A digital render of a robot hand releasing a pink butterfly into freedom.
Getty Images For Unsplash+

Let me start with two truths that don’t get said together often enough.

First: The concerns about artificial intelligence are legitimate. Environmental costs, job displacement, hallucinations, bias baked into training data, the erosion of human connection—all represent genuine, grounded, evidence-based fears we need to address. As mission- and values-driven leaders and organizations, we can’t be expected to leverage any new tools that don’t conform to what we hold most dear, especially when these concerns are likely to be disproportionately felt by the marginalized communities nonprofits exist to serve.

Nonprofit leaders and their organizations are being asked to do more with less while the ground shifts beneath them, and the promise is that AI can help.

Second: The majority of mission-driven organizations are failing to strategically leverage AI to meet pressing, rapidly expanding demands in these turbulent times. In fact, although 92 percent of nonprofits are using AI in some capacity, the AI Equity Project 2025 survey discovered only about 15 percent currently have an organizational policy for responsible AI use. Of those using AI, an overwhelming majority of nonprofits use it ad hoc, without shared infrastructure or workflows, which means most of the value is being absorbed by individual employees instead of by nonprofits as organizations.

Let’s explore why this may be, and what forward-thinking nonprofit leaders can do about it to prepare their organizations as we leave the age of planning and enter the era of navigation, when building more resilient, responsive nonprofits has never been more important.

The gap between buzz and action

You already know the buzz is deafening. You can’t attend a single conference or any trade publication without half the sessions and articles mentioning AI in the title. Everyone wants to know how it can help them navigate what are, by any measure, extraordinarily unstable times. Nonprofit leaders and their organizations are being asked to do more with less while the ground shifts beneath them, and the promise is that AI can help.

If we want to move forward, we have to stop talking around people’s concerns and start talking through them.

But given all that interest, why is the actual adoption of AI as an organizational strategy so rare? What we have instead is individual employees experimenting on their own: drafting emails in ChatGPT, cleaning up grant language, summarizing meeting notes, and much, much more. That’s not nothing. But that’s where it stops, and the state of the sector is that just about every organization has a local AI practitioner who answers questions and serves as its go-to resource, instead of crafting a cohesive, integrated strategy designed to mitigate concerns and risk, maximize efficiencies and effectiveness, and ensure the new technology is responsibly leveraged.

While leaders may think they can get to this later, that’s largely because most EDs are unaware that without a policy to establish guardrails, there are no controls to protect sensitive data and address a wide range of other potential pitfalls. Not only that, we’re missing out on crucial opportunities for shared learning and strategic alignment with our missions. And that’s a problem, because ad hoc adoption without guardrails is how bias gets replicated and trust gets eroded.

So why the gap? In my experience working with hundreds of nonprofits and foundations on this exact question, the answer isn’t apathy. It’s a combination of overwhelm, legitimate fear, and a sector-wide conversation that has been far too abstract to be useful.

Name the fears, then get practical

If we want to move forward, we have to stop talking around people’s concerns and start talking through them. Below are the ones I hear most often, and what I think honest engagement looks like.

Job loss

This is the big one, and it’s personal. People are more worried than excited about AI throughout society, and within the nonprofit sector that anxiety is amplified by already-thin staffing and chronic underfunding. The honest answer is that AI will change roles. Some tasks will be automated, as occurs during every technical innovation. But for most small nonprofits, the real story is not displacement, it’s capacity. When you’re a three-person organization and one of those people spends 12 hours a week on data entry, donor acknowledgment letters, and formatting reports, AI doesn’t replace a person. It gives that person a good chunk of those hours back to do the relational, strategic, creative work that no algorithm can replicate. Just like calculators and spreadsheets before, this new tool can enable us to do more of what only we can do, including investing in the donor relationships that are the lifeblood of any nonprofit. 

Hallucinations and accuracy

Large language models (LLMs) make things up: In fact, they don’t actually know anything, but instead AI represents knowledge statistically. That’s a fact, not a scare tactic. Under the hood, these systems produce text by predicting what words are likely to come next, based on patterns across an enormous body of writing. There is no separate fact-checking step built into the process. Fluency and accuracy come out of the same machinery, which means a fabricated statistic arrives the same as a factually correct one. That has one non-negotiable implication: Any nonprofit using AI to generate grant narratives, impact reports, or any public-facing communications needs a human review layer, full stop. But this is a workflow design question, not a reason to avoid the technology entirely. We edit human-written drafts too. The key is building review into the process rather than treating AI output as the final product. Your mantra should be “first not final” when using these tools. Names, dollar figures, dates, citations, quotes, and anything at all about your own programs must get checked against a source document before they leave the building.

Environmental impact

The energy costs of training and running large AI models are real and growing. For organizations with environmental justice commitments, this matters. If you look at the data, the incremental impact of using AI (asking ChatGPT a question) is tiny—researcher Andy Masley offers more specifics on this point. Those per-query numbers don’t tell the whole story. They represent the cost of using a model that already exists and conveniently leave out the enormous resources poured into training these models in the first place, and the data centers that make the technology actually work. So be honest and action-oriented about limiting your organization’s environmental AI footprint.

