no claims about donor behaviour |
1. Rough meeting notes to minutes and an action table
Minutes are a good first job because the facts are already in your notes. The risk is that a tidy summary quietly turns "discussed" into "decided", so the prompt separates the two.
Turn the rough notes below into draft minutes for a nonprofit board meeting.
Format: attendance by role; decisions; actions as a table (action, owner role, due date); items for the next agenda.
Record something under "Decisions" only if the notes say "agreed" or "approved". Anything proposed or discussed goes under "Discussed, not decided".
If an owner or a date is missing, write [CHECK] instead of guessing. Use roles, not names. 250 words maximum.
NOTES:
Present: Chair, Treasurer, Secretary, 2 trustees, Project Lead.
Treasurer: bank balance fine, grant instalment arrived. Suggested moving part of reserves to a higher-interest account. Chair wants to see options first.
Agreed: volunteer day on the second Saturday, Project Lead to book the hall.
Polytunnel quote $2,300, trustees happy in principle, need a second quote.
Secretary to send the AGM notice. Chair asked for the safeguarding policy to go on next agenda.A typical result:
Draft minutes. Present: Chair, Treasurer, Secretary, two trustees, Project Lead.
Decisions: (1) The volunteer day will be held on the second Saturday, and the Project Lead will book the hall. (2) Part of the reserves will be moved to a higher-interest account.
Discussed, not decided: the polytunnel quote of $2,300. Trustees are favourable in principle and a second quote is needed.
Action Owner Due Book the hall for the volunteer day Project Lead [CHECK] Obtain a second polytunnel quote [CHECK] [CHECK] Send the AGM notice Secretary [CHECK] Compare higher-interest account options Treasurer [CHECK] Next agenda: safeguarding policy; polytunnel decision.
The second "decision" is wrong. The notes say the Treasurer suggested moving part of the reserves and the Chair wanted options first, so nothing was agreed. The rule asked for "agreed" or "approved", and the draft slipped anyway. Read the Decisions list line by line against your notes every time, because a minute that records a decision nobody took is a record of the wrong thing. Treat the output as draft minutes, to be approved at the next meeting like any other. The same check applies if a note-taking tool produces the first draft instead of you: its summary is a draft too.
2. A long email thread, summarised with the evidence beside it
Threads with a hall hire, a supplier or a funder grow long, and the current position hides in the middle. The prompt makes ChatGPT quote the line each statement came from, so a wrong summary is easy to catch.
Below is an email thread with names replaced by letters (A is us). Summarise it in five lines or fewer: what has been agreed, what is still open, and who is waiting on whom.
After each statement, quote in brackets the line of the thread it came from. If the thread does not say, write [not in thread].
Then draft a reply from us of 80 words or fewer that proposes the next step. Do not agree to anything that is not already agreed in the thread.
THREAD (identifiers removed):
[paste the thread here]What came back for a thread about hiring a hall for a fundraising lunch:
Agreed: the hall is booked for the 12th (B: "we'll put you down for the 12th"). Agreed: capacity of 60 (B: "60 seated is fine"). Open: cost of the kitchen hire ([not in thread]). Open: insurance certificate, which B has asked us to send (B: "can you send over your certificate"). Waiting: B is waiting on us for the certificate.
Reply: Thank you for confirming the 12th. I will send our certificate today. Could you let us know the cost of using the kitchen on the day, so we can set the ticket price?
The quoted evidence makes one slip easy to find. Search the thread for B's line about the 12th and read what comes next: "we can hold the 12th for you until Friday, once we have your certificate". So the date is held on a condition, not booked, and the summary's "Agreed" was too generous. A quote does not prove a claim by itself; it tells you exactly where to read the surrounding lines. The reply is sensible but needs the same correction ("Thank you for holding the 12th") before it goes. This pattern, with a quoted source for every claim, works for any summary you cannot afford to get wrong.
3. A one-page policy, drafted from how you really work
Small nonprofits often run on unwritten rules. ChatGPT can turn the rules you describe into a clean first draft, but it will fill any gap with something plausible, so the prompt marks gaps instead of filling them.
Draft a one-page volunteer expenses policy for a small nonprofit, in plain English, using only HOW WE ACTUALLY WORK below.
Sections: who can claim; what we reimburse; limits; how to claim; who approves.
