AI Fluency for Nonprofits
There is a gap in the sector between AI enthusiasm and meaningful implementation. This edition closes it with one organizing principle: every AI efficiency gain should ultimately translate to greater impact for the communities you serve. Mission-centered thinking guides everything here.
- Anchor every AI decision to mission and community impact
- Research and write grants with your authentic voice, verified
- Validate AI analysis against known-answer data first
- Draft the seven-part organizational AI policy
Welcome, mission first
Nonprofits face a specific version of the AI question: limited time, limited money, high stakes for the people served, and donors watching how both are spent. The answer is not more enthusiasm; it is fluency anchored to mission.
Write the document you will open every AI collaboration with: your mission, vision, and values first, then who you serve, your programs, and your constraints. Close with the sentence: "If AI could help me ___, I would spend more time ___," aimed at higher-impact, human-centered work. That sentence keeps efficiency honest.
- Efficiency is only a win if it converts to impact for the people you serve.
- Mission, vision, and values open the context document, not tooling.
The framework, for mission work
The 4Ds in full live in Framework & Foundations. Each competency has three subcomponents (Delegation: problem, platform, task; Description and Discernment: product, process, performance; Diligence: creation, transparency, deployment), and each shows up daily in nonprofit work: delegating grant research, describing your program to a model, discerning a funder-facing draft, disclosing AI's role in a report.
Map your current AI use, or intended use, to the competencies. Then answer honestly: is the one you gravitate to where you most need to develop, or are you avoiding a harder competency that would unlock more progress?
Researching with AI
Picture an executive director, Maria, expanding a housing nonprofit from one city to another. She needs policy differences, funding landscapes, and compliance requirements, a research load that used to take weeks.
The research prompt recipe
- The specific policy or question.
- Your organizational context: who you serve and why this matters.
- What you need to know: beneficiary impact, funding, compliance, advocacy angles.
- Timeframe and geography, explicitly.
On any AI research output: flag at least two claims for verification before use, note which community perspectives are missing from the answer, and mark anything that smells stale or too general. Stretch: ask the AI to track down the original source of one claim, then compare how accurately the claim represented it.
- Context in the prompt is what turns generic research into your research.
- Two flagged claims and a missing-perspectives check, every time.
Writing with AI
Now picture James, running an environmental justice organization, with an emergency grant deadline 48 hours away. AI can get a credible draft on the table in an hour. What makes it fundable is everything he adds afterward.
The authentic-voice toolkit
- Upload past successful proposals so the AI writes toward your proven voice.
- Inject the details AI cannot know: your track record, partnerships, staff expertise, actual community relationships.
- Make revision requests specific, never "improve this."
The grant-draft discernment checklist
- Verify every data point.
- Generic language versus your actual voice: which is on the page?
- Catch deficit-based framing about the people you serve, and rewrite it.
- Does the draft address the funder's priorities, or just describe your program?
- What is missing that only you know?
Own the final result. If it goes out under your organization's name, you stand behind every word.
Your AI grant draft describes the community as “trapped in cycles of dependency”. What is the catch?
Generic AI text often frames communities by their deficits. Rewriting toward strengths and agency is a discernment move the sector specifically needs.
- Past successful work is the best style guide you can give a model.
- Deficit framing is a discernment catch the sector specifically needs.
Privacy and data
A new privacy consideration arrived with AI: training. Patterns from data you share could influence a model's future outputs, depending on the tool and plan. Different tools have different rules; match your tool to your task, and read the terms before sensitive material goes anywhere.
- You can usually get the full benefit without sharing sensitive information by breaking tasks into component parts.
- Work backwards from your actual goal to determine what data is truly necessary. Pattern analysis rarely needs personally identifying information at all.
- If something goes wrong: delete the conversation, request deletion from the platform, follow your organization's protocols. A provider without deletion and training controls is a warning sign in itself.
Take a donor spreadsheet, a participant survey, or a grant report. Annotate it two ways: which fields are personally identifying, and which are analytically necessary for the task you have in mind. Note the worst case if each identifying field leaked. Then choose the tool tier and the verification steps that match.
- Training is a privacy consideration now; read the tool's terms first.
- Work backwards from the goal; most analysis needs no PII.
Data analysis, validated
Meet Rio, analyzing program data at a veterans' services organization. The method that makes AI analysis trustworthy is simple and transferable:
Test AI against data you already understand. Before trusting AI with new analysis, validate it on past data where you know the correct results. Build a set of validated approaches, documented, and reuse them. Validation builds confidence, but it does not eliminate responsibility.
Start with your own public communications: ask AI to analyze what themes and language perform best across your newsletters and posts. No sensitive data involved, and the patterns are directly usable. Stretch: audit your messaging against your stated mission and values, and build a messaging guide from the gaps it finds.
- Validate on known answers before trusting new analysis.
- Documented, validated approaches beat one-off trust every time.
Workflow automation
Emily runs events at a community foundation, and the gala inbox is drowning her. Before building an AI email responder, she asks the question that defines this whole lesson:
Task delegation means asking "should AI do this?", not just "can AI do this?"
The method
- Audit the real workload first. What are the patterns, the frequency, the degree of standardization?
- Sort into three buckets: AI can handle (documented, standardized); AI assists, human decides; human should handle (high-stakes, emotional, complex judgment).
- Test with real past emails, not invented ones. The gaps you find in your instructions are normal and necessary; that is the description improving.
All three diligences apply to automation: be intentional about what you automate, review outputs before they go out (especially early), and be honest about AI's role, especially if something goes wrong.
- "Should AI do this?" governs every automation decision.
- Three buckets: handle, assist, human. Test on real history.
The organizational AI policy
"Human in the loop" means something specific in mission work: you decide what problems AI helps solve, you evaluate value alignment, and you maintain the relationships and real-world impact that define nonprofit work. A dependency test worth running quarterly: can we explain what the AI is doing? If yes, that is healthy augmentation. If not, rework the process until you can.
At its best, AI cuts through noise so you can focus on the handwritten card, the site visit, the conversation. At worst, it automates that human touch away. Set cultural norms about what happens with saved time.
The seven-part organizational AI policy
- Platform awareness. Approved and prohibited tools; what retention and training terms are acceptable at each data-sensitivity level; who keeps this current.
- Task delegation. What is appropriate for AI, what stays fully human, who decides new use cases, and how gray areas get resolved.
- Expectations and capacity. Where saved time goes; realistic expectations per role; capacity across the team rather than one designated "AI expert"; how failures are handled.
- Quality and oversight. Who reviews what, verification steps per content type, mistake handling, ongoing monitoring.
- Transparency. What stakeholders, funders, and served communities need to know; disclosure and attribution in grants and reports.
- Values alignment. Mission first; the dignity of the people served; and the cases where you will not use AI even when it would be more efficient.
- Compile and maintain. Draft it with AI's help, review it as a team, and set revisit dates.
- The dependency test: can we explain what the AI is doing?
- Seven parts, and part six includes when not to use AI at all.
Next steps
Pick one real task this week: a grant application, a donor report, program data. Run it through the loops. The framework is a loop, not a line, and your certificate is a conversation-starter inside your organization.
Course quiz
Attribution. Adapted from the AI Fluency courseware developed in collaboration with Anthropic, CC BY-NC-SA 4.0. This adaptation © 2026 AI Literacy Foundation, shared under the same license.