5 Habits for Writing Better Nonprofit AI Prompts (+ Free Templates)

Most nonprofit AI prompts are one sentence long. “Write a thank-you email for a donor who gave $250.” You paste it in, you get something back in four seconds, and it’s … fine. The message is grammatically correct, warm in roughly the way a grocery-store greeting card is warm, and interchangeable with the thank-you email every other nonprofit in your ZIP code sent last week.

So you rewrite it yourself, which raises a fair question: what did the AI actually do for you there?

Not much. That’s not really the tool’s fault, though. You gave it a low-effort prompt, and it gave you a low-effort answer. If these things were able to be disappointed, the chatbot would be disappointed too (but fingers crossed they never emote). 

Luckily, the solution here is simple: Use a better prompt, and you’ll get a better answer. 

Here are five habits that separate a prompt that saves you an hour from one that costs you 20 minutes and your good mood. They work in ChatGPT, Claude, Gemini, or whatever’s built into the tools you already pay for. Let’s get into it!

Donors Support Their Favorite Nonprofits Using AI

Working with AI tools at a nonprofit can feel nerve-racking, if only because there’s a kind of ambient “common knowledge” that donors are just waiting to pounce the second they detect even a whiff of AI use from the orgs they support.

As it turns out, though, that’s not really the case. We surveyed 2,007 nonprofit donors about AI and how they felt about a nonprofit using AI for different kinds of work, like drafting emails, analyzing data, etc. 

The most common response in all seven scenarios we tested was that … it made no difference to their trust. And in every one, the donors whose trust went up actually outnumbered the ones whose trust went down. About 94% generally approve of nonprofits using AI. 

All in all, we found that:

• When donors ranked eight AI-assisted tasks from most to least acceptable, brainstorming ideas for emails and other communications landed third, ahead of financial forecasting and ahead of drafting or editing those communications outright. 
• On thank-you messaging specifically, 37.8% of those donors said AI involvement would increase their trust in the organization, while 22.8% said it would decrease it.
• Impact updates performed the best of any task we tested, with 42.8% saying their trust would go up and only 18% saying it would go down.
• Appeals for donations had the narrowest margin of the seven tasks, and even there, more donors said their trust would rise (34.3%) than fall (26.5%).

So there you go. Your donors aren’t waiting in the bushes outside your office to whip out their homemade badges and arrest you for AI use. And if that means you’re going to start using AI prompts even more than you did before … then these five habits are exactly what you need. 

1. Give It the Context It Can’t Possibly Have

A language model knows a great deal about the average nonprofit. But it knows next to nothing about your nonprofit. 

An LLM doesn’t know that your spring event is a pancake breakfast, that your board chair hates exclamation points, or that the donor you’re writing to has given $50 every February since 2019 and always mentions her late husband in the memo line. 

And without any of that detail, the model will do what it’s designed to do: write the most statistically average nonprofit language available. That’s where the flat, could-be-anyone tone that we’re all getting better at recognizing comes from—not to mention the odd AI-isms like “It’s not this, it’s that” and vocab choices that you keep having to edit out.

So, instead, front-load your prompt with details. Specifically, a well-written AI prompt will always include these four angles:

• Say who you are in one line, using your mission rather than a tagline. A tagline might be fun and pithy, but your mission statement will contain way more actual context for the model to work from.
• Say who’s reading this and what they already know about you. A monthly donor in her third year and a prospect who found you through a Google search just yesterday need completely different emails.
• Give it the one concrete, current detail that this message will rely on. That could be a goal amount, a date, a particular program—whatever it is, it should be real, relevant, and specific.
• Tell it what you want the reader to do next. Should they make a gift, take a survey, or register for an event, etc.?  If the answer is “nothing,” say that too

Compare your basic “write a thank-you email for a $250 gift” prompt with this one:

We’re a youth arts nonprofit in Eugene, OR. I’m thanking a recurring donor who gives $25 a month and has for three years. Her gifts support our after-school ceramics program, which served 84 kids this spring. I want her to know the program is expanding to a second site in the fall. No ask in this email.

