What nonprofit teams use it for
Each job below starts from material you already have and says where a person takes over. None of it needs a technical background.
Grant narratives from program data
Paste the funder’s questions with their word limits, your program data as counts and dates, and past narratives the funder has already accepted. Ask for each answer within its limit, using only those sources, with [NEED DATA] wherever the data does not cover a question. The gaps are the useful part: they tell you what to pull from the database before the deadline, not after it.
Draft answers from approved contentCheck a draft’s claims against your sourcesSummarize a long documentPrompt Tester
Donor thank-you letters and appeals
A prompt makes it practical to send a thank-you within days that names the gift and says what it paid for, instead of a polished letter a month later. Give it the gift record and one true detail from the program, and ask for a letter that sounds like the person who runs it. Appeals work the same way: a real story told with permission, and a specific ask. The receipt paragraph gets checked against the gift record before anything is sent.
Edit a draft without losing your voiceProofread for errors onlyPrompt GraderMy Prompts
Impact reports
An impact report is a set of claims, and each one should trace to a row in your data or a sentence in an evaluation. Draft the narrative from the numbers, then run the claims check against the same data, so “more families than ever” stays only if last year’s figure says so. Participant feedback goes through a themes prompt that counts and quotes.
Check claims against your dataFind themes in survey responsesChoose the right chartTurn an interview into a story
Volunteer communications
Shift reminders, orientation notes and the message that a site is closed on Saturday are short, frequent and easy to check. Turn the coordinator’s notes into a newsletter, and what the Saturday lead knows into a written procedure the next lead can follow.
Turn rough notes into a newsletterTurn a process into a procedurePlan a training session for adults
Board summaries
Board members read the packet the night before, if at all. Put the finance report, the program update and the open decisions on one page, with the questions the board must answer first. Budget-versus-actual commentary explains which lines moved and why, from the treasurer’s notes, and marks the lines with no stated cause instead of inventing one.
Summarize for a busy readerWrite variance commentaryWrite a decision memo with optionsTurn meeting notes into action items
Plain language and translation
Eligibility rules, consent forms and program notices are often written for funders, not for the people who use the program. Rewrite them in plain language with the key terms kept, then translate from the approved English version into the languages your community reads. Ask a fluent staff member or volunteer to compare two or three models blind in Model Compare before you settle on one.
Rewrite expert text in plain languageTranslate for local readersModel Compare
A worked example: a year-end grant report
Take a hypothetical food pantry with three staff and about 60 volunteers. Its largest funder, a community foundation, wants a year-end report: six questions of up to 250 words each, plus outcome figures. The data lives in a spreadsheet of monthly distribution counts, and last year’s report sits in a shared folder.
- 1Export counts, not peopleMonthly totals by program and site, with no names, addresses or case notes.
- 2Draft from your own materialThe funder’s questions, the counts and last year’s accepted report, with [NEED DATA] for gaps.
- 3Grade the promptThe free estimate checks the limits and what to do when the data is silent. No key needed.
- 4Test it on last yearRun last year’s questions and data with a free key, and compare the drafts with the report you sent.
- 5Check every claimEach sentence checked against the data, with anything unsupported flagged for a person.
- 6Save it for next yearKeep the prompt with its version history, so next year starts from a tested draft.
The test in step four is simple. Each test case is one of last year’s questions with that year’s data, and a Length check holds each answer to its word limit. On one case, delete a month of data and put [NEED DATA] in its expected column; a Contains text check set to {{expected}} then fails any draft that fills the gap with a guess. An AI judge check asks whether every number in the answer appears in the program data. Read its reasons, because a judge can miss a number, which is why step five exists.
Running the test needs an API key. Google AI Studio, OpenRouter and Groq give free ones, explained in the API key guide. A free tier suits monthly counts with no names in them, though you should expect a rate limit now and then. It does not suit case notes or donor records.
What it costs
The chart prices a hypothetical five-person team making about 30 drafts a working day between them (letters, grant answers, newsletters, board summaries and translations) over 20 working days. Each call sends about 3,000 tokens of notes, data and a fixed house-style block, and gets about 800 back. The house-style block is about 30% of the input and can be read from cache. That is 600 calls, about 1.8 million input tokens and 480,000 output tokens a month.
A hypothetical team: 30 drafts a working day for 20 days. The Gemini and Groq models shown also have free tiers with daily limits. Monthly cost of 600 calls with about 3,000 tokens in and 800 out each, 30% of the input read from cache. List prices checked October 20, 2026; your bill may differ.
At this volume even the most expensive model in the chart stays under $15 for the month, and the cheapest is under a dollar, so the API bill matters less than the time it takes to review each draft. Choose the model on what your test shows, set a spending limit on the key, and try your own numbers in the cost calculator.
If your team would rather use a chat app than API keys, three vendors run nonprofit programs. Google says Google Workspace for Nonprofits now includes premium AI features, such as the Gemini app, at no cost. Anthropic’s Claude for Nonprofits offers eligible organizations up to 75% off its Team and Enterprise plans. OpenAI for Nonprofits discounts ChatGPT Business and Enterprise seats. Eligibility and prices change, so check each vendor’s page before you apply.
