92% of nonprofits use AI. Only 4% have a process.
A benchmark survey of 346 nonprofits found near-total AI adoption and almost no process behind it. The numbers hold up. The company that published them also sells the fix.
After reading this you can tell whether your team is in the 4% with its AI use written down or the 81% improvising it, do the twenty-minute version of closing that gap, and read the underlying report knowing who paid for it.
Virtuous and Fundraising.AI surveyed 346 nonprofit organizations in December 2025. Of those, 92% are using AI in some capacity. Just 7% report major improvements in their ability to achieve their mission.
Most of the coverage stopped there and called it a disappointment. The more useful number is on page 12.
Only 4% have documented, repeatable workflows. Another 81% use AI individually, on an ad hoc basis.
Who paid for this
Virtuous sells fundraising software with AI built into it. Fundraising.AI is an independent collaborative. So the finding that nonprofits need structured AI adoption comes, in part, from a company that sells exactly that. The report closes with a chapter on Virtuous products and a link to their policy template.
That conflict doesn’t make the survey wrong. Every figure below was checked against the report PDF rather than the press release, and they all hold up. So read the recommendations knowing who is making them, and judge the counts on their own.
The data is still worth your time, because almost nobody else is collecting nonprofit-specific numbers on this at all.
Three things to know before you quote any of it:
- “AI” here means generative AI. ChatGPT, Claude, Copilot, Gemini, and AI features inside software you already pay for. Anything that was mainly automation got excluded, so that 92% is narrower than it sounds.
- The sample skews small. Nine in ten respondents have budgets under $20M a year, and more than a third under $1M. That’s good news if you’re a small shop. These numbers describe organizations like yours rather than the sector’s giants.
- It leans fundraising. Fundraising leaders made up 31% of respondents, ahead of executive leadership at 24% and operations at 18%.
What “using AI” actually looks like
The survey asked how AI gets used across the organization. Respondents could check more than one box, so these overlap and don’t add to 100.
- 65% reactive and individual: one-off prompts, personal experimentation.
- 18% operational: team workflows and shared prompts.
- 7% strategic: AI built into goals, budgets, and the numbers they report on.
And 47% have no AI governance policy at all.
Put it together and you get a picture most people will recognize. Someone in your development office worked out a prompt that turns messy call notes into a usable donor update. It saves them an hour a week. Nobody else knows it exists. When they leave in March, it leaves with them.
The report is blunt about the cost. Knowledge walks out the door with people. Teams solve the same problem twice. New staff start from zero.
For organizations that haven’t started, lack of training is the top barrier, cited by 48% of them. But for the ones already using AI daily, the barriers change: privacy and security at 32%, time at 31%, staff skepticism at 19%. Training isn’t what’s stopping them. At 92% adoption the question of whether to use this stuff is settled, and what’s missing is everything after that.
What the 4% have
What they have is a document. Someone wrote the steps down. Which tool, which prompt, what a person checks before it goes out, what to do when it comes back wrong. It might be one page in Google Docs. That’s the whole differentiator.
Once it’s written down, you can hand it to the next person, fix it when the model changes underneath you, and argue about whether the check is strict enough. None of that works on a habit living in one person’s head.
Size helps less than you’d expect. Organizations under 50 staff report moderate impact slightly more often than large ones, 41% against 34%, despite having less of everything. The report attributes that to complexity rather than money.
The twenty-minute version
Pick the AI task somebody on your team already does most often. If this survey is any guide, that’s donor communications: 62% of organizations use AI there, more than anywhere else.
Write down four things:
- The prompt itself, copied exactly. Paste the real thing, however scrappy it looks.
- Which tool, which model, and which plan. All three change the answer you get.
- What someone checks before it goes out. Be specific. “Looks fine” isn’t a check.
- The last time it was wrong, and how you found out.
That’s it. Four bullet points in a shared doc, and by this report’s own definition you’ve moved from the 81% into the 4%.
Policy takes longer, though probably less time than you think. For the 47% without one, what’s missing is usually a decision rather than a document: what staff are allowed to paste into a chat window. That can be settled in a staff meeting well before anyone writes anything formal.
Why this publication exists
This is the first edition, so here’s what the rest will be.
The 4% finding is dull, and it’s the entire reason this publication exists. What separates those organizations is that somebody wrote down what works. Writing it down is the part nobody has time for.
Every edition is one of these three:
- Playbooks
- One job, run end to end on real work, including where it broke.
- Risks and red flags
- What breaks, what leaks, and which claims don't hold up.
- News that matters
- Sector news, and what it changes for your team.
This edition is the third. Most weeks it’ll be the first.
A playbook means one job, run end to end on real work. A grant report. A batch of thank-yous. A board summary. What arrives on Tuesday is the written-down version: the prompt, the model and plan it ran on, what it cost, how long it took with the editing included, and where it broke.
The failures matter most, and they’re the part you can’t get from a product page, which is why nothing here gets written up before it gets run. You should be able to paste any of it straight into your own shared doc and skip a year of working it out yourself.
Which, by this survey’s own count, puts you ahead of 96% of the organizations in it.
Sources
The 2026 Nonprofit AI Adoption Report, Virtuous and Fundraising.AI, published February 16, 2026. A benchmark study of 346 nonprofit organizations surveyed in December 2025.
Every figure above comes from the report itself, pages 9, 10, 12, 13, 14, and 15, rather than from secondary coverage. Virtuous hosts it on their own site, which asks for your details before the download, and also publishes an interactive version. NonProfit PRO covered the release.