Vault / course/courses/ai-fluency-for-nonprofits.md

updated 2026-06-25

Course: AI Fluency for Nonprofits

Mirrors: Anthropic Academy — AI Fluency for Nonprofits · https://anthropic.skilljar.com/ Audience: Staff, volunteers, and leaders at mission-driven organizations. No technical background needed. · Time: ~60 min + project Prereqs: claude-101 and a claude.ai account (setup-checklist). · Backing notes: ai-fluency-framework-foundations, ai-capabilities-and-limitations Project: p03-ai-fluency-delegation-audit

This course applies the 4D framework — Delegation, Description, Discernment, Diligence (an open educational resource by Rick Dakan & Joseph Feller) — to the reality of nonprofit work: high mission, low budget, lean staff, sensitive data. The promise here is concrete: AI lets a small team do the work of a larger one. The catch is that "lean and fast" must never mean "careless with the people you serve."

Learning objectives

After this course you can:

Apply the four Ds to nonprofit operations under real resource constraints.
Identify the highest-value AI use cases for a mission-driven org.
Use AI to accelerate grant writing, donor communications, and program operations.
Discern AI output where stakes (funder trust, beneficiary impact) are high.
Protect beneficiary and donor data with the Diligence the mission demands.

Module 1 — The 4D framework, under constraints

🎞 Frame 1 · Delegation = capacity, the scarcest nonprofit resource · ⏱ ~3 min

🎬 Scene — A two-person program team facing a grant deadline, a donor newsletter, and a board report due the same week. They delegate the first drafts to AI and reclaim their evenings.

🧠 Concept — Delegation for nonprofits is fundamentally about capacity. Staff time is the binding constraint. Delegate the time-sink drafts and admin; keep the relationships, the mission judgment, and anything touching the people you serve.

🖼 On screen

Delegate to AI	Keep human
First drafts of grants, reports, posts	Relationships with funders & community
Summarizing long documents & research	Decisions about beneficiaries' lives
Reformatting, translating, data cleanup	Mission strategy & values calls
Brainstorming campaign ideas	The final word that carries your name

✅ Checkpoint — Name the single most time-consuming writing task at your org that's safe to delegate a first draft of.

🎞 Frame 2 · Description: encode your mission and voice · ⏱ ~3 min

🎬 Scene — A generic AI appeal letter sounds like every other charity. A described prompt — mission, tone, the specific program, the audience — sounds like you.

🧠 Concept — Description means giving AI your organizational context: mission, voice, the program, the audience, and the format. Nonprofits live or die on authentic voice; the more of your identity you describe, the less generic the output. (Deeper: prompt-engineering-basics.)

🖼 On screen

❌ "Write a donor thank-you email."
✅ "Write a 150-word thank-you to a recurring donor who gave $50
    to our after-school literacy program. Warm, specific, no
    jargon. Reference that their gift funds books, not overhead.
    Match our hopeful, plain-spoken voice."

✅ Checkpoint — List the three facts about your org AI most needs to write in your voice.

🎞 Frame 3 · Discernment & Diligence: stewardship is the mission · ⏱ ~3 min

🎬 Scene — An AI grant draft cites an impressive but wrong outcome statistic. Catching it before it reaches the funder protects years of trust.

🧠 Concept — Discernment is judging AI output before it represents your org; Diligence is the responsibility around it — verifying claims, protecting data, and being honest with funders and community. For nonprofits these Ds are stewardship: a fabricated stat to a funder or a leaked beneficiary record is an existential risk. (Deeper: ai-capabilities-and-limitations.)

🖼 On screen

Before it leaves the building:
  Discernment → accurate? on-mission? truly in our voice?
  Diligence   → stats verified? no private data exposed? honest?

⚠️ Gotcha — A made-up impact number in a grant is not a typo — it's a credibility (and possibly legal) failure. Verify every figure.

✅ Checkpoint — Why are the stakes of Discernment higher in a grant than in an internal memo?

Module 2 — High-value use cases

🎞 Frame 4 · Grant writing, accelerated · ⏱ ~3 min

🎬 Scene — A program lead pastes the funder's guidelines and the org's past report; AI drafts a tailored proposal that the lead then sharpens with real numbers.