A prompt to start from
A thank-you letter that carries its own receipt paragraph. The double-brace words are blanks you fill in from the gift record.
You draft thank-you letters for {{org_name}}, a nonprofit that
{{mission_in_one_line}}.
Gift details (the only facts you may use):
- Donor name as it should appear: {{donor_name}}
- Date received: {{gift_date}}
- Cash amount, or a description of donated items: {{gift}}
- What we gave in return, with our good-faith estimate of its
value, or "none": {{benefits}}
- Program the gift supports: {{program}}
- One true detail from that program, from our notes:
{{program_note}}
Write a letter of 150 to 200 words that:
1. Thanks the donor by name in the first sentence.
2. Says what the gift supports, using only the program note.
Add no numbers, outcomes or stories that are not in it.
3. Ends with a short receipt paragraph that states our name,
the date, and the cash amount or a description of the
donated items (never their value), and says either that no
goods or services were provided or what was provided and
its estimated value.
4. If the donor paid more than $75 and received something,
adds that only the amount above that value is deductible.
5. Sounds like the person who runs the program, not a brochure.
If any gift detail is missing, write [MISSING: field name] in
its place instead of guessing.The receipt rules come from the IRS pages linked under Guardrails. The person who keeps the gift records checks every amount and date before the letter goes out.
Guardrails
Practical guidance, not legal or tax advice. Your funders’ agreements, your own privacy policy and your state’s rules add to what is here.
Donor and client records stay out of free tools
Free tiers are fine for drafts built from counts and public information. They are not the place for giving histories, case notes or anything that identifies a client. Google’s Gemini API terms, for example, say that on unpaid services Google uses what you submit to improve its products, that human reviewers may read it, and that you should not submit sensitive, confidential or personal information (the EEA, Switzerland and the UK are treated differently). A thank-you letter needs a first name and a gift amount, not a donor file. The risk checklist covers the rest.
Some programs have confidentiality rules in law
Grantees under the Violence Against Women Act may not disclose personally identifying information about the people they serve without the person’s informed, written, reasonably time-limited consent (34 U.S.C. 12291). If you run health services, the healthcare administration page covers HIPAA. When in doubt, a case note does not go into an AI tool at all.
Report what happened, not what sounds good
Every outcome figure in a grant or impact report points to a row in your data. If the model wrote a number nobody can trace, it comes out, however right it sounds; a confident invented figure is a hallucination, and a funder may hold you to it next year. For federal awards the stakes are written down: the official who signs a financial report certifies that it is true, complete and accurate, and that false information can bring criminal, civil or administrative penalties (2 CFR 200.415).
Stories and photos need permission
Use a participant’s story, name or photo only with their permission, and change details when they ask. Do not present an AI-generated image as a real participant, or a composite story as one person’s.
Receipts follow IRS rules
For a single gift of $250 or more, the IRS lists what the written acknowledgment must include: your organization’s name, the cash amount or a description (not the value) of donated items, and either a statement that no goods or services were provided or a description and good-faith estimate of what was. For a payment over $75 that bought a benefit, a disclosure statement must say that only the amount above the benefit’s value is deductible.
Translations get a fluent reader
Have someone fluent read every translated notice before it goes out. If your organization receives federal financial assistance for a health program, the Section 1557 rule requires a qualified human translator to review machine translation of documents critical to a person’s rights, benefits or access (45 CFR 92.201).
Where to start
Start with volunteer emails and plain-language rewrites: short, frequent and easy to check against what you meant to say. Board summaries and newsletters come next, drafted from documents you already trust. Grant narratives and impact reports follow once you have a tested prompt and the claims check in place, and donor receipts come last, with the gift record beside every letter.
Key takeaways
- Draft from your own records: program data as counts, past accepted narratives and real notes from the field.
- Every outcome number traces to a row in your data; mark gaps [NEED DATA] instead of filling them.
- Keep giving histories, case notes and anything that identifies a client out of free tiers.
- A thank-you letter can carry the IRS acknowledgment, but the facts come from the gift record and a person checks them.
- Start free: free API keys for testing, and nonprofit programs from Google, Anthropic and OpenAI for chat apps.
Related templates, guides and posts
Questions
Can a small nonprofit use AI without paying for it?
Yes, for most of the work on this page. The templates, the Grader’s free estimate, the diff, the library and the cost calculator need no key. Google AI Studio, OpenRouter and Groq give free API keys for testing, and Google, Anthropic and OpenAI run nonprofit programs for their chat apps. Free tiers may use what you send, so keep donor and client records off them.
Is it safe to paste donor information into an AI tool?
Only into a tool whose terms you have read and that covers personal data, and only what the draft needs. A thank-you letter needs a first name, a gift amount and a date, not a giving history. Case notes and anything that identifies a client never go into a free tier.
Can AI write our grant proposal?
It can draft one from your material: the funder’s questions, your program data and narratives you have written before. It should not supply outcomes, statistics or stories. Ask it to mark every gap, check every claim against your data, and have the person who knows the program read the final version.