What AI still can't do (and probably won't)
What AI is genuinely changing about fundraising, and the five things it cannot reach. The conversation in the room. The voice that sounds like the organisation. The judgement that comes from years of doing the work. Donors are noticing. Recent research suggests they are marking down on perceived authenticity when they sense AI in the room.
Virginia Simpson, Signalwork
AI is reshaping the back end of fundraising in ways that are real and worth paying attention to. The drafting, the synthesis, the briefing layer.
It is not reshaping the front end. The conversation in the room. The unspoken pressure a funder is under. The trust that took eight years to build. The voice that sounds like the organisation and not like every other organisation.
Donors are noticing. Recent research suggests they are marking down on perceived authenticity when they sense AI in the room. The boundary is not theoretical. It is showing up in click-through rates.
Last month a senior fundraiser asked me what AI had changed about her work, and what it had not. She gave a confident answer. Walking the dogs that evening, I kept thinking about how she had only got half of it right.
The half she got right is the easy half. AI is genuinely changing the back end of fundraising. The pre-call brief that used to take an hour now takes ten minutes. The funder report that used to need a week now needs an afternoon. The pattern across thirty stakeholder interviews that used to require a small team can now be surfaced by one person paying close attention to a workflow. None of this is hype. It is happening, in fundraising teams across the sector, and it is making people more productive at the parts of the job that benefit from being faster.
The half worth thinking harder about is the front end. The conversation, the relationship, the trust. The bit that determines whether the money actually moves. There are five places where I keep watching senior fundraisers expect AI to help them, and it does not, and I am increasingly sure it will not.
1. AI cannot read a room
A funder pauses before answering a strategy question. The pause is half a second longer than the previous pause. There is a slight tightening of the jaw, a glance at the wall behind the camera. None of this reaches the transcript. None of this reaches a meeting summary. The conversation continues. The pause is not technically anything. It is just air.
Anyone who has spent time in fundraising rooms knows that the pause was the most important thing that happened in the meeting. It signalled an unspoken pressure the funder is under, a constraint they have not articulated, sometimes a doubt about the relationship that they would never raise on the record. The fundraiser who notices the pause and follows it up gently does the work of the meeting. The one who does not, sends a polite follow-up email and wonders six months later why the renewal did not come through.
AI tools are getting good at transcripts and summaries. They are getting good at extracting themes. They are not getting good at hesitation, because hesitation is not in the recording. It is in the gap between the recording and what the person actually meant. That gap is where most of the relational signal in fundraising lives.
2. AI cannot keep your organisation sounding like itself
There is a thing happening to fundraising prose at the moment that I find quietly alarming. Appeals are getting more polished. Case statements are getting cleaner. Cover letters are getting more confident. And they are all starting to sound like each other.
The 2025 Donor Perceptions of AI study, which surveyed just over a thousand US donors, found that around a third said the use of AI in a charity's communications would make them less likely to give. The applications they were most worried about were not predictive analytics or chatbots. They were AI-generated appeals and AI bots posing as humans. A January 2026 paper in the Journal of Business Ethics ran a Facebook field experiment with 118,000 impressions and found significantly lower click-through rates on AI-labelled charitable advertising. The mechanism, the researchers argued, was that donors infer extrinsic motives, efficiency, cost-cutting, when they sense AI in charitable communication, which reduces perceived authenticity, which reduces giving.
The behavioural reading of this is interesting. Donors are not necessarily anti-AI. They are pattern-matching on authenticity, and AI prose is failing the test. Not because it is incorrect, but because it is too smooth. The organisation that sounds like every other organisation does not sound like itself, and donors notice.
I do not think this is solved with disclosure. I think it is solved at the craft level, by a person who is willing to make the prose worse on purpose, in the specific direction that makes it sound like the people who actually do the work.
3. AI cannot fix a misunderstood problem
Most of the AI projects in fundraising right now are bolting tools on top of broken foundations. The CRM is half-empty. The institutional knowledge lives in three people's heads. Funder context is fragmented across email, Slack, and somebody's notebook. The instinct, when AI arrives, is to add it to the stack. The result is that the AI gets faster at producing slightly more confident versions of the wrong answer.
