AI became familiar.
Donors did not become comfortable.
Three years of asking U.S. donors the same questions about artificial intelligence in the organizations they support. Familiarity rose eighteen points. Willingness to give did not move.
- 3,057 U.S. donors across three annual waves
- 1,020 respondents in 2026
- Fielded September 15–19, 2026
- Self-funded, independent research
Preferred citation: Chappell, N., & Koshy, C. (2026). Donor Perceptions of AI: Third Annual Study. Distributed by Fundraising.AI.
Familiarity and acceptance turned out to be separate judgments
The comfortable assumption in 2024 was that donor resistance was a novelty effect: they had not seen enough AI to trust it, so give them time. Three years of data say something different.
Three years at a glance
| Measure | 2024 | 2025 | 2026 | Trend |
|---|---|---|---|---|
| Very familiar with AI | 31.6% | 41.7% | 49.7% | +18.1 |
| Less likely to give after a charity announces AI | 31.4% | 32.2% | 32.3% | +0.9 |
| More likely to give after that announcement | 9.0% | 14.2% | 11.0% | +2.0 |
| Uncomfortable with AI-personalized appeals | 39.8% | 38.8% | 40.3% | +0.5 |
| Transparency about AI is “very important” | 69.4% | 69.2% | 77.0% | +7.6 |
| Sees potential for AI to improve charities | 49.4% | 51.5% | 54.2% | +4.8 |
| Does not see that potential | 32.2% | 33.0% | 36.2% | +4.0 |
Independent annual samples, not a panel: 2024 n=1,006 · 2025 n=1,031 · 2026 n=1,020. Trend labels describe direction across the reported years, not statistical significance.
Donors learned what AI is. They did not change their minds.
One objection deserves an answer up front. The 2024 sample was older, with 42.7% aged 60 or above. But the 2025 and 2026 samples are nearly identical in age (29.0% and 29.2%), and familiarity still rose eight points between them while giving intentions did not move at all. The pattern is not a sampling artifact.
A person can know exactly what a language model does and still not want one in her relationship with a cause she loves.
Among donors who call themselves very familiar with AI, 29.0% would still be less likely to give — and 85.0% of that same group call transparency very important. Sophistication raises the bar rather than lowering it.The line donors draw is not AI. It is concealed identity.
Most findings in this study divide donors. This one unites them. Asked which AI applications in charities concern them, 716 of 1,020 donors selected bots portrayed as humans — twenty points clear of anything else on the list.
Which AI applications concern donors
2026 respondents selected 2.1 applications on average. For 175 donors, 17.2% of the sample, a bot presented as a person was the only application they marked at all.
What sets this concern apart is not its size. It is its reach.
Cut the same question sixteen ways and a majority selects it in fifteen of them.
Selection rates from the 2026 dataset. Groups are defined by responses to other questions in the same survey. The chatbot itself draws modest concern. The chatbot wearing a name and a face draws the most concern in the study.
Donors want to know when they are dealing with a machine, and they want a person to remain accountable.
Across three years and three different question formats, human review of AI decisions and a clear public explanation have taken the top two places every time. Third-party audits trail in every year.Three donors sit on every list
Percentages describe a crowd. The individual responses let us see how attitudes cluster inside one person, and three groups emerge that a nonprofit will meet on any donor file. A single message about AI cannot serve all three.
The open
Would give the same or more after an AI announcement. Younger and more familiar with AI. Still, 55.6% of them select the bots-as-humans concern.
The persuadable middle
Say an AI announcement would not change their giving, or that it depends on how the AI is used. The largest group in every wave, and the one your decisions are actually addressed to.
The resistant core
Would give less and are uncomfortable with personalization. Seven in ten see no potential in AI for charities, yet 41% describe themselves as very familiar with it.
The group most leaders underestimate
When the 2026 wave asked donors to rank six things that might make them more comfortable with a charity’s use of AI, 21.0% put “nothing would make me comfortable” in first place — ahead of a public explanation, human review, evidence, opt-outs and audits. This is not a group that needs a better explanation. It is a group that has decided.
In 2025, when the same options were offered as a checklist, only 3% ticked that box. A ranking forces a first choice and a checklist does not, so the two figures are not comparable. One donor in five choosing “nothing” over every concrete safeguard is a finding on its own terms. The landscape is polarized in both directions: 42% of donors ranked “nothing” dead last, the most common last-place choice by a wide margin.