Where you would need a figure or rule I have not given, write [WE DECIDE: what is needed] instead of inventing one.
End with a list headed "Assumptions to confirm".
HOW WE ACTUALLY WORK: Volunteers can claim travel by public transport and materials bought for sessions, with a receipt. We pay by bank transfer within 14 days. The Project Lead approves claims up to $50; the Treasurer approves anything above that. We do not pay for meals. Mileage: we have not decided.A typical draft, in part:
Who can claim: any volunteer who has agreed in advance to carry out a task for the project. What we reimburse: public transport fares and materials bought for sessions, with a receipt. Mileage will be reimbursed at the standard approved rate. We do not reimburse meals. How to claim: send your receipts to the Project Lead. We pay by bank transfer within 14 days. Approval: the Project Lead approves claims up to $50, and the Treasurer approves anything above that.
Assumptions to confirm: that "agreed in advance" is how you want to define eligibility; that receipts are always required.
"Mileage will be reimbursed at the standard approved rate" is the trap. You told ChatGPT you hadn't decided. It resolved the gap by pointing at a rate that your organisation has never adopted, and it did not list this in the assumptions. The right line is "[WE DECIDE: mileage rate, or no mileage]". Keep this prompt for plain-English drafting of rules you already apply. For anything that carries legal weight, such as safeguarding, data protection or employment, start from a template from your umbrella body or regulator and use ChatGPT only to tidy the wording. A policy is not adopted until the trustees approve it.
4. A funder update that shows the variance honestly
A funder reading many updates wants each one short, with the numbers set against the plan and a plain account of what went worse. Follow this prompt's figures closely, because they come back in prompt 9, rewritten for donors.
Write a 150-word progress update for a funder, from the FIGURES below only.
Say what we planned, what happened, and what changed. Show each percentage and how you worked it out. Also show spend excluding the one-off item.
Explain a variance only with the reason I have given; if I gave none, write [reason needed].
Include anything that went worse than planned. Do not describe the programme as successful, impactful or transformative.
FIGURES: We are at month 6 of a 12-month grant of $9,000. Planned: 24 weekly growing sessions for 40 families. Delivered so far: 22 sessions, 46 families. Two sessions cancelled (one for severe weather, one because the lead volunteer was unavailable). Spent so far: $6,420, including the polytunnel ($2,300), bought up front. Saturday sessions average 17 attendees; Wednesday evenings average 9.What came back:
We planned 24 weekly growing sessions for 40 families. At month 6 of 12, we have delivered 22 sessions (92%, from 22 ÷ 24) and reached 46 families (115% of the target, from 46 ÷ 40). Two sessions were cancelled, one for severe weather and one because the lead volunteer was unavailable.
We have spent $6,420 of the $9,000 grant (71%, from 6,420 ÷ 9,000). Excluding the polytunnel, which cost $2,300 and was bought up front, spend is $4,120 (46%), in line with the halfway point. Saturday sessions average 17 attendees against 9 on Wednesday evenings, so we will move the remaining Wednesday sessions to Saturdays. We expect to finish within budget.
The arithmetic is worth checking line by line, and here it holds: 22 ÷ 24 is 91.7%, 46 ÷ 40 is 115%, 6,420 ÷ 9,000 is 71.3%, and $6,420 − $2,300 = $4,120, which is 45.8% of $9,000. The finding that spend looks ahead of time but is actually on track once the one-off purchase is removed is the useful insight, and you would not want to bury it. Two sentences, though, are invented. "We will move the remaining Wednesday sessions to Saturdays" is a plan nobody made, and "we expect to finish within budget" is a forecast nobody gave. Cut both, or make them true on purpose. For longer reports built from the same figures, see turning charity data into impact reports.
5. An answer cut to a funder's word limit without losing the numbers
Online forms cut you off at a word or character count. Trimming by hand is slow because every sentence feels necessary. ChatGPT is good at cutting repetition, but it will also cut the number the funder scores on, so the prompt makes it account for every figure.
Cut the answer below to 250 words or fewer for a funder's online form. Keep every number, the name of every partner organisation, and the one sentence that states the need.
Cut first: adjectives, repeated points, and sentences that explain what the funder already knows.
After the trimmed answer, list what you cut under "Removed", and list every number in the original under "Numbers kept" with "kept" or "dropped" beside each.
