About three in four nonprofits are now experimenting with generative AI, according to Salesforce’s Nonprofit Trends Report. That is a striking adoption number for a sector that historically lags on tooling. The same report flags the constraint that explains everything else about the sector’s AI rollout. Most nonprofits do not have the budget or the training capacity to do more than experiment.
The gap between “we are trying it” and “we have changed how we work” is wider in the nonprofit sector than in any other profession the major reports cover. That gap is the story.
Where small-to-mid nonprofits are actually getting value
The use cases that work for resource-constrained nonprofits are the ones that turn one person’s hour into something closer to half an hour, without requiring custom integration or specialist staff.
Grant writing is the cleanest example. A development director with a draft narrative, a logic model, and a funder’s published priorities can use ChatGPT or Claude to compress, restructure, and tighten an LOI in under an hour, when the same task used to take a working day. The grant officer still has to know the funder, the field, and the program. The model does not invent strategy. It accelerates the version of the strategy the writer already has in their head.
Donor communications and acknowledgement letters are the other obvious win. Mail-merging is old. What is new is being able to draft something that reads as written for that specific donor, referencing their giving history and stated interests, in seconds rather than hours. Most CRMs serving nonprofits (Bloomerang, Salesforce Nonprofit Cloud, Neon One) have shipped or are shipping AI features that pull this off without the user having to know any prompting craft.
Case stories and impact reporting are a more interesting use. Program staff have always struggled to turn raw program data and beneficiary anecdotes into the narrative reporting that boards and major donors expect. AI is unusually good at this, with the caveat that the stories must be reviewed carefully for accuracy and consent. The value is not in the writing itself, it is in compressing the time between “we did the work” and “we have a story to tell about the work.”
Where the bigger gains are stuck behind the budget wall
The use cases that show up in case studies from large nonprofits look different. Predictive donor scoring, segment-level personalisation across email and direct mail, automated stewardship sequences that vary the touch by predicted lifetime value, and program-side measurement that turns service delivery data into outcome metrics in real time.
These are real and they work. They also require either a vendor with serious nonprofit-specific AI capability (DonorSearch, Gravyty, iWave, Virtuous) or a data team that can build them in-house. Both cost money. The smaller the nonprofit, the less likely either is on the table.
This is the structural pattern in the sector right now. Big nonprofits with mature development operations are running AI-enabled fundraising programs that look more like enterprise marketing automation. Small and mid nonprofits are using ChatGPT to draft thank-you letters and getting real but smaller wins. The gap will not close on its own.
The skills problem inside the function
Salesforce’s research keeps surfacing the same constraint, which is staff training. Most nonprofit fundraisers and program staff have had no formal AI training. They are figuring it out from YouTube and from each other. This produces predictable failure modes.
Drafts that read as obviously AI-written, which donors notice and dislike. Prompts that pull “general” output rather than nonprofit-specific output, because the staff member did not know to feed in tone, context, or examples. Privacy mistakes, like pasting beneficiary data into public AI tools without checking whether the vendor uses it for training. Over-reliance on a model’s first draft when a second pass would have caught a factual error or a tone problem.
The nonprofits getting this right are running short, internal training that covers the practical craft. What good prompts look like for the work the team actually does. What never to paste into a public model. How to check the output for accuracy. This kind of training is cheap to deliver and pays back fast. It also tends not to happen because the executive director’s calendar is full of fundraising and the program director’s is full of service delivery, and nobody owns workforce development inside small nonprofits.
What boards and major donors are starting to ask
A pattern has emerged in major-donor conversations and grant applications during 2025 and into 2026. Sophisticated funders are asking nonprofits how they are using AI, what governance is in place, and whether AI use is disclosed in fundraising communications.
This is not gotcha territory. It is funders trying to understand whether the organisation is keeping up with a tool that affects fundraising productivity, donor experience, and the cost structure of the development function. A nonprofit that says “we have not started” is now a different signal than it would have been in 2023.
Boards are starting to ask the same question, sometimes clumsily. The right answer is not to claim a sophisticated AI strategy that the organisation does not have. It is to be specific about what the team uses, what guardrails are in place, and where the next investment goes.
What changes in the next twelve months
Two things move in 2026 and 2027 that nonprofit leaders should plan for now.
The CRMs serving the sector finish embedding AI as a default feature rather than an add-on. Most of the routine fundraising work (segmenting, drafting, sequencing) becomes a checkbox in the platform rather than a separate tool. This will narrow the gap between organisations with AI talent and organisations without, but only for teams that actually configure and use the features.
Major-donor and grant-funder expectations harden. The funders most invested in operational excellence will start expecting some level of AI fluency from grantees, the way they came to expect basic outcome measurement and financial transparency over the past decade. The nonprofits that do not invest in their own teams will fall further behind on the kind of fundraising that requires sophistication.
The work itself is not changing in spirit. Donors still give to people, mission, and trust. But the ratio of time fundraisers spend on the relationship versus the back-end paperwork is shifting in the right direction for the first time in years, and the teams that capitalise on that shift will spend their next decade meaningfully ahead.