What Nonprofit Leaders Should (and Shouldn’t) Use AI For

AI can absolutely make your nonprofit faster and sharper. It can also make you confidently wrong. The difference is leadership, not the tool.

Last week, I sat in on a leadership meeting where someone said, “We should just have ChatGPT write the grant.” Everyone laughed. Then nobody laughed. Because the idea wasn’t a joke anymore. It was a stress response.

Here’s the moment I see over and over: a smart nonprofit team is buried, hiring feels impossible, reporting deadlines do not care about your capacity, and suddenly, AI looks like a miracle intern who never sleeps.

But then the intern starts making things up. Not maliciously. Just… confidently. That’s the first leadership lesson of March: AI is not a replacement for judgment. It’s a force multiplier for whatever system you already have. If your system is messy, it multiplies mess.

And that’s why this month matters. Interest in AI spikes right about now because leaders are trying to plan the year, protect capacity, and keep up with funder expectations. Let’s demystify what AI is actually good for in a nonprofit, what it’s terrible for, and the guardrails you need so your excitement doesn’t turn into a headline you regret.

1) Start with the truth: AI is “prediction,” not “understanding.”

Most generative AI tools work by predicting what text (or image) should come next based on patterns in data. That’s why they can write a decent donor email in 20 seconds and cite a study that doesn’t exist.

This is not a minor quirk. It is the whole game. AI can be useful and risky at the same time, and leaders need to hold that tension without getting dramatic.

If you want the broader context on why adoption is accelerating (and why governance matters), Stanford’s 2025 AI Index is the best “big picture” snapshot out there.

2) The best nonprofit AI uses are the boring ones

If your team is new to AI, do not start with “high stakes” work. Start with “high drag” work. The tasks that steal time but don’t require your soul.

Here are the safest, highest-value starting points I’m seeing across the sector:

Drafting and editing (with a human owner).

First drafts of job posts, event run-of-show documents, board packet summaries, donor thank-you notes. You still need a human to make it sound like you and verify facts.

Summarizing and extracting.

Turn long meeting notes into action items. Pull themes from survey comments. Create a 1-page brief from a 12-page report (then have the actual program lead confirm it).

Brainstorming and outlining.

Campaign themes, workshop agendas, naming options, interview questions. AI shines when you need momentum, not answers.

Translation and accessibility support.

Draft plain-language versions of program info, create alternative reading levels, translate content (then have a bilingual human validate).

Basic data cleanup and pattern spotting (with guardrails).

Reformatting data, generating charts, and identifying outliers. Keep a human in charge of interpretation, because context matters.

Sector research backs up that nonprofits are using AI first for communications and productivity, while concerns cluster around privacy, accuracy, and bias. That combination is exactly why “low-risk productivity” is the right on-ramp.

3) The worst nonprofit AI uses are the tempting ones

Here’s where nonprofit leaders get into trouble. Not because they are reckless, but because they are exhausted.

Do not use AI as a decision-maker.

AI can support decisions. It cannot be the decider on program eligibility, HR actions, client risk scoring, or anything that could harm someone if it’s wrong.

Do not feed sensitive data into tools you haven’t vetted.

Client information, donor data, protected health info, student records, internal financials, legal matters. If you cannot clearly explain where the data goes, who can access it, and how it is retained, you do not put it in.

Do not outsource trust.

If AI writes donor communications, grant narratives, or public statements, a human still owns the relationship. Authenticity is not “nice.” It is strategy.

Do not treat AI outputs as “research.”

AI is a starting point for research, not a source. If your team is using AI for facts, require citations and verification every time.

Nonprofit and foundation leaders consistently name security, accuracy, staff expertise, and bias as core concerns, even while usage expands.

4) A simple “Should we use AI?” test that actually works

Most teams overcomplicate this. Here’s the framework I recommend because it fits on a sticky note:

The R.I.S.K. test

R = Reversibility: If it’s wrong, can we undo it easily?

I = Impact: If it’s wrong, who gets hurt? How badly?

S = Sensitivity: Does it involve confidential or regulated data?

K = Knowledge requirement: Does it require lived context, judgment, ethics, or nuance?

If a task is highly reversible, low-impact, low-sensitivity, and low-judgment, AI is usually fine with review. If it’s the opposite, AI can still support the work, but only inside a defined process with human accountability.

