How can nonprofits use AI on a small budget?

Most small nonprofits run on a few staff and a lot of goodwill. The jobs that drain them are rarely the mission itself; they are the forms, reports, newsletters and minutes around it. That is exactly where a general chat assistant helps most, and where the free versions are often good enough.

The trick is to pick tasks where the AI drafts and a person decides. A model can turn twenty pages of funder guidelines into a checklist in a minute. It cannot know your programs, your community or your numbers, and it will sometimes state things that are wrong with full confidence. So the human stays the author; the AI is the fast first draft.

Where the hours go, and where AI fits

None of these needs a paid plan to try. Start with one task for a month, note the time saved, then decide whether a subscription is worth it.

Worked example 1: a grant summary

Copy the funder's public guidelines (not your application) into a chat and use a prompt like this:

You are helping a small community nonprofit decide whether to apply for a grant.
Below are the funder's published guidelines.

Summarize them as:
1. Who is eligible, in one or two sentences.
2. What the money can and cannot pay for, as two short lists.
3. Every deadline and required document, as a checklist.
4. Three questions we should ask the funder before applying.

Quote the exact sentence from the guidelines for each deadline.
If something is unclear or missing, say so instead of guessing.

[paste the guidelines here]

The last two lines do real work. Asking for the exact sentence behind each deadline lets you check the summary in seconds, and asking the model to flag gaps reduces confident guessing. Always compare the dates with the original before you plan around them.

Worked example 2: a newsletter from rough notes

Write a 250-word newsletter update for our volunteers and supporters.
Tone: warm, plain language, no jargon, no exaggeration.
Use only the facts in my notes. Do not invent numbers or quotes.
End with one clear way to help this month.

Notes:
- summer reading club ran 6 weeks at the library
- volunteers needed for September book sort, Saturdays
- thank our partner, the local library

Notice what is missing from the notes: no donor names, no children's names, no addresses. You add those yourself, in your own email tool, after the draft is approved. (The figures in this prompt are illustrative.)

Spending little, wisely

Free tiers of the main chat assistants are enough for drafting. A paid team plan becomes worth discussing when several people use AI every week and you need shared settings and stronger data terms. OpenAI's ChatGPT Team, launched in January, is one example: it costs $25 per user per month billed annually (or $30 monthly), and OpenAI says it does not train its models on Team data or conversations. For a board of volunteers, one or two seats for the people who write the most may be all you need.

Write a one-page AI policy before anyone pays for anything: which tools are approved, what may never be pasted in, and who reviews AI-assisted text before it goes out. It costs nothing and prevents most problems.

Which free tools are safe for donor data?

The cautious answer is: treat no free consumer tool as a place for donor data unless your organization has checked its terms and settings and decided it is acceptable. Donor records hold personal information (names, contact details, giving history, sometimes health or family circumstances behind a gift), and your privacy obligations don't disappear because a tool is convenient.

Canada's federal, provincial and territorial privacy authorities set out principles for organizations using generative AI in December 2023, and they translate well to a small charity. Use only tools that respect privacy law. Where possible, put anonymized or de-identified information in prompts rather than personal details. Enter sensitive personal information only when you are authorized to. And unless there is a reason, don't keep prompts or reuse them for other purposes. The authorities also point out that an inference the AI makes about an identifiable person, such as "likely to give again," is itself personal information.

In practice that means three habits. Replace names with placeholders ("Donor A gave in March") before asking for help with a thank-you letter, then fill in the real details yourself. Check each tool's data settings, including whether conversations may be used to improve its models and whether you can turn that off. And if you need AI to work directly with donor records, use a business plan with written data terms, or a feature inside the donor database you already trust, after your board has approved it. In Québec, Law 25 adds specific duties for organizations handling personal information, so check them before connecting any new tool to your lists.