🧠 Concept — Grant writing is the highest-leverage AI use for many nonprofits: it's time-intensive, formulaic in structure, and repeated across funders. Delegate the structure and first draft; you supply the verified outcomes, the real story, and the final voice.

🖼 On screen

"Here are the funder's priorities and our last annual report.
 Draft a proposal that maps OUR programs to THEIR priorities.
 Leave [bracketed placeholders] for every statistic so I can
 fill in verified numbers. Don't invent any data."

⚠️ Gotcha — Tell AI to leave placeholders for numbers. Never let it generate the statistics — you fill those from your records.

✅ Checkpoint — Why instruct AI to leave bracketed placeholders for every stat in a grant?

🎞 Frame 5 · Donor & community communications · ⏱ ~3 min

🎬 Scene — One impact update becomes a donor email, a board summary, a social post, and a printed flyer — same facts, four voices, minutes of work.

🧠 Concept — AI excels at repurposing one set of verified facts across channels and audiences. For a lean comms team, this turns a single update into a full campaign. You verify the facts once; AI handles the formats.

🖼 On screen

"From this verified impact update, produce: (1) a 120-word donor
 email, (2) a 3-bullet board summary, (3) a short social post,
 (4) a one-paragraph flyer blurb. Keep every fact identical."

✅ Checkpoint — What's the one thing you must do before repurposing an update across channels? (Verify the facts once.)

🎞 Frame 6 · Program ops & research · ⏱ ~3 min

🎬 Scene — A coordinator uses AI to summarize 30 pages of policy, draft a volunteer onboarding guide, and scan a funder landscape report.

🧠 Concept — Behind the mission is a mountain of operational text — policies, reports, onboarding docs, research. AI clears that backlog: summarizing, drafting SOPs, comparing options, and surfacing themes — freeing staff for the work only humans can do.

🖼 On screen

Ops wins:
  • Summarize long policy/research into a 1-page brief
  • Draft SOPs, onboarding guides, FAQs from your notes
  • Compare grant/vendor options in a decision table
  • Theme open-ended survey responses (de-identified)

✅ Checkpoint — Name one operational document at your org that AI could draft from your existing notes.

Module 3 — Data sensitivity, ethics & doing more with less

🎞 Frame 7 · Beneficiary data is sacred · ⏱ ~3 min

🎬 Scene — A caseworker about to paste a client's name, situation, and address into a chat for "help wording an email" — and stops.

🧠 Concept — The highest Diligence duty in nonprofit work is protecting the people you serve. Never put identifiable beneficiary data (names, health, immigration status, addresses, case details) into AI tools without a vetted, approved arrangement. De-identify before you delegate.

🖼 On screen

Data rule of thumb:
  Could this identify or endanger a person we serve?
   ├─ Yes → do NOT paste it; de-identify or don't use AI here
   └─ No  → proceed with normal Discernment
Strip: names, locations, dates, anything uniquely identifying.

⚠️ Gotcha — Vulnerable populations can be harmed by a data leak in ways a typo never could. When in doubt, leave it out.

✅ Checkpoint — Name three details you'd strip before asking AI to help with a client-related email.

🎞 Frame 8 · Honesty with funders & the public · ⏱ ~3 min

🎬 Scene — A board debates whether to mention AI use in a grant. They decide transparency builds trust, not doubt.

🧠 Concept — Diligence includes honesty about how the work was made. Funders increasingly expect that impact claims are real and verifiable. Use AI to draft, but stand behind every claim as if you wrote it — because, accountably, you did.

🖼 On screen

Nonprofit integrity test:
  • Is every statistic in this document independently verified?
  • Would it survive a funder asking "where's this number from?"
  • Are we honest about outcomes, including the disappointing ones?

✅ Checkpoint — What question should every AI-drafted impact claim survive before it ships?

🎞 Frame 9 · Doing more with limited budget & staff · ⏱ ~3 min

🎬 Scene — A three-person org maps where the hours go, then redirects AI at the biggest time-sinks — and finds a day a week.

🧠 Concept — The strategic move isn't "use AI everywhere"; it's targeting AI at your biggest capacity drains so freed hours flow back to mission. Many providers also offer nonprofit pricing or grants — Diligence includes finding the affordable path.