ANSWER (380 words):
[paste your full answer here]The "Numbers kept" list that came back:
| Number in the original | Status |
|---|---|
| 46 families | kept |
| 22 sessions | kept |
| 35 volunteers | kept |
| 600 supporters | dropped |
| $225 per family ($9,000 ÷ 40) | dropped |
Dropping "600 supporters" is fine; it doesn't help the case. Dropping "$225 per family" is not, because cost per beneficiary is the number many funders compare across applications. Ask for it back: "restore the cost per family and cut a different sentence". The draft also stated it was 248 words, and a word processor counted 271. Models miscount words, so paste the trimmed answer into your own word count before you paste it into the form, and check the form's own counter, which may count differently. For the surrounding method, how to use AI to write grant applications goes step by step.
6. First-gift thank-you templates for three gift bands
A first gift is the moment supporters decide whether to give again, and a prompt-written thank-you works only if it sounds like a person. Ask for templates, not letters: three bands, merge fields, and impact facts you supply.
Write three thank-you email templates for first-time donors, one per gift band: under $50, $50 to $249, and $250 and over. 110 words maximum each.
Use these merge fields exactly as written: [first name], [gift amount], [campaign]. Do not guess any donor detail.
Say what a gift does using only IMPACT FACTS. Do not mention tax, receipts or deductibility. No exclamation marks. Do not call donors family or heroes.
IMPACT FACTS (ours): $25 buys seed and compost for one family bed for a season. $75 covers tools and gloves for a month of sessions. $250 covers a set of six raised beds.The first band, as it came back:
Hello [first name],
Thank you for your gift of [gift amount] to [campaign]. It is the first time you have given to us, and it matters. A gift like yours buys the seed and compost for a family growing bed for a season, which is where a lot of this work starts. A tax receipt will follow shortly. If you would like to see where the money goes, you are very welcome at any Saturday session.
With thanks, [Chair's first name]
"A tax receipt will follow shortly" breaks the brief, and it is the most important slip in the set. Whether a receipt applies depends on your organisation's status and the rules that cover it, so the sentence is only true if you have confirmed it, and a promise about timing is an extra commitment. Delete it unless your adviser has given you exact wording. A second, softer point: "a gift like yours buys the seed and compost for a family growing bed" overstates a small gift's reach if the gift is $10. Ask ChatGPT to match the impact line to the band, or write the band-one line yourself. The method behind this is covered in donor thank-you letters that still feel personal.
7. An appeal letter built from one story and one ask
An appeal that tries to say everything gives the reader nothing to do. One true story and one ask gives them something. The prompt provides both and forbids decoration.
Write a one-page appeal letter (about 280 words) from the Chair of a community garden nonprofit, using only the STORY, NEED and ASK below.
Open with the story, using the participant's own words as quoted. Then the need, then one clear ask with the amount, then a P.S. that repeats the ask and the reply date.
Do not add any detail to the story. No statistics. No guilt, and no words such as "heartbreaking". Greeting: [first name].
STORY (told to us in writing, with the participant's agreement; no name): A participant in her seventies, who joined after retiring, says: "The garden is the first place since I retired where someone expects me." She now runs the Wednesday seedling table.
NEED: A polytunnel would let us run sessions all year; we need $1,200 more to cover the build.
ASK: $40 towards the build; 30 gifts of $40 close the gap. Reply by [date] using [link].The opening of what came back:
Dear [first name],
When a woman in her seventies joined our garden after retiring, she told us something that has stayed with me: "The garden is the first place since I retired where someone expects me." Her eyes lit up as she said it. Today she runs our Wednesday seedling table, and the polytunnel that is half built behind her will let her keep it going all year.
We still need $1,200 to finish the build. That is where you come in...
Two details are invented: "her eyes lit up" and "the polytunnel that is half built". Neither is in the STORY, and the second is a factual claim about your project that a donor could check. An appeal built on an embellished story breaks trust the moment someone who was there reads it. Strike both, and keep the rule you will reuse: every sentence about a person or a place must trace to what you pasted. The story also needs the person's agreement, which is why the prompt says so. For the sum, say you mail 600 supporters and 24 give (24 ÷ 600 = 4%), at an average of $38: that is $912, which leaves $288 of the $1,200 still to find. Run that sum before the letter is printed, because it tells you whether the ask, the mailing list or the target needs to change. Those are placeholders, so use your own last appeal's response rate. For more on keeping an appeal in your own voice, see charity appeals and social posts that keep your voice.