Bridgespan and the Center for Effective Philanthropy published research last year that landed on something similar. Three quarters of nonprofits believe their funders have little to no understanding of their AI-related needs. Fewer than 20 percent have ever discussed AI with a funder. The conversation that should be happening is not which tool to buy. It is whether the underlying knowledge work is being done at all. AI multiplies what is already there. Where what is there is thin, AI gives you thin output, faster.
This is the bit fundraising leaders find hardest to hear, because the appeal of AI is precisely that it appears to fix the underlying problem without fixing the underlying problem. It does not. There is no shortcut.
4. AI cannot be trusted blind
AI hallucinates. Everyone working with these tools knows this in the abstract. The risk is not that it sometimes invents things. The risk is that it invents things confidently, in the same fluent register as everything else, and the human reading the output is not paying close enough attention.
A 2024 Stanford study tested legal AI tools, including ones marketed as hallucination-free, and found they invented citations on roughly one in six queries. Legal AI sits in a domain with rigorous training data and clear ground truth. Fundraising sits in a domain with neither. The seams are wider. The errors are quieter. The likely scenario is not that AI fabricates a funder. It is that AI subtly mis-remembers what a funder said in a previous conversation, and the fundraiser walks into the next meeting with a slightly wrong picture of the relationship.
The behavioural risk here is interesting. Fluent AI output triggers the same trust responses as fluent human output. The reader's brain does not flag the source. Which means the discipline of checking has to be imposed at the workflow level, by a person, every time. The technology will not impose it on itself. It will get better. It will not get reliable in the way humans need it to be reliable.
5. AI cannot replace the judgement that comes from doing the work
This is the one I am least sure about, and the one I think matters most.
There is a class of fundraising knowledge that lives in nobody's notes, because the people who hold it cannot describe how they hold it. The instinct that this funder will say yes if asked at the right moment in the right tone, and will say a permanent no if asked five minutes earlier. The pattern recognition across years of conversations that lets a senior fundraiser hear a phrase and know what is coming next. The judgement about which board member to put in the room.
The economist David Autor wrote a paper a decade ago about why automation has not eaten knowledge work the way mid-century economists predicted. The argument, in plain English, is that there is a category of human work where the practitioners cannot describe how they do it. Not because they are hiding the secret. Because the knowledge does not exist in language. It lives in the doing. Polanyi called this tacit knowledge in 1966. The phenomenon has not changed since. The tools have got cleverer at the parts of work that can be written down. The parts that cannot be written down have stayed where they were.
Fundraising is dense with this kind of knowledge. The brief is in the brief, but the read is in the room. The pattern is in the pattern, but the meaning is in the years.
What I am left with
Two things, mostly.
The first is that I think the sector is about to spend a few years adopting AI for the wrong reasons. Because the funder asked. Because the board read an article. Because the consultant said. The output will be faster, more confident, and slightly worse. Whether anyone notices will depend on how closely the people in the room are paying attention.
The second is that the parts of fundraising work AI has touched least are the parts that determine whether the money actually moves. The conversation that goes somewhere unexpected. The phrase a funder uses that the fundraiser will not understand for six months. The judgement call that cannot be shown its workings. I do not know if these are durable advantages or just things experienced fundraisers happen to do. I am increasingly sure they are durable.
The question I think senior fundraisers should be asking, and that I would be interested to hear behavioural psychologists weigh in on, is not what AI can do for fundraising. It is whether the people doing the work still know what they are doing well enough to notice when AI gets it wrong.
SignalWork is a strategic advisory that surfaces what funders, stakeholders, and programme partners actually think but will not say publicly. We have conducted over 320 confidential senior interviews across the UK, Europe, Southern Africa, and the US. Clients include Centre for Public Impact, Carbon Trust, Tony Blair Institute, and Co-Impact.
If this is sitting with you, get in touch.