Search for AI training for nonprofits and page one is a catalog. Nine listings: two free courses, three priced from $500 to $1,500, two that publish no price, and two round-ups. Every page prices the course. None prices the thing your development shop is short of, which is the two people in it.
That omission explains the sector’s headline number. The 2026 Nonprofit AI Adoption Report from Virtuous and Fundraising.AI, built on 346 organizations surveyed in December 2025, found 92% of nonprofits using AI and 7% reporting major improvement in their ability to achieve their mission. Courses are not the scarce resource. Hours are, and so is a shared method: the same report puts 81% of use on an ad hoc basis and 47% of organizations with no governance policy.
What the market sells
Every price below was read off the provider’s own public page in October 2026. These catalogs change at least annually, so re-check before quoting a figure internally.
| Option | Price as listed | Time |
|---|---|---|
| AI Fluency for Nonprofits, Anthropic Academy with GivingTuesday | Free | 6 hours, 5 modules |
| AI Skills for Nonprofits, NetHope with Microsoft | Free | Self-paced |
| Google for Nonprofits training | Free | Self-paced |
| AI for Nonprofits, UT Austin | $595 | 2 days, instructor-led |
| Maximizing AI for Nonprofit Efficiency, UW–Madison | Not published | 12 hours, hybrid |
| Leading in the Age of AI, Kellogg | $1,500 | 4 live half-days, 12 CFRE credits |
| AI for Nonprofits Certificate, NTEN | $500 member, $1,000 non-member | 13 one-hour courses plus an assignment |
| A facilitated session on your own grant files | $2,000–15,000, market range | Half day to two days |
Read that as four products, not eight competitors. The free courses teach a person what these tools are. The university options add an instructor and a schedule, and the certificate adds a credential. A private session is the only one that opens the proposal you are writing this month. Buyers compare all four on price and pick the cheapest, which is sound only if they were substitutes. The same confusion recurs function by function in our role-by-role view of what each team needs.
The number nobody on that page prices
The Bureau of Labor Statistics puts the median wage for fundraisers at $34.88 an hour in May 2025 across 140,900 jobs. Its June 2026 release puts benefits at 31.5% of total civilian compensation, so a loaded development hour is roughly $51. Two full-time staff carry about 4,160 scheduled hours a year between them, near $212,000 of capacity, and that is the budget line every training decision draws on.
| Option | Invoice | Staff hours | Cost of those hours | Loaded total |
|---|---|---|---|---|
| Both staff, free 6-hour course | $0 | 12 | About $610 | About $610 |
| Both staff, NTEN certificate, member rate | $1,000 | About 30 | About $1,530 | About $2,530 |
| One seat, Kellogg, 4 half-days | $1,500 | About 17 | About $870 | About $2,370 |
| One facilitated half day for both, two follow-ups | $2,500, low end of market | 12 | About $610 | About $3,110 |
Two things fall out. The free course is not free: six hours for two people is about $610 of capacity, worth spending and worth a budget line so nobody treats it as having cost nothing. And the certificate is the dearest option once hours are counted, while being advertised as nearly the cheapest. Thirteen hours plus an assignment, twice over, is $1,530 of fundraising time in the season one of them writes the year-end appeal.
The spread between cheapest and dearest is about $2,500 loaded, against $212,000 of capacity. The expensive mistake is not the wrong course. It is spending any of those hours without a shared method coming out the other end.
What two people can cover
A two-person shop is usually a development director and a grants coordinator, and between them they own the grant pipeline, individual giving, one or two events, the database, board support, and impact reporting. Larger teams split that among specialists; yours cannot. The pattern is narrow: these tools extend tasks where a source document exists and a human checks the output, and do nothing where the input is a relationship or the answer must be certified.
| Work in a two-person shop | What a trained pair can cover | Where the hour still goes |
|---|---|---|
| Grant narrative, first draft | Assembled from your boilerplate against the funder’s questions | Checking every number |
| The same narrative for a second funder | Reshaped to new questions and word limits, where most of the saving sits | Judging what this funder wants |
| Funder guidelines and RFPs | Summarized into a brief, with eligibility and deadlines pulled out | The go or no-go call |
| Interim and final reports | Drafted from your program notes and output data | Verifying before signature |
| Appeal and segment variants | Three or four tone and length variants from an approved base | Segments and ask amounts |
| Acknowledgments and stewardship letters | Templates that read like a person wrote them, by gift level | The handwritten line |
Price your own version: if your last full foundation proposal took 30 hours across the two of you, that is about $1,530 of capacity, and a realistic first-year gain is a share of the drafting portion, not the whole 30. The honest pitch is not a third hire. It is roughly one more funder relationship’s worth of written output a quarter, often the difference between a pipeline that grows and one that holds. The same arithmetic in a communications function lands the same way, as our look at in-house marketing teams shows.
