Each prompt below uses the skeleton OpenAI's prompt engineering guide recommends: **Role** (register, expertise), **Context** (real program data, real numbers, redacted where required), **Constraints** (word count, reading level, banned phrases, dignity rules), **Format** (markdown, plain text, table). Every sample output below was tested against a redacted version of a real human-services organization's program data.
### Prompt 1 — Grant LOI from a program brief
```
You are a senior grant writer who has written 200+ funded letters of inquiry to family foundations in the $500K-$10M asset range.
Program brief (paste 1-2 pages, no beneficiary names):
[brief]
Funder context:
- Name: [foundation].
- Priority area we map to: [one phrase].
- Typical first-grant size from public 990: $[X].
- One recent grantee whose work resembles ours: [name + one sentence].
Write a 2-page letter of inquiry in this order:
1) Greeting + 1-sentence reason for writing (no flattery).
2) The problem in our service area in 3 sentences, with one publicly cited statistic.
3) Our program in 2 paragraphs — what we do, what we have already done.
4) The ask: $[amount] over [period], with a one-line use-of-funds breakdown.
5) Why this funder, in 2 sentences referencing their stated priority — not their net assets.
6) One-paragraph organizational credibility (years operating, budget, audited).
7) Close + named contact.
Reading level: educated general audience, not academic. Banned phrases: "unprecedented", "truly transformative", "world-class", "underserved populations" (use the specific community instead), "at-risk youth".
```
**Why it works.** "$500K-$10M asset range" rules out the wrong register — a $50M foundation reads differently than a regional family fund. The banned-phrase list does more than the role line; "underserved populations" is the single phrase most program officers cite as a tell that the writer does not know the community.
**Sample output (redacted).** Opening: "We are writing to introduce our literacy program in [county], which served 312 families last year and which we believe maps directly to your stated priority of early-childhood education in rural [state]." Ask paragraph: "We are requesting $85,000 over 18 months to extend the program from 4 to 7 sites, with 72% to teacher stipends, 18% to materials, and 10% to evaluation."
**Guardrail.** Do not paste beneficiary names, case notes, or unredacted donor pipeline information. The funder's name and the program brief are safe — both are or will be public.
### Prompt 2 — Donor thank-you, tone variant by gift size
```
You are a development director writing donor thank-yous. Voice: warm, specific, dignity-preserving, never gushing.
Donor (use ONLY initials and tier — strip name, address, email before pasting):
- Initials: [A.B.]
- Gift tier: [select one — $25, $100, $500, $1,000, $5,000, $25,000].
- Repeat donor (Y/N) + years giving: [Y, 4 years].
- Specific program designation: [program] OR "unrestricted".
- One thing the database tells you about why they gave (volunteer history, board member's friend, attended event): [one phrase].
Produce three versions:
A) Standard letter (under 250 words).
B) Hand-written-style note (under 80 words) for a board member to copy.
C) Email version (subject under 45 chars + under 150 words body).
Rules:
- Mention the program by name and one specific use of funds.
- Repeat donors: acknowledge the year count in one sentence, no boilerplate.
- Major gifts ($5,000+): reference the specific program use and offer a named follow-up call within 14 days.
- Never describe a beneficiary by name, demographic, or case detail.
- Banned phrases: "Words cannot express", "Your generosity", "Thanks to donors like you", "On behalf of", "Truly grateful".
Output each version with a one-line note on which tier of donor it fits.
```
**Why it works.** Gift-size segmentation is the practice Bloomerang's donor retention benchmarks and Classy's State of Modern Philanthropy report most often correlates with second-gift conversion. Generic thank-yous read like form letters; specifically naming program use signals the gift reached its destination.
**Sample output (redacted).** $1,000 standard: "A.B., your $1,000 designated to the literacy program covers reading kits for 14 families. You have given for four years now — that consistency is what lets us forecast site staffing in February rather than scrambling in June."
**Guardrail.** Strip donor PII before pasting. Initials and tier are enough for tone; the database holds the real name when the letter is merged.
### Prompt 3 — Board-report narrative from program data
```
You are a chief of staff to a nonprofit executive director. You write board-meeting narratives that surface decisions, not activity logs.
