AI Can Draft Your Appeal. It Can't Tell You What Your Donors Care About.

If you work in a small development shop, you've probably already asked an AI tool to write something for you. You're in good company. In a December 2025 survey of 346 nonprofits by Virtuous and Fundraising.AI, 92% said they use AI tools in some capacity, and donor communications was the most common use (62%). The same survey found that 47% of organizations have no AI governance policy.

So most of us are using it, and about half of us are making up the rules as we go.

I think AI is genuinely useful for fundraisers with more work than hours. But the way most people use it puts the machine in charge of the wrong part of the job. AI is good at producing words. It has no idea which words matter to your donors, because that knowledge lives in your donor file, your program staff, and your conversations with the people who give. The workflow that works keeps the knowing with humans and hands the typing to the machine.

Where AI actually saves time

For a one- or two-person shop, these are the places I'd use it without much hesitation:

First drafts from a real brief. Staring at a blank page is where a lot of appeal-writing time disappears. If you give the tool a solid brief (more on that below), it can hand you a workable draft in a minute. You'll rewrite a good portion of it, but rewriting is faster than starting.

Segment variations. Once you have a strong master appeal, AI is very good at adapting it: a version for lapsed donors that acknowledges they've been away, a version for monthly donors that doesn't ask them to make an extra gift as if they've never given, a version for first-time donors. That's the kind of work small shops usually skip because there's no time.

Cleanup. Shortening a 900-word letter to 600, simplifying long sentences, drafting ten subject lines so you can pick two to test. Tedious work, done quickly.

Summarizing your own material. Program reports, staff updates, and interview transcripts can be long. AI can pull out candidate stories and details so you know who to follow up with. (See the privacy section before you paste anything in.)

Where it hurts

Generic voice. Left to itself, AI writes the average of every fundraising letter it has seen. You get "our dedicated team works tirelessly" and "your generous support makes a difference." Donors have read those sentences hundreds of times. They don't register.

Invented details. This is the one that worries me most. When a model doesn't have a detail it needs, it often supplies a plausible one: a child's name, a statistic, a quote from a volunteer. In a blog post about productivity, that's embarrassing. In a fundraising appeal, it's a donor reading a story that didn't happen, with your signature at the bottom. If you've read about the story crimes that cost nonprofits trust, invented details belong near the top of the list.

Donor data privacy. Don't paste identifiable donor information (names, addresses, gift histories, notes from a major donor visit) into an AI tool unless you understand that vendor's data terms and your organization has a policy that allows it. Terms differ between tools, and often between free consumer accounts and paid business accounts, on questions like whether your inputs are retained or used to train models. Read the actual terms for the account you're using. When in doubt, strip the identifying details. "Donor A, gave $250 three years in a row, lapsed in 2025" gets you nearly the same help as the donor's real name.

Losing the donor's perspective. AI writes from whatever point of view you feed it. If your inputs are internal program reports, the output will sound like an internal program report: a lot of "we" and "our," not much "you." The draft becomes about the organization's needs instead of what the donor makes possible.

A workflow with human checkpoints

Here's the process I'd recommend for an appeal or a donor email. The AI handles two of the seven steps. Humans own the other five, and those are the ones that make the letter work.

1. Human: write the brief. One page. Who is receiving this, and what did they last give to? One true story, with details confirmed by the staff member closest to it. The specific ask and what a gift accomplishes. Checkpoint: could someone on your program staff confirm every fact in this brief?
2. AI: draft from the brief only. Include this instruction in your prompt: "Use only the facts in this brief. If you need a detail I haven't provided, write [NEED: description] instead of inventing it." It won't stop every invented detail, but it helps, because it gives the tool a way to admit a gap instead of filling it.
3. Human: fact-check. Highlight every name, number, date, and quote in the draft. Each one should trace back to the brief. If it doesn't, cut it or verify it. Don't assume the tool copied numbers correctly, either.
4. Human: read it as the donor. Pick a real person on the list and read the letter as if you were them. Count the "we" and "our" against the "you." Ask whether the donor is the one making something happen, or a spectator to the organization's hard work.
5. AI: produce segment versions and cleanup. Now that the master is right, generate the lapsed, monthly, and first-time variations, plus subject lines.
6. Human: check the variations. Variations are where invented details tend to sneak back in, so give each one a quick version of step 3.
7. Human: read it aloud and sign off. The person whose name is on the letter should read the final version out loud. If they wouldn't say a sentence to a donor in person, it comes out.

The first time through, this takes longer than accepting a draft as-is. After a few rounds, the brief gets faster to write and the drafts need less work, because you're feeding the tool better material.

What a weak draft looks like, and how to fix it

Take a hypothetical county animal shelter writing its year-end appeal. Someone types "Write a year-end fundraising appeal for our animal shelter" and gets this:

Every day, our dedicated team works tirelessly to care for animals in need. Last month, a puppy named Max arrived at our shelter scared and alone. Thanks to generous supporters like you, Max is now thriving in his forever home. Your gift today will help us continue this important work.

It reads fine. It's also unusable. Max doesn't exist; the tool made him up. The first sentence is about the staff. And "this important work" asks the donor to fund something vague.

Now the same shelter writes a brief: a real dog (a senior shepherd mix the staff named Juniper, found tied to a trailhead sign in August), the real medical need (heartworm treatment), the real cost the shelter's vet confirmed for one stage of treatment, and the instruction to use only those facts. After a human edit, the opening reads:

In August, a hiker found a gray-muzzled shepherd mix tied to a trailhead sign. She's about nine, and she has heartworm. Our staff named her Juniper. Treatment comes in stages, and each one costs money we haven't raised yet. A gift from you before December 31 can pay for her next stage, and for the next dog who arrives the same way.

Every detail is checkable, and the donor is the one who acts. The AI still saved time: it drafted the structure and the segment versions, and the humans supplied what only they could know.

Start with a half-page policy

Don't build a full AI strategy before your next appeal. Write a half-page policy instead: what donor information never goes into an AI tool, which accounts staff may use, and who reviews AI-assisted copy before it goes out. Then try the brief-first workflow on one email.

If you want a fuller picture of what an AI-first approach to fundraising can look like, read Vibe Fundraising. And if your brief keeps coming up short on real stories, the 10-minute story interview is a simple way to get them from your program staff. AI can't do that part for you.

Want AI to do more of the heavy lifting? Build Your AI Fundraising Assistant and learn how to turn AI into a fundraising teammate that knows your donors and your voice.