8. A lapsed-supporter email that doesn't point at their giving
Writing to people who have gone quiet is delicate. The usual mistakes are guilt and surveillance: "we noticed you haven't given" tells the reader you are watching their record. The prompt bans both.
Write a short email (110 words maximum) to supporters who gave in the past but not in the last 13 months. Greeting: [first name].
Do not mention their gift, their giving history or the time since they last gave. Do not write "we miss you" or "we noticed".
Include one update from UPDATE, one small ask from ASK, and a plain line saying they can stop hearing from us by replying "stop".
UPDATE: This year 46 families grew vegetables with us, up from 31 last year.
ASK: Come to the open day on [date], or give $10 to cover one family's seeds.What came back:
Hello [first name],
A quick update from the garden: this year 46 families grew vegetables with us, up from 31 last year. Your support has never mattered more, and we would love you to see it for yourself. Our open day is on [date], and everyone is welcome. If you would rather help from home, $10 covers one family's seeds.
If you would prefer not to hear from us, reply "stop" and we will remove you.
"Your support has never mattered more" is a stock line of fundraising copy, and it is a claim you can't back up. Cut it. The rest follows the rules: no mention of the person's giving, one number, one invitation, one small ask, an easy way out. The "stop" line matters in its own right: it respects the reader, and it keeps your list made up of people who want to hear from you.
9. A quarterly update that donors will read to the end
This uses the same figures as prompt 4, with a different reader. A funder wants variance against a plan. A donor wants one number and a person. Compare the two outputs and you can see how much of good writing is choosing what to leave out.
Write a 180-word update email for donors, from the FIGURES and QUOTE below.
Open with the single most important number. Include the quote exactly as written. Include one thing that did not go to plan and what we are doing about it, only as stated.
No adjectives that claim impact, such as life-changing or transformative. Greeting: [first name].
FIGURES: 46 families so far (target 40); 22 of 24 sessions delivered; about 310 kg harvested so far (weighed at the shed).
QUOTE (shared with agreement, no name): "My daughter now asks to eat the beans she grew."
DID NOT GO TO PLAN: Two sessions cancelled. We are running two make-up sessions on [dates].What came back:
Hello [first name],
46 families have now grown food with us this year, against a target of 40. One parent told us: "My daughter now asks to eat the beans she grew."
So far we have run 22 of 24 planned sessions and harvested about 310 kg. Two sessions were cancelled, and we are running two make-up sessions on [dates]. Every family now eats fresh vegetables from our garden every week.
Thank you for making this possible.
"Every family now eats fresh vegetables from our garden every week" is the slip. It is a claim about 46 households' diets that no figure supports; the garden's harvest does not tell you who ate what. The rest is accurate, and the opening is right: the most important number first, then a person. Note too that the quote is attributed to "a parent" by ChatGPT, although the input said only "shared with agreement, no name". Is the speaker a parent? The quote implies it, but you didn't say, so check with the person before you publish a label.
10. A monthly-giving invitation, with the sums shown
Monthly giving is a natural next ask for people who have already given, and it reads best when each level has a concrete meaning. Ask ChatGPT to do the sums, then check them.
Write an email of 140 words maximum inviting one-off supporters to give monthly. Offer three levels: $5, $10 and $20 a month. For each, work out the yearly total and how many family beds it covers, and show the sums.
Say they can change or stop at any time by replying to this email. Do not make any claim about how donors behave, how long monthly donors stay, or how much monthly giving raises.
Also give five subject lines of 45 characters or fewer.
FACTS: Seed and compost for one family bed for a season costs $25.The working it showed, laid out:
| Monthly gift | Per year | Beds covered ($25 each) | Left over |
|---|---|---|---|
| $5 | $60 | 2 beds ($50) | $10 |
| $10 | $120 | 4 beds ($100) | $20 |
| $20 | $240 | 9 beds ($225) | $15 |
The sums are right (12 × $5 = $60, and $60 ÷ $25 gives two beds with $10 over, and so on). In the email body, ChatGPT added "most monthly donors stay with us for years". That is a claim about donor behaviour that you have not supplied, and the prompt banned it. Cut it. A sum you can use for planning, with its own assumptions stated: if 20 of your 600 supporters switched to $10 a month, that would be 20 × $10 × 12 = $2,400 a year. The 20 is a guess for illustration; your past response to similar invitations is the only honest number to plan from.
