Nonprofit AI Adoption Statistics

Nonprofit AI statistics are unusually unreliable, for a reason worth stating before any of the numbers. Almost every figure in circulation comes from a survey run by a company selling AI software, distributed to that company’s own audience, and answered by people who opted in to a survey about AI.

That does not make the findings wrong. It does mean the provenance is part of the fact, and this page gives it every time.

Adoption

The most cited source for 2026 is a report published by Virtuous and Fundraising.AI, surveying 346 nonprofits with fieldwork in December 2025. Virtuous sells an AI fundraising product.

The adoption figure is quoted as both 74% and 92% across the report’s own summaries, which is a good illustration of why a single headline percentage should be treated carefully. The difference is what counts as using AI. Count any individual staff member experimenting once and the number is very high. Count organizational use and it falls.

The finding that survives either framing is the gap between adoption and effect. Somewhere between six and seven percent of organizations report that AI has made a major strategic difference, against three quarters or more who use it. Almost everyone has tried it. Almost nobody has changed anything structural.

Where it is actually used

Writing dominates, which matches what the tools are actually good at and what costs nothing. This is worth holding in mind against the marketing for donor intelligence platforms: the sector’s real AI activity is overwhelmingly drafting, not prediction.

Governance, which is where the gap is widest

Nearly half of organizations using these tools have written nothing down about how they may be used. Set that against the adoption figure and the position is that most of the sector is using AI without having decided what may go into it.

That is the single most actionable number on this page, because it is the cheapest to fix. A policy costs nothing but an afternoon and a board vote. We publish a free AI use policy template written to be adopted as it stands.

Maturity

Three fifths of the sector is in the middle band: using these tools routinely, with nothing written down and no way of telling whether they are helping. That is the condition in which people quietly paste donor lists into consumer chatbots, and it is also the condition in which software is bought on impression rather than evidence.

What funders think about AI written applications

This one matters commercially and comes from Candid rather than a vendor.

Two thirds of funders have not decided. That is a live risk rather than a settled position, and it argues for checking each funder’s rule per application and assuming disclosure is expected where a funder is silent. A mismatch discovered after an award is a serious problem with a relationship you depend on.

Capability and budget

Seventy percent want the benefit, sixty percent cannot evaluate what they are being sold, and four percent have budgeted to learn. That combination explains a great deal about how nonprofit software is currently bought.

The context these numbers sit in

One figure from outside the AI literature is worth attaching, because it underlies most of the case made for donor intelligence software. A peer reviewed study of 1,663 US fundraisers found median tenure in the current job of 2 years and mean tenure of 3.6 years, with 20 percent intending to leave their organization within the year.

Note what that is not. It is not the 16 month figure quoted throughout the sector, which no peer reviewed source supports. If you are citing fundraiser turnover to justify a purchase or a grant application, cite the real number.

Why the adoption and impact figures are so far apart

Three quarters or more of organizations use these tools. Six or seven percent report a major strategic difference. That gap is the most interesting thing in the data and the survey’s own breakdown explains most of it.

Seventy three percent of use is individual and ad hoc. One person finds a tool useful, uses it for their own work, and nothing about how the organization operates changes. Twenty seven percent have documented, repeatable workflows, and it is overwhelmingly that group reporting more than marginal benefit.

Put plainly, the constraint is organizational rather than technological. An assistant that saves a communications officer four hours a week produces four hours, not a transformation, unless somebody decides what those hours are now for. Most organizations have not had that conversation, which is why the modal result is faster rather than different.

This has a direct implication for buying. If ad hoc individual use is producing marginal returns, the fix is not usually a more expensive tool. It is deciding which specific recurring task is being handed over, who owns it, and what the freed capacity is redirected to. That is free, and it is the step almost nobody takes before spending.

What to measure in your own organization

Sixty percent of nonprofits say they cannot assess AI tools. Most of that gap closes with four numbers you can collect yourself, without expertise.

The third row is the one to run before any purchase decision and again after ninety days. A tool with a low error rate on your own data has earned trust on the material you cannot check. One that gets three in twenty wrong on verifiable cases is not safe on anything else, whatever the survey figures say about the category.

How to use these figures

Three cautions, in order of importance.