Same task, wildly different output. 

Just Don’t Ever Put Donor Data in an Unapproved Tool

Since we’re telling you to give your tool specific context about the person receiving a given message, we also have a duty to say this: You should never paste real donor data into a general-purpose AI tool. 

Instead of information like full name, mailing address, email address, and gift history, use placeholders instead ([DONOR NAME], [GIFT AMOUNT], [PROGRAM]). Let the AI write the structure, and fill in the real details yourself afterward. 

So what’s an unapproved tool? At the ground level, it’s anything you’re using under a personal account. Even if you click the box that says “don’t use my data to train your model,” that is not enough to protect your supporters’ data.

If your org wants to use a tool like, say, Claude, then the org needs to sign up for a business or enterprise account. Those subscriptions come with data security agreements that personal ones do not. 

At the next level, both the tools that are approved for donor data and the ones that are not should be outlined in your nonprofit’s AI policy. If your org has yet to write one of those, then—good news—we can help you get a real head start. 

Just fill in and download our AI governance policy template for a complete first draft.

2. Make It Ask Questions Before It Writes Anything

This is the habit that’s going to improve your prompting more than anything else—and it’s because it works on you, not the AI. 

Whenever you’re writing a prompt, be sure to include this: “Before drafting, ask me for everything you need. Ask all your questions at once, then wait for my answers.” 

When you do this, your AI chatbot becomes more than a tool for simply spitting out first drafts. It’s now more like a colleague that is poking and prodding your communications strategy to make sure there aren’t any holes.

For example, writing a year-end appeal might spark a question about which specific program outcome from the past 12 months the letter should point to. If you don’t have an answer, this is your cue to go find one. 

It also means you’ll have to deal with fewer placeholders to fill in later. A model left to guess will cheerfully write “your gift helped us serve [X] families this year” and leave you to go find X. A model that asks first for the real number will just include it in the draft. 

This is a key pattern behind every prompt in our library. Each one opens by asking you for critical information, waits for real answers, and only then starts to draft. 

3. Stop It From Guessing.

Left to their own devices, AI tools will fill any gap you leave. They don’t experience uncertainty the way we do, so they’ll invent program outcomes, manufacture quotes, and hand all of it back to you in a paragraph so tidy and well organized that you’ll read straight past it the first time. And possibly every time after, too.

Unfortunately, people tend to really, really dislike it when basic info in your communications is straight up false. For a social post, that can be an embarrassing annoyance. In an annual report, a grant narrative, or an impact story about a real person, it could very well cost you a funder relationship or a participant’s trust.

So, even though it might feel a little silly, write the prohibition into the prompt:

Do not invent details, quotes, statistics, or outcomes that I haven’t given you. Ask me for anything essential that’s missing instead of using a placeholder. Flag anywhere you’re inferring something rather than stating what I told you directly. Use only the numbers I provide, and don’t estimate, extrapolate, or round.

That last part—”don’t estimate, extrapolate, or round“—is crucial. Your real numbers are almost always more persuasive than invented ones, because they’re specific and even slightly odd in the way true things are. “84 kids this spring” sounds true. “Hundreds of young people each year” sounds … less so.

One note on phrasing all of this. Prompts work better when you describe what you want than when you list what you don’t, since telling a model to avoid a concept puts the concept right there in its attention. 

Lead with the positive version and save the explicit bans for the handful of cliched phrases or AI tics you never want to see again. 

4. Explain What “Done” Looks Like

If an AI tool has no idea what the finished product should look like, it’s going to do what it does best (see the previous section) and just make one up. If you don’t tell it you want only 150 words, you could very well end up with 600. If you don’t say what shape, you’ll get an essay when you want bullets; and so on.