The worry moved from the database to the conversation
The 2024 and 2026 waves asked donors to select up to three concerns, which makes them directly comparable. Read on a like-for-like basis, two concerns moved.
| Concern about charities using AI | 2024 | 2026 | Change |
|---|---|---|---|
| Lack of human touch in interactions | 60.2% | 64.1% | +3.9 |
| Privacy and data security risks | 60.0% | 54.0% | −6.0 |
| Job displacement of charity workers | 43.3% | 44.4% | +1.1 |
| Potential for bias in AI decision-making | 39.1% | 37.3% | −1.8 |
| Over-reliance on technology | 38.3% | 33.2% | −5.1 |
Both years asked donors to select up to three. The 2025 wave used a select-all format and a sixth option, so it is excluded from this comparison. We read this as a change in what donors picture when they hear the word AI. Three years ago the picture was a database. Today it is a conversation.
The human-touch concern climbs with age, from 58% of donors under 45 to 72.5% of those aged 60 to 75. It is especially strong among the 36.4% of donors who say they have become more selective about which charities they support: 72.5% of that group select it. Selective donors are watching for exactly this.
A practical agenda for the next ninety days
Nothing here asks an organization to slow down. Done well, these steps are how a nonprofit earns the room to use AI more, not less. Donors who know what they are dealing with, and who can always reach a person, are the donors who let you keep building.
- Fix the identity first
Audit every donor-facing automation you run today, not only the ones you are planning. For each, ask whether it has a human name, a face, a signature, a voice, or an opening message that a reasonable donor would take as a person. If the answer is yes, the fix is not a line in your privacy policy. It is the identity itself.
- Disclose at the door, and again at the gift
A plain-language page on your website saying what AI is used for, what it is not used for, who reviews its output, and how to reach a person. Then repeat it in one or two sentences whenever an automated system asks for money or personal information. A donor who read the website in March should not have to remember it when a voice agent calls in December.
- Make access to a person a designed feature, not a fallback
Every automated interaction needs a dependable route to someone who can take responsibility, and the handoff has to work the first time. Measure it the way you would measure a donation page: time to reach a human, how often the handoff fails, and what donors say afterward.
- Build the first business case on a mission improvement, not a donor-facing one
Reports that arrive sooner with fewer errors. Event questions answered accurately the same day. Measure whether it happened before extending the tool toward donor contact. A useful internal application is a foundation, not a license.
- Stop writing one message about AI for “donors”
Write for the skeptic who will read your disclosure looking for a reason to leave, and for the enthusiast who will read it looking for evidence you know what you are doing. Each is now more than a third of your list, and they sit at the same tables. Keep a general, human, non-automated path alive for the donor who has already decided.
“We are transforming our work with AI” gives a supporter nothing to judge.
“Software helps organize the reports our grantees send us. Our program team checks the information and writes the conclusions you receive” gives them a division of labor and a person who is accountable. Use the second sentence only when it is true.Methodology
Each wave surveyed U.S. adults who reported giving to charity in the previous twelve months, recruited through a managed online panel with automated and manual quality checks.
- Samples: 1,006 respondents in August 2024, 1,031 in August 2025, and 1,020 between September 15 and 19, 2026.
- Screening: the 2026 file contains 4,649 records. 2,977 were screened out for not having given; remaining incompletes, duplicates and quality failures were excluded, leaving 1,020 completed U.S.-located responses.
- Design: the samples are independent. Each wave measures the donor climate in that year, not change within individuals, so broad patterns deserve more confidence than small year-to-year movements.
- Comparability: several questions changed format between waves. Where formats match we compare directly; where they do not we compare direction only. The 2024 sample was older (42.7% aged 60+) than 2025 and 2026 (29.0% and 29.2%), so the 2025-to-2026 comparison is the cleanest in the series.
- Who is in the sample: 78.5% of 2026 respondents gave $500 or less, and 76.6% report household incomes below $100,000. The findings speak to broad donor engagement and should not be treated as a complete account of major-donor expectations.
- What surveys measure: stated attitudes and intentions. A donor who says she would be less likely to give is not a forecast of a specific revenue loss, and an association between familiarity and comfort does not prove that education causes acceptance.
Read the full 2026 report
The complete study includes the full subgroup tables, the three-year reference page, the five questions to ask before any donor-facing automation goes live, and where we believe the next three years are heading.