This is essentially what formal risk frameworks are trying to operationalize. The NIST AI Risk Management Framework (AI RMF 1.0) is worth borrowing from even if you never say “NIST” out loud in a staff meeting.

5) Your real work is governance, not prompts

Most nonprofit AI conversations get stuck at “What tool should we use?” That’s the wrong question.

The leadership questions are:

• What do we consider acceptable use?
• What data is off-limits?
• What decisions must remain human?
• How do we document and audit AI-assisted work?
• How do we protect equity and avoid harm?

McKinsey’s research continues to show a gap between AI experimentation and scaled value, and it repeatedly points to governance, talent, and operating model issues as the bottleneck, not tool availability.

Translation for nonprofits: You don’t need a fancy AI initiative. You need clean rules and a grown-up process.

6) What an AI policy should cover (without becoming a novel)

A solid nonprofit AI policy is not a legal dissertation. It’s a practical agreement that protects people, trust, and mission.

At a minimum, your policy should clarify:

1. Approved tools (and who can approve new ones)
2. Banned data categories (client, donor, HR, financial, legal)
3. Human review rules (what requires sign-off and by whom)
4. Disclosure expectations (when you will say AI was used)
5. Quality controls (fact-checking, citations, bias checks)
6. Equity lens (how you’ll assess disparate impact)
7. Training requirements (who must be trained before using AI)

If you do only one thing this month, do this: create a one-page “AI Guardrails” sheet and require every staff member to read it.

TAKE ACTION (6–10 practical steps)

• Inventory where AI is already being used (you’ll be surprised).
• Create a short “approved use cases” list that starts with low-risk tasks.
• Define “no-go” data categories and make them non-negotiable.
• Require human review for anything external-facing or mission-critical.
• Establish a fact-check rule: AI outputs need sources and verification.
• Pilot one workflow for 30 days (example: board packet summaries).
• Add a short disclosure line for communications where it’s appropriate.
• Train managers first so enforcement is consistent and not chaotic.
• Assign one owner for AI governance (not “everyone,” not “IT”).

Support to keep you moving forward

If your team is feeling both excited and intimidated, that’s normal. This is exactly where good leadership shows up: you create clarity and reduce risk so people can actually experiment responsibly. If you want help doing this without spinning your wheels, the EmpowerED coaching approach at C&A: Social Impact Consulting can support you in building a practical AI policy, training your managers, and setting up a simple governance process your team will actually follow.

For extra grounding, I also recommend reading the NIST AI RMF’s framing of risk, then translating it into plain-language rules your staff can use without needing a compliance degree.

Recommended Resources

Books

• Human Compatible: Artificial Intelligence and the Problem of Control by Stuart Russell
• Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy by Cathy O’Neil
• Noise: A Flaw in Human Judgement by Daniel Kahneman, Olivier Sibony, Cass Sunstein
• The Alignment Problem: Machine Learning and Human Values by Brian Christian

Listen/Watch

• TED Talk: “The era of blind faith in big data must end” (Cathy O’Neil)
• Stanford HAI talks/interviews connected to the AI Index (browse from the report hub)

Articles/Reports

• Stanford HAI: 2025 AI Index Report
• NIST: AI Risk Management Framework
• Salesforce: 7th Edition Nonprofit Trends Report (2025)
• CEP: AI With Purpose

Frequently Asked Questions

1) What is the best nonprofit AI leadership approach in 2026?

The best nonprofit AI leadership approach is to start with low-risk use cases, set clear rules for sensitive data, and require human review for high-impact work. Treat AI as a support tool, not a decision-maker, and build basic governance before scaling.

2) Should nonprofits let AI write grants?

AI can help draft outlines, polish language, and summarize program descriptions, but it should not be the final author. Grants require accuracy, credibility, and an authentic voice, and AI outputs must be fact-checked and owned by a human.

3) Is it safe to put donor or client data into AI tools?

Not by default. Unless your organization has vetted the tool’s data handling and privacy terms and established clear rules, donor and client data should be treated as off-limits for generative AI.

4) Do nonprofits need an AI policy?

Yes. Even a one-page policy reduces risk, sets expectations, and prevents inconsistent use across staff. A practical AI policy is now part of responsible nonprofit management.

If you want to use AI without creating chaos, start with guardrails. If you’d like support building a practical AI policy and training your managers, reach out through C&A: Social Impact Consulting, and we’ll help you design an approach that fits your mission and capacity.