Where it stops
Four categories, and a session promising to move any of them is overselling.
The ask. The visit, the call, the close. Nothing in a chat window moves a major gift conversation, and the hours freed by faster drafting are worth having because they go here.
Donor or client records in a tool you have no agreement with. Giving histories, beneficiary case notes, anything under HIPAA if you deliver health services or FERPA if you run education programs, plus confidentiality clauses in your grant agreements. The guardrails are mostly missing: NTEN and the Bridgespan Group surveyed 917 nonprofit staff and executives in summer 2026 and found rules on AI data entry in place at 39.95% of organizations and staff training at 18.41%.
Capacity and wealth judgments. Ask a general-purpose assistant what a prospect can give and you get a confident number with nothing behind it. Prospect research is a licensed-data problem.
Certification. Someone signs the grant report and the 990. A draft stays a draft until a human has checked every figure against the system of record, and the signer is accountable either way.
The first dollar is not a course
Two credible free curricula exist, so a shop that has spent nothing is not short of instruction. It is short of two things, both prerequisites to any paid session paying for itself.
A one-page data rule. Named categories, not a general instruction to be careful: which document types may be pasted into which account, what never leaves your systems, who decides the unclear cases. Three quarters of organizations in the 2025 TechSoup and Tapp Network benchmark had no AI strategy, and at two people you need the page, not a strategy. Why written policy and real behavior come apart is the subject of our guide to safe use.
A boilerplate library the tools can work from. Mission language, program descriptions, outcome data by year, past successful proposals, budget notes, bios, in one place and current. Most shops have this scattered across four drives and three inboxes, and assembling it is the highest-return day of work available, because every drafting gain above depends on it.
Do those first. If the pair then work through the free six-hour course with the library open and a live proposal in front of them, a paid session may prove unnecessary this year, and no provider will tell you so. A session earns its price in one case: both have tried the tools, both hit the same wall, and neither can judge whether the output is good. More self-paced video does not fix that, and it is the gap between access and capability we set out in why licenses are not capability.
What a two-person shop should not buy
A thirteen-course certificate for both people. About $1,530 of fundraising time on top of $1,000 in member tuition. Send one if a credential matters to your board, and have them brief the other.
Predictive analytics or wealth-screening modules. Built for files with tens of thousands of records and a data team. On a few hundred active donors the model has nothing to learn from.
An AI strategy engagement, or a platform subscription bought before a method exists. At two people the strategy is the one-page rule plus a decision about which part of the grant cycle comes first. With 81% of sector use ad hoc, another tool does not fix it, and 48% of non-users named lack of training as the barrier.
A session that will not open your own documents. A feature tour and an ethics slide are free in six hours.
What we would do with a two-person team
Assume 180 staff, two in development, four foundation deadlines a year, and a year-end appeal.
- One day assembling the boilerplate library. About $410 of staff time, and the prerequisite for everything else.
- The one-page data rule, written by the executive and development directors together. Two hours: donor records, client case notes, unreleased financials, anything a grant agreement restricts.
- Both people through the free six-hour course, library open, a live proposal in front of them. About $610 of capacity, not passive viewing.
- One facilitated half day, only if step three leaves them unable to judge their own output. The agenda is your next proposal, your last funder report, your appeal.
- Two one-hour follow-ups at three and six weeks, each bringing something that went wrong. About $200 of time, and where the method survives or lapses.
- A named owner with two hours a month, here the development director, with written sign-off that those hours exist.
Loaded cost lands near $1,200 if you stop after step three and near $4,300 if you add the session and its follow-ups, against $212,000 of annual capacity. The board is a separate audience: oversight questions are not a working method, which is the case for treating leadership separately made in our view of executive and leadership programs.