Quarter program data (paste table — aggregate numbers only, no individual case data):
- Households served: [Q-1: X, Q: Y]
- Program completion rate: [Q-1: X%, Q: Y%]
- Average cost per outcome: [Q-1: $X, Q: $Y]
- Staff turnover in program: [number this quarter, number last quarter]
- Three open decisions in front of the board: [list]
Write a 500-word board narrative in this order:
1) Headline finding (one sentence, plain English).
2) What changed quarter-over-quarter, with the actual numbers and the direction.
3) The one metric trending wrong, with the operating reason (not blame).
4) The one metric better than expected, with what to learn from it.
5) The 3 open decisions framed as: question + 2 options + recommended option + the risk if we choose wrong.
6) One sentence on what staff will track next quarter.
Reading level: a board member who is brilliant in their field but not a nonprofit-operator.
Banned phrases: "As you can see", "We are pleased to report", "Moving forward", "Going forward".
```
**Why it works.** Board reports decay into activity recaps. Forcing "the one metric trending wrong" plus a recommended option converts the meeting from update to decision. The reading-level cue matters — board members are senior leaders in their own fields, not your field.
**Sample output (redacted).** Headline: "Program completion rose to 71% but the cost per outcome rose with it — we need a board decision on whether to keep the higher-touch model or revert."
**Guardrail.** Aggregate numbers only. Never paste individual case files into the prompt.
### Prompt 4 — Year-end appeal copy
```
You are a year-end appeal copywriter. Voice: direct, story-led, dignity-forward, no poverty porn.
Context:
- Organization: [one-sentence mission].
- The one specific program outcome from the past year, told as a number plus a one-line story (composite, no real names): [example].
- The dollar amount we need by Dec 31, and what it specifically buys: [example: "$140,000 keeps the after-school site open through May"].
- The match, if any (matching donor + cap): [optional].
Produce a 5-piece appeal:
1) Direct-mail letter (650 words, signed by ED).
2) Email 1 — early November, soft introduction, 220 words.
3) Email 2 — Giving Tuesday, urgency + match if applicable, 180 words.
4) Email 3 — Dec 30, last-chance, 140 words.
5) Donate-page hero (headline 12 words + subhead 25 words + 3 bullet outcomes).
Rules:
- Open every piece with the specific outcome — never with "As 2026 comes to a close".
- Composite stories must be flagged with one line at the bottom: "Names and identifying details changed to protect privacy."
- Banned phrases: "In these uncertain times", "Now more than ever", "Help us continue", "Your support means everything".
```
**Why it works.** Specificity beats sentiment in Classy's State of Modern Philanthropy benchmarks — appeals naming a specific use of funds out-convert generic ones. The composite-story disclosure line is the dignity guardrail.
**Sample output (redacted).** Email 2 subject: "Today, $1 becomes $2 for [program]." Email opens: "Last Tuesday, the after-school site served 47 kids — for $4.20 per kid per afternoon. Today, every dollar you give doubles."
**Guardrail.** Composite, anonymized stories only. Never use a real beneficiary's name or photo without a signed, current release on file.
### Prompt 5 — Volunteer onboarding document
```
You are an HR-light operations lead writing a one-page volunteer onboarding doc for a [type] nonprofit.
Role: [one-sentence description, e.g., "after-school tutor, 2 hours/week, in-person"].
Program context: [one paragraph — public-facing language only].
What the volunteer must know on day 1: [bulleted list, e.g., where to park, who to find, confidentiality, mandatory-reporter status if applicable, what to do if a child discloses harm].
Write the doc with these sections:
1) What you'll do this shift (3 bullets).
2) Who you'll meet (named role, not named person — fill in at print).
3) What we'll never ask you to do (3 bullets — protects volunteer + org).
4) Confidentiality, in plain English (4 sentences).
5) Mandatory-reporter notice if the role qualifies (use state-specific wording — flag for [STATE] insertion, do not invent the statute).
6) The 3 things to do before leaving each shift.
7) Who to call after hours (named role, not named person).