Name the publisher. When you quote the adoption figures in a board paper, say they come from a vendor survey of 346 self-selected organizations. It changes how a trustee weighs them, correctly.

Do not compare across sources. These surveys ask different questions of different populations. Movement between one study and another is not a trend.

Watch the date. The fieldwork here is December 2025. In a category changing this fast, a figure eighteen months old describes a different market.

What this data does not tell you

Worth knowing before anyone builds a strategy on it. There is no reliable breakdown by organization size, which is the variable that matters most, since a $400,000 food bank and a $40m health system are both counted here as one nonprofit. There is no comparable prior year study from the same sample, so nothing on this page is a trend. And self-reported impact is not measured impact: the organizations claiming major benefit were not audited, they were asked.

Questions people ask

What percentage of nonprofits use AI?

Between 74% and 92% as of December 2025, depending on how the question is framed, from a survey of 346 organizations published by Virtuous and Fundraising.AI. Virtuous sells an AI fundraising product, and the sample was self-selected, so treat these as indicative rather than representative of the sector.

The wide range is instructive. Counting any individual who has experimented produces the high number. Counting organizational, repeatable use produces the low one, and only 27% report documented workflows.

Is AI actually working for nonprofits?

Mostly at the margins. In the same survey, 6% to 7% reported major strategic impact, 39% moderate improvement, 40% small improvement and 14% no impact yet.

The pattern is an efficiency gain rather than a transformation: things get faster, and the organization does not fundamentally change. That is a reasonable outcome and it is not what the marketing describes.

The organizations reporting more than marginal benefit are disproportionately those with documented workflows rather than ad hoc individual use, which suggests the constraint is organizational rather than technological.

How many nonprofits have an AI policy?

About half. In the December 2025 survey, 47% had no AI policy at all. Of those that did, 23% described it as cautious, 19% as enabling and 6% as restrictive.

Set against adoption figures of 74% or higher, that means a substantial share of the sector is using these tools with nothing written down about what may go into them. This is the cheapest gap on the list to close, and we publish a free policy template built for it.

Do foundations accept AI written grant applications?

Most have not decided. Candid found 23% of foundations will not accept AI generated applications, 67% are undecided or have no policy, and 10% will accept them.

Practically, check each funder's stated rule for every application, and where a funder is silent assume disclosure is expected. The downside of disclosing unnecessarily is close to zero. The downside of a mismatch discovered after an award is a damaged relationship with a funder you depend on.

Regardless of disclosure, every figure, citation and factual claim in an application must be verified at source. Fabricated references in a grant application are the worst version of this failure.

Are nonprofit AI statistics reliable?

Treat them with care. Most published figures come from surveys run by companies selling AI software, distributed to their own audiences, and answered by people who chose to respond to a survey about AI. Each of those three things pushes the numbers up.

They are still the best available evidence and they are useful for direction rather than precision. When you quote them, name the publisher and the sample size. A trustee weighing a purchase should know that an adoption figure came from a vendor survey of 346 self-selected organizations.

What do nonprofits use AI for most?

Writing. Content and communications lead at 68%, followed by data analysis and reporting at 42% and operations at 24%.

That matches both the capability of the tools and their cost. Drafting is the thing general purpose assistants do well at a free tier or about $20 a month, and it is where most organizations should start.

It is also worth setting against the marketing for donor intelligence platforms. The sector's actual AI activity is overwhelmingly drafting, not prediction, and the two carry very different price tags.

How much do nonprofits spend on AI training?

Almost nothing. Only 4% report an AI specific training budget, while 60% say they lack the in-house expertise to assess AI tools and 70% believe AI could reduce their workload.

That combination is the underlying problem. Organizations that want the benefit and cannot evaluate what they are being sold, with no budget to learn, buy on impression. Several hours of structured training across a team is cheaper than any of the software on the market and will change more.

Do fundraisers really leave every 16 months?

No. That figure is repeated constantly in this literature and no peer reviewed source supports it.

A study of 1,663 US fundraisers by Shaker and colleagues found mean tenure in the current job of 3.6 years, a median of 2 years, and mean tenure across all fundraising jobs of 3.9 years, with 20% intending to leave their organization within the year.

The turnover concern is real at the accurate number, because major gift relationships routinely take longer than two years to mature. Use the real one.