We have more leverage here than the conversation usually admits. A 65-fold spread between models means tool selection is a sustainability decision. Sending every routine task to the largest available reasoning model is the computational equivalent of driving a semi to pick up groceries; most of what I describe in this article, drafting, summarizing, reformatting, runs perfectly well on smaller and faster models. Ask vendors where their data centers sit, how they’re powered, what they’re doing about water, and whether they publish any of it. Put those questions in your procurement process next to the ones you already ask about data privacy. Nonprofits are a small slice of this market, but we are a vocal one, and demand signals are one of the few levers we hold. 

Staying human-centered

This is perhaps the most important concern, and the one most easily dismissed by tech enthusiasts. Nonprofits exist because of human relationships: with communities, donors, partners, and the people they serve. AI should never be the voice of your organization in high-stakes, trust-dependent interactions. But it can handle the back-office burden that keeps your team from showing up fully in those interactions.

Where the Value Actually Lives

The most useful thing I can do here is get specific. These are real applications I’ve seen work for small, resource-constrained nonprofits, many of which were shared by the 57 experts we interviewed when writing AI for Nonprofits.

Fundraising

AI can draft first versions of donor acknowledgment letters, segment your donor list based on giving patterns, and help you identify lapsed donors who are statistically most likely to re-engage. (Although it is important to remember that AI, in any case but particularly fundraising, still requires a human touch when possible.) One organization I worked with used a free-tier AI tool to analyze three years of donation data and identified a segment of mid-level donors who had been completely overlooked in their annual campaign. That single insight generated $14,000 in recovered revenue. For a $300,000-budget organization, that’s not trivial.

Grant writing

AI won’t write your grant for you (at least not well), and it shouldn’t. But it can help you repurpose language from previous successful applications, tailor your narrative to a specific funder’s priorities, and catch inconsistencies between your budget narrative and your line items. For small shops where the executive director is also the lead grant writer, this can cut drafting time by 30 to 40 percent.

Marketing and communications

Generating social media content calendars, repurposing a single blog post into multiple platform-specific formats, drafting newsletter copy, writing alt-text for images: these are high-volume, repetitive tasks that consume disproportionate staff time. AI handles them well, and the equity implication is significant. Organizations that can’t afford a communications director can still maintain a consistent, professional public presence.

Back-office operations

Summarizing board meeting minutes, generating first drafts of policies, cleaning and deduplicating data in your CRM, translating materials into multiple languages. These are the unglamorous tasks that eat small organizations alive. AI doesn’t make them exciting. It makes them faster.

The equity dimension we keep missing

Here’s what frustrates me most about the current AI conversation in the nonprofit sector: The equity analysis is almost always about the risks of AI, and almost never about the risks of not adopting it.

The organizations with the least capacity—small, community-based, led by people of color, operating in rural or under-resourced areas—are the ones that stand to benefit most from tools that reduce administrative burden. They’re also the ones least likely to have access to training, strategic guidance, or the institutional permission to experiment. When we frame AI exclusively as a threat, we inadvertently widen the gap between large, well-resourced organizations (which will adopt it regardless) and the grassroots groups that need every efficiency gain they can get.

An equity lens on AI doesn’t mean avoiding it. It means ensuring access is distributed fairly. It means funders covering the cost of training and implementation, not just programs. It means capacity-building organizations offering practical, jargon-free guidance rather than high-level thought leadership that doesn’t translate to a Tuesday morning at a two-person nonprofit.

Nothing is going to save the nonprofit sector except the people in it doing the hard, relational, deeply human work of building power and serving communities. But those people deserve tools that free them up to do that work.

A starting point, not a destination

If your organization hasn’t taken a strategic step toward AI yet, here’s what I’d recommend—not as a technology roadmap but as a leadership exercise:

  1. Audit your time. Have every staff member track their tasks for one week and flag the ones that are repetitive, high-volume, and low complexity, AKA “low joy” and “no joy” tasks. That’s your AI opportunity map.
  2. Pick one use case. Don’t try to transform everything. Choose a single workflow: donor acknowledgments, social media scheduling, meeting summaries, and pilot an AI tool for 30 days.
  3. Set a policy before you scale. Even a one-page document that addresses data privacy, human review requirements, and prohibited uses gives your team a framework for responsible experimentation. Fast Forward has a great, free AI policy generator that can help.
  4. Talk about it openly. The worst thing you can do is let AI adoption happen silently. Name the concerns, share what you’re learning, and make space for staff to say what’s working and what isn’t.

AI is not going to save the nonprofit sector. Nothing is going to save the nonprofit sector except the people in it doing the hard, relational, deeply human work of building power and serving communities. But those people deserve tools that free them up to do that work. And right now, too many of them are drowning in administrative tasks that a machine could handle in seconds.

The question isn’t whether AI belongs in the nonprofit sector, since it’s already here. The question is whether we’re going to be intentional about how it’s used or leave it to chance and let the inequities we claim to fight get quietly automated into our own operations.