Reading level: high-school graduate, no jargon. Under 600 words.
```
**Why it works.** Volunteer onboarding fails when it confuses the volunteer about boundaries. Naming "what we'll never ask you to do" pre-empts the situations that cause volunteers to ghost — and protects the organization from role drift.
**Sample output (redacted).** "What we'll never ask you to do: drive a participant in your personal vehicle, photograph a participant, or communicate with a participant outside the program platform."
**Guardrail.** Do not invent state-specific mandatory-reporter statute text — flag it for your operations director to insert from the verified source. The Child Welfare Information Gateway maintains state-by-state references.
### Prompt 6 — Theory of change articulator
```
You are an evaluation consultant helping a nonprofit write a one-page theory of change.
Program inputs (what we put in): [list — staff hours, dollars, materials, partnerships].
Activities (what we do): [list — sessions, home visits, training].
Outputs (what we count): [list — sessions delivered, families enrolled].
Outcomes (what changes for participants in 6-18 months): [list].
Impact (what changes in the community in 3-7 years): [one sentence].
The single assumption that, if wrong, breaks the whole logic chain: [one sentence].
Produce:
1) Narrative paragraph (under 180 words) that a board member can read aloud.
2) Logic-model table: Inputs | Activities | Outputs | Outcomes | Impact — one row per program track.
3) The three measurable indicators we would track to know the outcomes are happening — with the data source for each.
4) The two indicators that, if they moved the wrong way, would tell us our assumption was wrong.
5) One paragraph: what we do NOT claim our program causes (the negative-space discipline).
```
**Why it works.** Most theory-of-change docs over-claim impact. The "what we do NOT claim" paragraph is the credibility move — funders read it as a signal of evaluation maturity. The named breaking-assumption is what separates a theory of change from a wishlist.
**Sample output (redacted).** Negative-space paragraph: "We do not claim our after-school program raises district graduation rates on its own; we claim it contributes to literacy gains among enrolled participants, measured by the [assessment] administered at intake and at 12 months."
### Prompt 7 — Impact-report sidebar generator
```
You are a nonprofit communications designer writing the sidebars and callouts for a 12-page annual impact report.
Main copy (paste): [paragraph of program narrative].
Key numbers from the year (aggregate only): [list].
One composite participant story, anonymized: [paragraph].
Generate 6 sidebar components:
1) Three pull-quotes (12-22 words each) from the main copy.
2) One big-number callout (the number + 8-word context line).
3) One "by the numbers" stack: 4 stats with one-line each.
4) One quote callout from the composite story (under 25 words).
5) One "what this made possible" 3-bullet box.
6) One donor-recognition footer line for the page (no names — just "made possible by 1,247 donors and 3 foundation partners").
Do NOT generate new statistics. Pull only from the numbers I provided. If a sidebar would require an invented number, write "INSERT FROM DATA" instead.
```
**Why it works.** The "INSERT FROM DATA" instruction is the hallucination guardrail. Impact reports are the easiest place for a well-meaning AI draft to invent a number that survives layout review.
**Sample output (redacted).** Big-number callout: "4,200 — households served across 7 counties, up from 3,100 last year."
### Prompt 8 — Awareness-day social content calendar
```
You are a nonprofit social media strategist planning a 14-day campaign around [Awareness Day, e.g., World Refugee Day on June 20].
Organization: [one-sentence mission].
The specific connection between our program and the awareness day: [one sentence].
Primary channel: [Instagram / LinkedIn / X / Facebook].
Goal: [pick one — newsletter signups, donations, volunteer interest, policy advocacy].
Tone: [pick one — somber, energizing, educational, urgent].
What we will not do: misappropriate the day's frame, use stock photos of people not connected to our program, or center the organization over the community.
Produce a 14-day calendar table: Date | Channel | Format (post, reel, carousel, story) | Hook (under 12 words) | Body (under 60 words) | CTA | Asset note (what photo or graphic is needed — and how it must be sourced).
The 14 days must include:
- At least 3 days that link directly to our program's published outcomes.
- At least 1 day featuring a partner organization (not us).
- At least 1 day citing a credible external source ([UNHCR / Pew / Census / your sector body]).
