AI for nonprofits already means drafting updates, organizing notes, and preparing reports. CEP reports that almost two-thirds of surveyed nonprofits use AI, mainly for communications and internal work. The practical challenge is choosing a useful task, setting rules, and protecting people. AI for good starts there.
Backed By What is an independent explorer documenting how AI can make ordinary people's lives better. Read more about Backed By What. This guide recommends no specific paid products and contains no affiliate links. It approaches AI for social good through everyday work: what to provide, what to ask, and what a person must check.
Where nonprofits stand with AI right now
Most teams already use it, mostly for writing
CEP's AI With Purpose draws on a survey of 451 nonprofits in April–May 2025. Among the N=279 nonprofits that use AI, the most common uses were communications (84%), internal productivity (63%), and development/fundraising (61%). These are use-case percentages among AI users. See CEP's survey and use-case chart.
Few have a plan
The TechSoup/Tapp Network survey of 1,321 respondents, conducted in Q2–Q3 2024, reported that 76% lacked an AI strategy and 80% lacked an AI acceptable-use policy. Those dates matter: this describes an earlier survey, rather than a fresh measurement of today's teams.
Bigger budgets adopt faster
Among respondents reporting generative AI use, TechSoup's budget comparison lists 34% from organizations below one million dollars and 66% from those above that threshold. A small team can start with a narrow task and clear boundaries. Buying more software is a decision to evaluate after a trial.
6 practical ways nonprofits use AI today (with real workflows)
Use these suggested workflows as starting points for a supervised trial. Choose public or approved, non-sensitive material. Keep the original beside the draft so the reviewer can compare them.
Drafting grant applications and reports
- Input
- Provide the application questions, word limits, and approved program facts. Exclude donor identities and beneficiary records.
- What AI does
- Organizes the facts into an outline, drafts a section, or shortens an existing report. Ask it to flag missing information instead of filling gaps.
- What a human checks
- Verify every figure, outcome, and promise against organizational records. The program lead approves the commitments; the writer checks that the answer addresses the funder's actual question.
Email updates, social posts and plain-language updates
- Input
- Use an approved public announcement, its intended audience, and the action readers should take.
- What AI does
- Drafts a short email, social post, and simpler version from the same source. Ask it to preserve dates and instructions and avoid adding claims.
- What a human checks
- Compare each version with the announcement. Check dates, links, tone, and whether the next step is clear. Remove invented stories or language that portrays people receiving support as helpless. Publish only the approved version.
Meeting notes and board minutes
- Input
- Start with approved notes from a routine planning meeting. Confirm permission before using a recording or transcript.
- What AI does
- Separates discussion points, decisions, unresolved questions, and proposed actions into a draft record.
- What a human checks
- Confirm who agreed to each action. Check that suggestions have not become decisions. Have the responsible person approve minutes, then follow the agreed retention rules for source material.
Summarizing community feedback and surveys
CEP quotes a nonprofit leader's hope that AI could help identify community needs and measure program impact. Treat that as an aspiration to test.
- Input
- Prepare anonymized feedback about a public workshop. Remove identifying details and exclude sensitive responses.
- What AI does
- Suggests themes, groups similar comments, and flags responses that do not fit the main themes.
- What a human checks
- Read the original responses. Preserve minority views and uncertainty. Check any counts yourself, and avoid treating a set of comments as proof of what the whole community needs.
Accessibility: translation, captions, easier reading
Try AI for accessibility with a specific reader or listener in mind. Treat generated versions as drafts requiring review.
- Input
- Use a public event description, an approved video script, or a page explaining how to join an activity.
- What AI does
- Produces a draft translation, captions, or a version with shorter sentences and clearer steps.
- What a human checks
- A fluent reviewer checks translation meaning and local phrasing. Someone checks captions against the recording. Test whether the revised instructions preserve every essential detail and work for the intended audience before sharing.
Answering common questions (FAQ / chatbot) with a human fallback
The U.S. Chamber recommends routing complex chatbot questions to a person.
- Input
- Supply approved public answers about opening times, event access, and where to find information.
- What AI does
- Drafts an FAQ or answers from that limited material. Instruct it to say when the material does not contain an answer.
- What a human checks
- Test ambiguous questions and outdated information. Provide a clear route to staff, explain the system's limits, and prevent it from requesting personal histories or deciding whether someone qualifies for support.
What not to hand over to AI
Keep beneficiary personal information, individual case records, and donors' private information out of public AI tools. NTEN's policy template calls for authorization before entering employee, client, or vendor personal data, and removing or anonymizing personal data where feasible. For a first trial, use public material.
Keep decisions about who receives services with accountable people. Reviewers need the original context and a way to hear concerns or corrections. AI helping people should leave them able to question what happens.
