Readers think they can spot real writing, trusting their ear for voice, rhythm, style. They assume they know what human feels like on the page. So when a group of researchers pitted expertly trained fine arts writers against AI, the expectation was obvious, humans would win easily.  And they did…in the first phase.

When the models were only prompted in the standard way, expert writers and everyday readers both preferred the expertly written, human product and not by a little, 6 to 8 times more often. The AI had all the usual tells with competent sentences but a too frequent rhymying style, a one clichy too many vibe and predictable staccato. In other words, a machine noticeably imitating humanity with it’s own machine style.

Then the second finding arrived and overturned the entire premise. The researchers fine-tuned the LLM on a selection of top selling author’s, to teach it how to write really well and overwrite it’s machine like, out of the box tendencies.  The AI stopped sounding like an imitator and began sounding like the author. Expert readers and the everyday readers preferred the fine-tuned excerpts eight to one and the AI detection tools that easily caught the first round of writing could not tell the second round was AI.

The model trained on the deeper signal was no longer competing with expertly trained writers, it was beating them.

What’s the fundraising parallel?

The literary training worked because the authors had a stable stylistic identity and very high quality output.

That is not how fundraising copy works. Most nonprofit writing is assembled, edited and compromised into something that isn’t a voice at all but a template.  And our extensive copy writing evaluation tool finds most copy scoring poorly on readability and narrative features.

Do you really want to fine tune a model on that?

A novelist’s voice is not just words. It is the psychological signature behind the words. Decisions about agency, control, intimacy, distance, timing, vulnerability, what is said and what is withheld. That is why the fine-tuned models worked. They learned not just surface moves but the underlying pattern.

Good fundraising copy has its own underlying pattern and it has nothing to do with brand tone.

Giving is autobiographical, every donor brings their own story to the decision. Their own map of what matters, what feels just, what feels possible and where they see themselves in the world. They give to reinforce identity, to express agency, to restore coherence, to meet an emotional or moral need, to feel connected to a cause that reflects something about how they want to live.

Three psychological drivers sit underneath that process.

Autonomy.
People give because they want to feel like agents in their own story rather than spectators or pawns in someone else’s crisis reel. The language that works affirms choice, control and personal authorship.

Competence.
Donors act when the message lets them feel capable and effective. They need to believe the action does something real. They need clarity, not theatrics. They need a story where their intervention works.

Relatedness.
Connection drives giving. Not manufactured intimacy or sentimental manipulation. Actual resonance. A sense that the story aligns with the way they understand people, the world and their place in it.

These are not abstract concepts, they are the architecture of why anyone gives at all.

Layer on identity, values and traits, and you have the real corpus worth training on. Not the organization’s archive of appeals but the donor’s

• internal logic
• their moral vocabulary
• their way of describing agency, fairness, responsibility, impact.
• the frames they gravitate toward
• the emotional temperature they trust.

This is the fundraising version of the author’s oeuvre, its signal deep enough to support fine-tuning.  This is the material with the same coherence the literary study exploited.

If you want an AI model that outperforms humans, give it the right raw material. Feeding it thirty years of direct response boilerplate will only produce a faster version of what you already have.

Feeding it donor psychology gives you a system able to construct stories that function as mirrors, where the donor recognizes themselves in the narrative.  A system that can vary message, tone, structure and emotional load by trait, by identity and by motivational pattern.

The researchers proved the model wins once the model sees the true pattern, fundraisers can emulate if they feed the machine the right one.

Kevin