What these prompts cost a small nonprofit to run
All ten are short enough to run on a free ChatGPT account. The case for paying is data handling and limits. Business-tier plans don't train on your content by default, whereas a personal account has a model-training switch in its privacy settings that you would need to turn off. Discounts exist, and eligibility rules vary, so check the current terms for your organisation before you apply.
- OpenAI: its help article for nonprofits lists ChatGPT Business Standard seats at $8 per user a month billed annually, or $10 monthly. Premium seats are not eligible, and verification runs through Goodstack. Who qualifies and what to check first goes through it.
- Anthropic: Claude for Nonprofits lists its Team plan at $8 per user a month, with a minimum of two seats and verification through Goodstack.
- Google: Google Workspace for Nonprofits has a $0 edition. Google has said its nonprofit editions include the Gemini app and Gemini Notebook (formerly NotebookLM) at no extra cost, but the offer page doesn't list them, so check what your account shows before you plan around them.
Prices are list prices as I read them in October 2026, and offers change, so confirm on each vendor's own page. Whichever you choose, the habit that protects you is the one at the top of this tutorial: merge fields in, donor data out. After each prompt, keep a short edit log of what you changed, and send yourself a test with real merge values before any batch goes out, to check that no [bracket] has survived into the final copy.
Questions small nonprofits ask about these prompts
Which of the ten prompts should we try first?
Start with the minutes (prompt 1) or the funder update (prompt 4). Both work from notes and figures you already hold, involve no donor data and are easy to check against the source. Move to donor letters once your merge fields have been tested with a send to yourself, because a [first name] that survives into a real letter is the mistake supporters notice first.
Should we tell donors a letter was drafted with AI?
Funders' application guidance may ask about AI use, so read it for each bid. For donor letters, check whether any disclosure rule applies to your organisation. Whatever the rules say, supporters care about honesty: never use AI to invent a story or quote, and have a named person read and sign every letter that goes out.
How do we share these prompts with volunteers?
Keep them in one shared document with the never-paste table at the top, and give one person ownership of it. A ChatGPT or Claude Project can hold standing instructions such as your voice and banned words, so everyone starts from the same settings. Review who has access whenever a volunteer leaves.
Further reads
- Using AI to Spot Donors Who Are About to Lapse — Find which supporters are drifting before you write to them.
- ChatGPT for Nonprofits: Volunteer Recruitment Posts and Rota Messages — Volunteer recruitment posts and rota messages, drafted the same way.
- Claude AI for Charities: Five Admin Jobs Set Up With Projects — Setting up Claude Projects for five recurring admin jobs.
- Fireflies for Nonprofits: Can It Write Board Minutes Safely? — Whether a note-taking tool can write board minutes safely.
- How to Write an AI Policy for a Small Charity — A short AI policy for staff and volunteers to sign.
- AI Tool Discounts for Charities and Non-Profits: What to Claim — Which AI tools offer nonprofit pricing and what to claim.
- Charity AI Risks: Data Protection, Deepfakes, and Donor Trust — Data, deepfake and donor-trust risks to know about.
- Charity Complaints Policy Template: Draft One With AI and Check It — Drafting a complaints policy with AI, then checking it.
- HeyGen for Nonprofits: Donor Update Videos and Disclosure Rules — How a small nonprofit can make donor update videos in HeyGen: avatar consent, ready-to-copy disclosure wording, translation, and when to film a real face.
- AI Implementation for Nonprofits: A Phased Plan for a Small Team — A 20-week pilot-to-rollout timeline for a small nonprofit team: inventory, pilot, policy, training, review and trustee sign-off.
- ChatGPT Prompts for Churches: 10 Admin Jobs, No Sermons — Ten admin-only ChatGPT prompts for a church office, each with a worked example, and a firm line: no sermons, no member data.
- Introduction to AI Skills for Nonprofits: What Staff Learn First — A manager's order of teaching for a small charity: donor-data rules, prompting basics, a review habit, and a first-week practice plan for five staff.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's help article on OpenAI for Nonprofits; Anthropic's Claude for Nonprofits page; Google Workspace for Nonprofits offer page (all re-checked 28 October 2026). All organisation details, figures and quotes in the examples are illustrative.