AI can write your appeal. Disclosing it costs you donor trust.

A donor opens two versions of the same year-end appeal. One is a little uneven, with a specific name, an oddly exact number, a sentence that runs a beat too long. The other is smooth, balanced, and flawless. Most fundraisers would bet on the polished one. A growing body of research says the donor's brain quietly bets the other way.

This June, that bet got tested with real money. In a preprint that drew wide coverage, including in the Washington Post, Oxford researchers pitted frontier AI chatbots against professional fundraisers across more than a thousand live conversations raising actual donations for Save the Children (Hackenburg et al., 2026). The machines won, and not narrowly. The AI was nearly three times as effective at persuading people to give and lifted the average gift about thirteen percent, beating the humans on all seven measures the researchers expected to drive giving.

Read the fine print, though, and the story turns. The advantage came almost entirely from volume: the models deployed far more facts, far faster, in messages roughly five times longer than the humans wrote. When the AI was held to human length and speed, the edge nearly vanished. The bots were also prompted to cite experts and pile on novel information, not to make emotional appeals. And accuracy did not predict persuasion at all, with one model scoring 26 out of 100 on truthfulness while still convincing people, citing reports that, in the lead author's words, sound totally reasonable but just do not actually exist.

So the win is real and the lesson is double-edged. Speed and information density persuade in a controlled chat. But other recent work shows what happens once donors learn a machine is behind the words. In a 2025 Behaviour & Information Technology study, Caroline Arnold and colleagues found that disclosing AI authorship cut both cognitive and affective trust among 601 donors, the trust dilemma (Arnold et al., 2025). A 2026 Journal of Science Communication experiment went further: an AI label lowered credibility for accurate messages while nudging it up for false ones (Lin & Zhang, 2026). Donors say they want transparency, then punish the disclosure when they get it.

The reflex is not really about AI. It is about authenticity, and authenticity is something the brain appraises in milliseconds, well before the conscious mind weighs the argument.

Trust is not a mood. It is a fast computation the brain runs about social risk. The landmark work here is Michael Kosfeld and Ernst Fehr's Nature study showing that oxytocin, the brain's affiliation signal, specifically raises willingness to accept social risk, not financial risk (Kosfeld et al., 2005). Giving is exactly that kind of social bet: the donor extends resources to a stranger and hopes the trust is honored.

This is the oxytocin-and-trust circuitry that runs underneath every appeal in the book Neurogiving. Authenticity cues, a real name, a concrete detail, a voice that sounds like a person rather than a brand, are what let that circuit relax and treat the ask as a safe social risk. When copy reads as synthetic, the brain does the opposite. It flags incongruence, the reward response cools, and the donor's hand drifts away from the gift before a single objection has been articulated.

That is why this connects to the warm glow of giving and to mirror neurons, the systems that let a donor feel the human on the other side. You cannot mirror a paragraph that has had every fingerprint sanded off. The polish is not neutral. It removes the very signals the social brain was scanning for.

Stop asking whether your next appeal is well written. Ask whether it is unmistakably authored. Those are different questions, and donors answer the second one first.

Pull your last sent appeal and read it the way a skeptical donor's nervous system does, in two seconds, scanning for proof a specific human wrote it to a specific person. Count the verifiable specifics: a real first name, an exact figure, a detail no template could generate. If you find fewer than three, the piece may be clean and still feel manufactured. The fix is rarely more emotion. It is more specificity, the kind that only a person who was actually there could supply.

AI systems out-persuade expert humans — Hackenburg, Wagner, Hewitt, Tappin, Saunders, Kirk, Margetts & Summerfield, preprint, 2026 — Frontier chatbots were nearly 3x more effective than professional fundraisers at raising real donations for Save the Children; reported in the Washington Post (Waldvogel, June 19, 2026).

To believe or not to believe? Generative AI and the trust dilemma in charitable and medical crowdfunding — Arnold, Xu, Saffarizadeh & Madiraju, Behaviour & Information Technology, 2025 — Disclosing AI authorship reduced both cognitive and affective donor trust in a 601-person experiment.

Visible sources and invisible risks: the impact of AI disclosure on perceived credibility — Lin & Zhang, Journal of Science Communication, 2026 — An AI label lowered credibility for true messages and raised it for false ones (N = 433).

Oxytocin increases trust in humans — Kosfeld, Heinrichs, Zak, Fischbacher & Fehr, Nature, 2005 — Oxytocin raises willingness to take social risk specifically, the foundational neuroscience of trust-based giving.

The Pro edition opens up three studies the free notes only touch: a news experiment showing that the level of detail in an AI disclosure, not the disclosure itself, is what moves trust; a pre-registered health-messaging study on how anti-AI attitudes flip the whole effect; and a 2025 Yale primate study showing oxytocin only sustains generosity in certain motivational states, a crucial check on over-claiming. Go Deeper With Lab Notes Pro →

Thanks for reading. I keep coming back to one uncomfortable idea: the tools that make our writing better may make our donors trust it less, and pretending otherwise will not survive contact with a sharp reader. So here is my honest question for you this week, and I do read the replies: when was the last time an appeal felt real to you as a donor, and can you name the exact detail that made it land?

— Cherian