Be boringly explicit about all of it:

• Length. Give it a number. “Under 150 words” is far better than “keep it brief.” 
• Format. Name the thing you want. For example: an email with a subject line, three labeled sections, bullets with no paragraphs underneath them. Even better, you can specify content flow, like “open with the outcome, then the ask, then one clear next step.” 
• Tone. Don’t just throw out an adjective and call it a day, because telling a model to be “friendly” will only produce the average of every friendly sentence ever written. Give it a situation instead: “Warm and specific, the way you’d write to someone you’ve met twice.” 
• The ask. Pick the one thing you want this person to do and tell the model to ask for that and nothing else. Leaving that part of the message open-ended can get you back a call to action buffet: a donate link and a volunteer sign-up and an invitation to follow you on Instagram.

This is also where you set the reading level. For board members and funders, you can use slightly more technical language. But for anything a client or a supporter reads, tell the model “plain language, sixth-to-eighth-grade reading level, short sentences, no acronyms without explanation.” 

5. Turn It Loose on Its Own First Draft

Close the prompt by pointing the model back at what it just wrote: “After your first draft, review it against the rules above, cut anything generic or overlong, and give me the tightened version.”

This gives you an automatic second pass. It won’t catch everything, but it reliably catches the throat-clearing opener, the two sentences that say the exact same thing in different words, and the closing paragraph that unnecessarily summarizes the email the reader just finished reading.

Of course, the next step is for you to do a pass as well. The final draft has to sound like you, and the more important a message is, the more of your actual words should be in it. 

This is the one thing donors in our survey were unambiguous about. 71.6% said it’s moderately or very important that a human being is involved in creating the messages they get from a nonprofit, and only 9.1% said it didn’t matter to them either way. 

Nobody is asking you to write every sentence from scratch. They want you in the room while it gets written. If you’d rather not rely on your own eye for that last pass, our AI validation checklist gives you something to check the draft against.

How Writing Better Prompts Can Lead To Stronger Relationships

It’s tempting to measure the effectiveness of how you’re using AI at your nonprofit by volume: more emails, more posts, more drafts per hour. But that’s not actually the metric you should be using. You should instead be judging it by the amount of time you’re spending with your supporters—is AI leading to more time, or less—because it’s those personal connections that are going to turn into long-term support. 

And when it comes to assessing lifetime donor value, it’s longevity, not the size of their first gift, that really matters.

For The Generosity Report, we tracked 99,522 donors across 2,748 nonprofits over five years. And we found that donors who gave all five years made up 45% of overall revenue, even though they only accounted for 11.7% of the panel. First-gift size, meanwhile, told us almost nothing about who would stay. 

Here’s an interesting point from the Generosity Report that should affect how you plan your communications (including your use of AI prompts): more contact did not translate into more support. Multi-org donors got 55 emails across all their organizations; single-org donors got 59 from one, and their giving grew at a slower pace than multi-org donors.

So the case for better prompts isn’t about speed, and it isn’t about increasing the number of messages you send. It’s about consistently creating and sending the kind of specific, obviously about them messages that foster relationships with your donors even as you have a million and one other pressing issues on your plate.

The time you save from using well-written AI prompts will be incredibly valuable. But if you want to use it to its fullest potential, you’ll spend it hopping on the phone, going out to get coffee, and just generally trading time in front of a screen for time in front of your donors.

Neon One’s AI Prompts Come With Best Practices Included

Guess what? You don’t have to take all the advice in this article and apply it to your prompts by hand. Neon One’s AI Xchange hub contains a whole library of 17 ready-to-use prompts, sorted into fundraising, marketing and communications, reporting, and operations, that have all these good habits built in. 

Start with one task you repeat every week: a thank-you email, a board meeting summary, a set of social captions. Paste the prompt in, answer what it asks, edit what comes back, and spend the hour you got back on somebody in your community. 

Frequently Asked Questions About AI Prompts for Nonprofits