🖼 On screen

Capacity audit:
  1. Where do staff hours actually go? (track a week)
  2. Which of those are draft/admin/summarize work?
  3. Delegate those first → measure hours reclaimed.
  4. Reinvest reclaimed hours in mission-critical, human-only work.

✅ Checkpoint — Where would one reclaimed hour per staffer per week go in your org?

Module 4 — Discernment where the stakes are high

🎞 Frame 10 · Ground AI in your real numbers · ⏱ ~3 min

🎬 Scene — Instead of asking AI for "typical outcomes," a director uploads the org's actual data and asks it to summarize that.

🧠 Concept — High-stakes Discernment means grounding AI in your verified material rather than its general knowledge. Upload your real reports and data; treat any number AI produces on its own as unverified until you trace it to a source. (Deeper: ai-capabilities-and-limitations.)

🖼 On screen

Trust ladder (nonprofit):
  Summarized from OUR uploaded data   → reliable, still spot-check
  General claim we can verify         → verify, then use
  A statistic AI generated alone      → assume invented; trace it
  "It sounds plausible to a funder"   → not good enough

✅ Checkpoint — Why upload your real annual report instead of asking AI what "typical" outcomes look like?

🎞 Frame 11 · Bias and the communities you serve · ⏱ ~2 min

🎬 Scene — An AI-drafted program description quietly uses deficit-framed language about a community. A staffer rewrites it with dignity.

🧠 Concept — AI can reproduce bias and stereotypes about the very communities you serve. Discernment includes reading output through the eyes of those communities and correcting framing that's stigmatizing, paternalistic, or inaccurate. Mission-aligned language is a human call.

✅ Checkpoint — Give one example of framing you'd watch for and fix in AI-drafted text about a community.

🎞 Frame 12 · You've got the nonprofit 4Ds · ⏱ ~1 min

🎬 Scene — Recap slide: delegate to reclaim capacity, describe your mission, discern high-stakes output, do the diligence on data and truth.

🧠 Concept — You can now stretch a lean team further while protecting the trust and the people at the heart of your mission. Next: audit it in the project.

✅ Checkpoint — Without looking, list the four Ds and one nonprofit example of each.

🛠 Project

Complete p03-ai-fluency-delegation-audit — Delegation & Discernment Audit (Nonprofit edition). You'll run a capacity audit to find your team's biggest time-sinks, draw the delegate/keep line, draft one real deliverable (a grant section or donor update) with placeholders for every statistic, build a beneficiary-data checklist, and verify three AI claims against your own records before they could reach a funder.

🧪 Self-check quiz

What are the four Ds, in order?
For a nonprofit, Delegation is fundamentally about which scarce resource?
Why instruct AI to leave bracketed placeholders for statistics in a grant?
What is the single highest Diligence duty in nonprofit AI use?
Name three details to strip before using AI on a beneficiary-related task.
What question should every AI-drafted impact claim survive?
Why ground AI in your real uploaded data rather than its general knowledge?

<details><summary>Answers</summary>

Delegation, Description, Discernment, Diligence. 2. Staff capacity / time. 3. So AI never invents numbers — you fill verified figures from your records. 4. Protecting beneficiary data — never paste identifiable info about people you serve into AI without an approved arrangement. 5. Names, locations/addresses, dates, and any uniquely identifying details (health/status/case specifics). 6. "Where's this number from?" — it must be independently verifiable. 7. AI's general knowledge can be wrong or invented; your data is verifiable and accurate to your work.

</details>

🎓 Certificate criteria

You've "passed" AI Fluency for Nonprofits when you can:

State the four Ds with a nonprofit example of each.
Identify your org's highest-value AI use cases via a capacity audit.
Draft a real deliverable with verified-only statistics (placeholders for the rest).
Apply a beneficiary-data protection checklist before delegating.
Complete p03-ai-fluency-delegation-audit and journal the time you reclaimed in learning-journal-template.

Tick this course off in progress and record the date you earned Anthropic's official certificate.

🔗 Sources & deeper notes

Official course: https://anthropic.skilljar.com/ (AI Fluency for Nonprofits)
Framework: open educational resource by Rick Dakan & Joseph Feller
Vault notes: ai-fluency-framework-foundations, ai-capabilities-and-limitations, prompt-engineering-basics
Project: p03-ai-fluency-delegation-audit · Related courses: ai-fluency-for-students, ai-fluency-for-educators