In CEP's survey, nonprofit concerns included misinformation or inaccurate results (73%), data security and privacy (67%), and bias (50%). Build checks around those risks: verify claims, restrict inputs, and review whose perspective a draft may overlook.
A simple responsible AI policy you can write in an afternoon
Start with a page your team can use while working. Name who approves tools, who reviews outputs, and where staff should report a mistake. A suggested three-column structure:
Outlines, rewrites, and summaries using public or approved non-sensitive material within agreed tool rules.
Anything shared externally, including grant drafts, translations, meeting minutes, and answers for the public.
Personal data in public tools, invented evidence, unverified publication, or automated decisions about access to services.
Follow NTEN's review guidance: someone with appropriate expertise checks accuracy, verifies against non-AI sources, and looks for bias. Give that person enough time to compare the draft with its sources.
Before approving a tool, use NTEN's vendor questions: ask what data it collects, how it stores that data, whether inputs can be anonymized, and whether data can be deleted after use. Record the answers. For deeper work, NIST's AI Risk Management Framework is intended for voluntary use.
How to start in 30 days with no new budget
This is a suggested trial plan. Many general-purpose AI assistants have free tiers, and some vendors offer nonprofit discounts. Start with tools already available to your team and check their data rules first.
- Week 1: Choose a task. List three writing tasks that consume staff time. Pick one with public inputs, a clear output, and a reviewer. Save an example of how you currently do it.
- Week 2: Run a small trial. Try that task with an existing tool. Save the prompt and source material. Record drafting and review time, including corrections.
- Week 3: Write the rules. Turn what you learned into a page covering allowed inputs, review, approval, and deletion. Walk through it with the people doing the work.
- Week 4: Compare and decide. Compare total effort and output quality with your earlier example. Continue, revise the workflow, or stop. Include the review burden in the decision.
For freely accessible learning resources, explore NTEN's AI resource hub and the data.org AI for Nonprofits Resource Hub. Choose a resource that helps resolve the question your trial raises.
What funders can (and mostly don't yet) help with
CEP says nearly 90 percent of foundation leaders report their foundations offer no funding or nonmonetary support for grantees' AI use. Nonprofits' top requests included staff education on AI fundamentals (68%) and funding for AI tools and software (63%).
Consider including AI training in a capacity-building request. Describe the task, the people who will review it, the data boundaries, and how you will judge the trial. Ask for support with staff learning and evaluation as well as access to tools. Make the request specific to your organization's needs.
FAQ
How are nonprofits using AI today?
Nonprofits mainly use AI for communications, internal productivity, and development or fundraising. In CEP's survey, those uses were reported by 84%, 63%, and 61% respectively of the N=279 nonprofits using AI. A practical starting point is drafting from approved material, followed by someone checking the result.
Is AI safe to use with client or beneficiary data?
Keep personal data out of public AI tools by default. NTEN recommends authorization for personal data inputs and anonymization where feasible. Before any approved use, check the tool's collection, storage, and deletion arrangements. For an initial trial, choose public information that does not identify anyone receiving support.
Do small nonprofits need an AI policy?
Yes. Start with a page that explains encouraged uses, uses needing human review, and prohibited uses. The TechSoup/Tapp survey of 1,321 respondents in Q2–Q3 2024 reported that 80% lacked an acceptable-use policy. Assign responsibility for approving tools and checking outputs, then revise the rules after your trial.
What should a nonprofit never use AI for?
Keep final decisions about access to services with people. Do not put sensitive personal information into public tools or publish unverified output. CEP reports concerns about inaccurate results, privacy, and bias. Give staff authority to stop a workflow when they cannot check its inputs, reasoning, or consequences.
Can AI help with grant writing?
Yes. CEP describes nonprofits using AI to draft or refine grant applications and reports. Try it for an outline or a clearer version of approved text. A person must verify facts, figures, and promises against organizational records, fill information gaps, and approve every commitment before submitting the application.
Where can nonprofits learn AI for free?
Start with freely accessible resources in NTEN's AI resource hub and the data.org AI for Nonprofits Resource Hub. For a framework to explore risks, use the voluntary NIST AI Risk Management Framework. Choose materials relevant to your current task, then discuss how their guidance fits your team's responsibilities.
Do nonprofits need to buy special AI software?
Not necessarily. Begin with an approved tool you already have, public inputs, and a narrowly defined task. Many general-purpose AI assistants have free tiers. Compare total drafting and review effort before buying anything. A purchase should address a demonstrated need and fit your data rules, staff capacity, and review process.
AI for good begins with a real need and care for the people affected. Choose a manageable task, protect the people you serve, and keep someone accountable for the result. Let the trial show whether a tool earns a place in your work. Explore AI for good in everyday life, or return to the journal for more notes on AI for good.