- At most 3 days with a donation ask.
- 0 days using a participant's name, face, or identifying detail without a signed current release.
```
**Why it works.** Awareness-day calendars fail by being self-promotional. The constraint floor (at most 3 ask days; at least 1 partner day) forces a calendar that doesn't read like a fundraiser. The asset-sourcing rule is the dignity guardrail.
**Sample output (redacted).** Day 4, LinkedIn carousel hook: "Resettlement starts with a lease. Here is what that took in 2025." CTA: "Read the report — link in comments."
### Prompt 9 — 990 plain-English summary for donors
```
You are a nonprofit finance translator. You write 1-page donor-facing summaries of Form 990 that are accurate and readable.
Paste the relevant 990 figures (these are PUBLIC — already on ProPublica Nonprofit Explorer, so safe to paste):
- Total revenue: $[X]
- Total expenses: $[X], split: program $[X], management $[X], fundraising $[X]
- Net assets, beginning and end of year: $[X], $[X]
- Largest single grant or program-service category: [name + $]
- Executive compensation (Schedule J line — public): $[X]
- Independent auditor: [Y/N]
- One operational note for context (e.g., "capital campaign year"): [optional]
Write the 1-pager:
1) Top-line in 2 sentences — revenue, expense, and whether the year ended in surplus or deficit (and why).
2) Program ratio: $X of every $1 went to programs. (Calculate from the numbers — do NOT make up the ratio.)
3) Management and fundraising ratio explained in 2 sentences — no defensiveness, no apology.
4) The largest program, what it is, what it costs, what it delivered (use my one-line outcome).
5) Executive compensation in 1 sentence — the number, and the context (board-approved, benchmarked).
6) The plain-English answer to "is this organization financially healthy?" — based on net asset trend and reserves.
7) One paragraph: what we would do differently with $100,000 more next year.
Reading level: a donor who has never read a 990. No accounting jargon. If a number would require a calculation you can't do from what I gave you, write "INSERT — needs CFO" instead.
```
**Why it works.** 990 anxiety is universal among small-shop directors. The "INSERT — needs CFO" instruction is the guardrail against invented ratios. Naming overhead without apology is the move that builds donor trust — the Overhead Myth letter signed by Candid, BBB Wise Giving Alliance, and Charity Navigator is the citation worth knowing.
**Sample output (redacted).** "Of every $1, $0.79 went directly to programs, $0.13 to fundraising, and $0.08 to management — in line with peer human-services organizations."
**Guardrail.** 990 figures are public. Donor names and pipeline notes are not — never blend the two in one prompt.
### Prompt 10 — Restricted-vs-unrestricted explainer for trustees
```
You are a nonprofit CFO writing a one-page memo for a new board member who has joined from the for-profit world and is asking, "why can't we just spend the reserves?"
Context:
- Our unrestricted net assets: $[X]
- Our temporarily restricted net assets and what they're restricted to: $[X], [list of restrictions]
- Our permanently restricted (endowment) net assets, if any: $[X]
- Reserves policy adopted by the board: [e.g., 6 months operating]
Write the memo:
1) The 3-sentence why: why the IRS, GAAP, and donor intent each create restrictions.
2) A plain-English analogy that does NOT condescend.
3) Our specific picture: which buckets we have and what each can and cannot fund.
4) The 4 questions a trustee should ask before voting on a withdrawal from any bucket.
5) Where the line between "prudent reserves" and "hoarding" actually is (cite the Nonprofit Finance Fund's published benchmarks).
6) What changed in FASB ASU 2016-14 net-asset reporting and what it means for our financial statements.
Reading level: a sharp executive new to the sector. No condescension, no jargon, no "as you know".
```
**Why it works.** This is the single most common board onboarding question — and the one most likely to cause an expensive misunderstanding. Naming FASB ASU 2016-14 (summary at FASB) is the credibility move. The Nonprofit Finance Fund publishes the operating-reserves benchmark.
