Template
Nonprofit AI Use Policy
A fill-in-the-blank policy your team can adopt in one meeting.
Adapted from Artificial Intelligence for Nonprofit Fundraising by Dale Nirvani Pfeifer.
Replace every [bracketed field] with your own answer. Delete anything that does not apply. Keep it short: a policy people read is worth more than a policy people file.
Before you start
Write this with staff, not for them. When the International Center for Journalists built its AI policy, a cross-functional group anchored the work in four principles before naming a single tool: do no harm, protect rights and privacy, use content with consent, and be transparent about AI use. Drafts circulated widely over about six months. The result did not slow experimentation. It made it possible.
1. Purpose and scope
This policy governs how staff, contractors, and volunteers at [Organization name] use artificial intelligence tools in fundraising and communications work. It exists so that our team can adopt useful tools with confidence, and so that our donors can trust how we work.
What counts as AI under this policy: any tool that generates text, images, audio, or video, or that scores, ranks, predicts, or recommends actions about people. This includes general assistants, AI features inside our CRM or email platform, and AI built into tools we already pay for.
Guiding principles. We commit to: [do no harm / protect rights and privacy / use content with consent / be transparent about AI use / add or amend].
Effective date: [date].
Applies to: [all staff / development team / named roles].
2. Approved and prohibited uses
Approved. Staff may use approved AI tools to: draft first versions of appeals, thank-you letters, and social copy; summarize documents and meeting notes; research funders and prospects from public information; clean and structure data; and generate ideas and outlines.
Prohibited. AI may not be used to: send any donor-facing communication without human review; make a final decision to solicit, decline, or deprioritize a specific person; generate impact claims, statistics, or beneficiary stories that are not independently verified; create synthetic images or voices of real beneficiaries, staff, or donors; or process the data listed in Section 4 in an unapproved tool.
Additional prohibitions for our context: [add any specific to your mission, for example clinical, legal, immigration, or safeguarding contexts].
3. Human review
A named person reviews and approves AI-assisted work before it leaves the organization. AI drafts. A human decides.
| Work | Review required | Approver |
|---|---|---|
| Donor communications | Every item, before sending | [role] |
| Grant applications and reports | Every item, plus fact check | [role] |
| Prospect scores and segments | Reviewed before outreach | [role] |
| Public content and impact claims | Every item, sources verified | [role] |
| Internal notes and summaries | Spot check | [role] |
The reviewer is accountable for accuracy, tone, and fairness, the same as for any other work that carries our name.
4. Donor data and privacy
Never entered into a general AI tool: donor names paired with giving history, contact details, health or beneficiary information, payment data, passwords, unpublished financials, or anything a donor shared in confidence.
Permitted: aggregated or anonymized data, published information, and our own already-public materials. Donor data may be processed inside [named approved systems], which are covered by our contracts and data agreements.
Vendor requirements. Before approval, a tool must confirm in writing that our data is not used to train its models, state where data is stored, and offer deletion on request.
If donor data is exposed: notify [role] within [24 hours] and follow our existing data-incident procedure.
5. Disclosure and transparency
We disclose AI involvement when it materially shaped something a donor would reasonably want to know about. Silence costs more than candor: donors who discover undisclosed AI use lose trust in the organization, not the tool.
We disclose when: AI generated a substantial portion of a published piece; AI features in imagery or audio; or AI is used in a decision process a donor is entitled to understand.
We do not routinely disclose: spelling and grammar assistance, internal summarization, or list cleanup.
Standard language we use: “[Draft your one sentence here, for example: This message was drafted with AI assistance and reviewed by our team.]”
6. Approving a new tool
Adoption should be a decision, not a drift. Before a new AI tool is used for organizational work, the requester submits it to [role] with answers to five questions:
- What specific task does this replace or improve, and how will we know it worked?
- What data does it need, and is that data permitted under Section 4?
- Where does human review sit in the workflow?
- What does the vendor commit to on training, storage, and deletion?
- What does it cost, including staff time to adopt it?
Approved tools are listed in [shared location]. Anything not on that list is not approved for organizational work.
7. Ownership and accountability
One person owns this policy: [name, role]. Committees may advise; a named person is accountable for keeping it current and answering questions about it.
Staff responsibilities. Use approved tools only, follow the review requirements, and raise anything uncertain rather than guessing. Nobody is penalized for asking.
Leadership sign-off: [name, role, date].
8. Review cadence
This policy is reviewed quarterly for the first year, then twice a year. AI tools change faster than annual policy cycles, and staff practice shifts as new tools appear.
Each review asks: what are people actually using, what went wrong or nearly wrong, what has changed with our vendors, and what needs to be added or removed.
Next review date: [date].
Owner: [name].
Adopting this in one meeting
Send the draft ahead. Spend the meeting on three questions only: what should AI never touch here, who signs off on donor-facing work, and when do we tell donors. Fill the brackets live, name the owner, set the review date, and publish it the same week. You can refine it at the first quarterly review.