How to tell in ninety days whether it worked
Pick the measures before anyone is trained, while nobody has an outcome to defend. A week of rough timekeeping beats a number reconstructed later.
- Hours from funder guidelines to a complete first draft. Same proposal type, same two people. The number that should move.
- Whether the data rule is live. Ask both separately what may be pasted into an assistant. Different answers mean it did not land.
- One accuracy check. Verify every figure in every AI-assisted document that went to a funder. One unverified number reaching a program officer costs more than the training.
- Whether the pipeline moved. One more qualified funder approached, or a report filed early.
Self-reported confidence is deliberately absent. Survey scores measure the session; these measure the organization. Turning a plausible gain into one you can defend at budget time is the subject of our guide to proving capability moved.
Frequently asked questions
How much does AI training for nonprofits cost?
Free to $1,500 for published options. The Coursera course from Anthropic Academy and GivingTuesday and the NetHope and Microsoft course on data.org cost nothing. NTEN’s certificate is $500 for members, UT Austin lists $595, Kellogg $1,500. A private facilitated session runs $2,000 to $15,000 at market rates. Add staff time at roughly $51 an hour, which is why a certificate is worth one seat and not two.
Are the free courses enough for a two-person development team?
Often, on two conditions: they work through the material with their own boilerplate library and a live proposal open rather than passively, and someone has written the one-page rule on donor and client data first. If both finish and still cannot judge whether an output is good enough to send, a paid session is worth it.
Who should be trained first in a small nonprofit?
The two who write the most, which in a development shop is the director and the grants coordinator: grant narratives, funder reports, and appeal copy all have a source document behind them. The executive director needs the data rule and the measures, a two-hour conversation, not a course.
What can AI realistically do for grant writing?
First drafts from your boilerplate against a funder’s published questions, reshaping a narrative to a second funder’s format and word limits, summarizing guidelines into a go or no-go brief, and drafting interim reports from program notes. It cannot supply outcome data you do not have, judge what a program officer wants, or certify a figure. Roughly a quarter of organizations in the 2025 TechSoup and Tapp benchmark were already using AI in grantseeking.
What donor data can we put into an AI tool?
Treat giving histories, contact records, client case notes, unreleased financials, and anything a grant agreement restricts as off limits unless a vendor agreement covers it. Program descriptions, published outcome data, public funder guidelines, and your own mission language are fine. Write the list down: NTEN and Bridgespan found data-entry rules in place at 39.95% of organizations surveyed.
How long should AI enablement for a small nonprofit take?
Six to twelve hours of learning per person, a day assembling the boilerplate library, and two short follow-ups at three and six weeks. A thirteen-hour certificate or multi-month cohort is built for an individual career, not a two-person operating team.
Does a two-person nonprofit need a written AI policy?
One written page, not a policy document: the document categories that may go into a tool, what never leaves your systems, who decides the unclear cases, signed by the executive director. The Virtuous and Fundraising.AI survey found 47% with no governance policy, and at two people the absence shows up as two different private habits.
Can AI replace a third development hire?
No. A trained pair gains back a share of drafting time, which funds about one more funder relationship’s worth of written output a quarter. That is not a salary. The work these tools do not touch, the ask and the stewardship relationship, is what a third hire would do.
Key takeaways
- Page one is a catalog of four products: free courses, instructor-led university courses, a credential, and a private session on your own files. Only the last opens the proposal you are writing now.
- Price the hours or the comparison is wrong. At a $34.88 median hourly wage for fundraisers and benefits at 31.5% of total compensation, a development hour costs about $51, and two staff carry roughly $212,000 a year of it.
- Fluency extends drafting-and-checking work where a source document exists. It does not touch the ask, capacity judgments, restricted data, or certification.
- Spend the first day on a boilerplate library and the one-page data rule. Every drafting gain depends on the library, and 92% adoption against 7% reported impact is what happens without them.
- Skip the wealth-screening module, the strategy engagement, and the second certificate seat. None of the three earns its hours at this size.
If you are weighing a free course against a paid session for a two-person development team, book a free AI readiness call. We will look at your grant calendar and your document categories and tell you whether you need to spend anything.
Arthur Wandzel