**Sample output (redacted).** "Of our $4.2M net assets, $1.1M is unrestricted reserves (six months operating, per board policy), $2.3M is temporarily restricted to specific programs and time periods, and $800K is the donor-restricted endowment. The reserves are not 'spare' — they are what keeps payroll moving in a delayed-reimbursement month."
### Prompt 11 — EOY tax-receipt template
```
You are a nonprofit finance administrator writing the year-end tax-receipt template for the mail-merge.
Context:
- Tax year: 2026.
- Organization name + EIN + address (public): [fill at merge].
- Gift type: [select — cash, securities, in-kind goods, in-kind services (NOT deductible by donor), vehicle].
- Quid-pro-quo gift (donor received goods/services in return — gala dinner, auction item)?: [Y/N + fair-market-value if Y].
- Acknowledgement deadline: January 31 of following year.
Write the template with merge-field brackets:
1) Greeting with [DONOR_NAME] field.
2) Confirmation of receipt: gift amount + date received + payment method (no card numbers).
3) IRS-required language for gifts $250+ — the "no goods or services were provided in exchange" line, OR the fair-market-value statement for quid-pro-quo gifts (cite IRS Pub 1771 plain-English).
4) For in-kind goods: state that the donor is responsible for valuation; do NOT state a value.
5) For in-kind services: state that donated services are NOT tax-deductible to the donor.
6) EIN + organization 501(c)(3) status line.
7) The one warm sentence connecting the gift to a specific program use — pulled from a merge field, not invented.
Produce 4 variants:
A) Single cash gift, no goods/services exchanged.
B) Recurring monthly donor (annual summary, no goods/services exchanged).
C) Gala / event ticket with fair-market-value deduction (quid pro quo).
D) In-kind goods gift.
Do NOT generate IRS legal text I did not provide. Flag any IRS-citation line as "VERIFY — IRS Pub 1771" so the finance director confirms before send.
```
**Why it works.** Tax-receipt language is one of the few nonprofit communications with real legal consequence — wrong language can disallow the donor's deduction. The "VERIFY — IRS Pub 1771" guardrail forces a human check. IRS Publication 1771 is the canonical reference for charitable contribution acknowledgement requirements.
**Sample output (redacted).** Gala variant body: "Thank you for your $500 gala ticket purchase. The fair market value of goods and services received (dinner and program) was $85. The deductible portion of your contribution for federal income tax purposes is $415."
**Guardrail.** Finance director or CPA reviews the IRS-citation lines before send. Never invent statute text.
### Prompt 12 — Urgent campaign brief
```
You are an emergency-response campaign strategist for a nonprofit.
The situation in 3 sentences (no beneficiary identifiers): [what happened, who is affected at our org, what window we have].
What we need: [dollars / volunteers / advocacy actions / shelter / supplies].
The deadline: [date].
What we will NOT ask for: [anything outside what we can responsibly deploy].
Produce a 1-page campaign brief, internal-facing, in this order:
1) The 1-sentence ask, the deadline, the channel mix.
2) The 3-message ladder: opening message (within 24 hours), midpoint update (day 3 or halfway), close (final 24 hours).
3) Audience segments and which channel goes to which.
4) The 4 honest-update milestones — what we'll tell donors at 25%, 50%, 75%, 100% of goal.
5) The dignity rules for any messaging that touches affected community members.
6) The 3 things we will do in the 30 days AFTER the campaign closes, by name, to honor what we asked for.
7) The 2 questions a journalist might ask that we should pre-draft answers to.
Reading level: an exhausted ED at 11 pm.
Banned phrases: "Tragic", "Devastating", "Unprecedented", "In light of recent events".
```
**Why it works.** Urgent campaigns fail two ways — by going silent at 60% to goal, or by extracting more than the org can responsibly deploy. The post-campaign 30-day commitments are the trust-keeper. The journalist pre-draft is the operational save when the campaign goes well enough to attract press.
**Sample output (redacted).** Honest-update milestone at 75%: "We are $18K from goal with 36 hours left. If we close the gap, here is exactly what the next 30 days look like for the 47 families on the waitlist."
**Guardrail.** Affected community members are not props. Names, photos, and identifying details require a signed, current release — and a frank conversation about what the person actually wants shared.